spb/wp35_uqo

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

modified README.md +10 −10

		@@ -3,7 +3,7 @@
3	3
4	4	**UQO Working Paper No. 35** — *A pre-registered test of coordination demand, institutional capacity, and instability in historical and modern societies, with a formal framework and agent-based experiments*
5	5
6		−[-b31b1b?logo=latex&logoColor=white)](paper/main.pdf)
	6	+[-b31b1b?logo=latex&logoColor=white)](paper/main.pdf)
7	7	[](pyproject.toml)
8	8	[](data/raw/README.md)
9	9	[](Makefile)
		@@ -11,7 +11,7 @@
11	11	[](https://uqo.ca)
12	12	[](mailto:simon-pierre.boucher@uqo.ca)
13	13
14		−> **TL;DR** — Can the mismatch between a society's coordination *demand* (scale) and its coordination *capacity* (administration, standardized rules and money, communication infrastructure, record-keeping) predict instability? Under a pre-registration frozen before any outcome was examined, we built the indices from Seshat Polaris-2026 (626 polities, 127 CrisisDB crises with pre-crisis predictors), the Ethnographic Atlas (1,291 societies) and a V-Dem/Maddison panel (14,637 country-years, 2,692 events). Capacity and scale are positively associated everywhere. The mismatch index is positively associated with crisis severity (0.20 log-odds/SD) and modern instability onsets (0.40/SD), stable across regions and eras — **but fails the pre-registered test**: it does not beat the scale-only baseline (Seshat) or the capacity-only baseline (modern) out of sample, a placebo does as well, past events load on it as strongly as future ones, and an Oster adjustment flips its sign. Five theorems make "institutions as codes" precise; 6,632 agent-based runs confirm them in silico and locate where institutions hurt (unpaid set-up, saturated hierarchies, unstructured task spaces). The universal theory is not supported; the narrower paper is what we wrote.
	14	+> **TL;DR** — Can the mismatch between a society's coordination *demand* (scale) and its coordination *capacity* (administration, standardized rules and money, communication infrastructure, record-keeping) predict instability? Under a pre-registration frozen before any outcome was examined, we built the indices from Seshat Polaris-2026 (626 polities, 127 CrisisDB crises with pre-crisis predictors), the Ethnographic Atlas (1,291 societies) and a V-Dem/Maddison panel (14,637 country-years, 2,692 events). Capacity and scale are positively associated everywhere. The mismatch index is positively associated with crisis severity (0.20 log-odds/SD) and modern instability onsets (0.39/SD), stable across regions, eras and cultural spheres — **but fails the pre-registered test**: it does not beat the scale-only baseline (Seshat) or the capacity-only baseline (modern) out of sample, the equal-and-opposite restriction it imposes is rejected in the modern panel, a placebo is not distinguishable from it in Seshat, past events load on it nearly as strongly as future ones, and it is not robust to an Oster-type adjustment; higher capacity even predicts shorter polity durations in Seshat. Five theorems make "institutions as codes" precise; 6,632 agent-based runs confirm them in silico and locate where institutions hurt (unpaid set-up, saturated hierarchies, unstructured task spaces). The universal theory is not supported; the narrower paper is what we wrote.
15	15
16	16	---
17	17
		@@ -41,14 +41,14 @@ Pr(Y456 in (t,t+5]) = Λ(α + β P_it + γ'X_it + μ_r + τ_t) # modern onsets
41	41	|---|---|
42	42	| corr(z(D), z(K)) across Seshat polities | 0.84 — P retains 17% of the joint variance |
43	43	| H1 Seshat: D on K (region + era FE) | 0.75 SD per SD of K (SE-based 95% CI excludes 0); EA: 0.70 log community size per jurisdictional level; modern within-country 0.15 |
44		−| H2-A Seshat crisis severity, β_P | 0.196 (SE 0.056, n = 127); with covariates 0.204 (0.085, n = 87); scale-only 0.257 (0.138); placebo 0.257 (0.138) |
45		−| H2-A out of sample | beats FE-only / K-only / complexity in ≥ 80% of folds; **not** scale-only (75%), persistence (69%), D+K (75%); loses to scale alone in 7/10 leave-region-out folds |
46		−| H2-C modern Y456, β_P | 0.398 (0.089, n = 14,637); capacity-only −0.632 (0.126); negative control (past events) 0.436 vs forward 0.475; Oster-adjusted −0.024; RV = 0.07 |
47		−| H2-C out of sample | never beats capacity-only (forward chaining 0/2; LORO 1/6 log-lik) |
48		−| H4 threshold | Seshat: hinge/spline beat linear in 10%/8% of folds; modern: 80%/87% but +0.001 nats (negligible) |
	44	+| H2-A Seshat crisis severity, β_P | 0.196 (SE 0.056, n = 127); with covariates 0.243 (0.086, common sample n = 93); scale-only 0.416 (0.173); D and K separately 0.44 / −0.13 (Wald β_D+β_K=0: p = 0.22); placebo 0.192 (0.166) vs P 0.290 (0.079) with the same controls |
	45	+| H2-A out of sample (50 polity-blocked folds) | beats FE-only / K-only / complexity in 84–86% of folds; **not** scale-only (42%), persistence (78%), D+K (44%); loses to scale alone in 8/10 leave-region-out folds |
	46	+| H2-C modern Y456, β_P | 0.392 (0.089, n = 14,637; 901 distinct event-years); capacity-only −0.631 (0.131); D and K separately 0.17 / −0.70 (Wald β_D+β_K=0: p < 0.001); negative control (past events) 0.436 vs forward 0.475 (passes the literal rule by < 1 SE); Oster-adjusted −0.030; RV = 0.07 |
	47	+| H2-C out of sample | beats capacity-only in 1/3 forward-chaining splits and 1/6 leave-region-out folds (log-lik) |
	48	+| H4 threshold | Seshat: hinge/spline beat linear in 10%/8% of folds (rejected); modern: 73%/80% — spline meets the rule exactly, +0.003 nats (not rejected, small) |
49	49	| H5 | β_P positive in every held-out macro-region, era, region and period |
50		−| Y2 polity hazard / Y3 succession | no association with P (HR 1.00–1.03) |
51		−| H3 (SIMULATED) | norms/records lower operating bits per success in 99.7% / 71% of paired runs and raise success in 99.3% / 100%; central hierarchy at half capacity: success −0.44; a 200-bit code is repaid only with s = 256, Zipf 1.5, T = 160; stale-code penalty 1.90 bits vs KL 1.98 |
	50	+| Y2 polity hazard / Y3 succession | no association with P (HR 1.00–1.03); **K associated with higher hazard** (HR 1.75; 1.71 with a periodization control) |
	51	+| H3 (SIMULATED) | norms/records lower operating bits per success in 99.7% / 71% of paired runs (success effects are parameters of the model); at equal budgets, central hierarchy at half the load: success −0.49, federated −0.29; a 200-bit code is repaid only with s = 256, Zipf 1.5, T = 160; stale-code penalty 1.90 bits/recipient vs KL 1.98 |
52	52
53	53	## 🗂 Repository layout
54	54
		@@ -139,4 +139,4 @@ Simon-Pierre Boucher — simon-pierre.boucher@uqo.ca — Département des scienc
139	139
140	140	## 📝 Notes
141	141
142		−Numbering: WP 35 (next free number after WP 34). Category folder `economie/` created for this paper. Adversarial reviews and responses: `docs/reviews/`. Final audit: `results/logs/final_research_audit.md`.
	142	+Numbering: WP 35 (next free number after WP 34). Category folder `economie/` created for this paper. Adversarial reviews (institutional economist, historical anthropologist, computational social scientist) and the response: `docs/reviews/` — they found real defects (a tautological placebo, silently failed CV folds, mislabelled rows, un-run robustness variants, simulation accounting) that were corrected (amendments A-002 to A-007 in `research/decisions.md`). Final audit: `results/logs/final_research_audit.md`.

modified configs/regions.yaml +13 −0

		@@ -18,3 +18,16 @@ era_blocks: # CE years, applied to the polity midpoint (start_year+end_year)/2
18	18	- {name: "1001-1500", min: 1001, max: 1500}
19	19	- {name: "1501-1800", min: 1501, max: 1800}
20	20	- {name: "post-1800", min: 1801}
	21	+# Alternative hold-out scheme added after the historical-anthropologist review (A-002): cultural spheres rather than
	22	+# geography (Sahel with sub-Saharan Africa; Southeast Asia separated from Oceania; steppe separated from Southwest Asia).
	23	+cultural_spheres:
	24	+ Europe: [Western Europe, Southern Europe, Central Europe, Northern Europe, Eastern Europe, Southeastern Europe]
	25	+ West_Asia_North_Africa: [Anatolia-Caucasus, Mesopotamia, Levant, Iran, Arabia, Afghanistan, Northeast Africa, Maghreb]
	26	+ Steppe_Central_Asia: [Turkestan, Mongolia, Siberia, Pontic-Caspian]
	27	+ South_Asia: [North India, Southern India, Eastern South Asia, Pakistan, Sri Lanka]
	28	+ East_Asia: [North China, South China, Northeast Asia]
	29	+ Southeast_Asia: [Mainland Southeast Asia, Maritime Southeast Asia]
	30	+ Oceania: [Polynesia, New Guinea]
	31	+ Sub_Saharan_Africa: [West Africa, East Africa, Central Africa, Southern Africa, Sahel]
	32	+ Mesoamerica_North_America: [Mexico, Mississippi Basin, Western North America, East Coast, Caribbean]
	33	+ Andes_Amazonia: [Andes, Amazonia]

modified data/processed/modern_h2_sample.parquet +0 −0

Binary file not shown.

modified data/processed/seshat_crisis_indices.parquet +0 −0

Binary file not shown.

modified data/processed/seshat_h2_sample.parquet +0 −0

Binary file not shown.

modified data/processed/seshat_indices.parquet +0 −0

Binary file not shown.

modified data/processed/seshat_polity_meta.parquet +0 −0

Binary file not shown.

added docs/reviews/response_to_reviewers.md +71 −0

		@@ -0,0 +1,71 @@
	1	+# Response to the three adversarial reviews — WP35 (2026-11-01)
	2	+
	3	+The three reviews (`review_institutional_economist.md`, `review_historical_anthropologist.md`,
	4	+`review_computational_social_scientist.md`) were produced by independent reviewer personas with access to the full
	5	+repository. All three recommended **major revision**. None endorsed the paper, and none was asked to. This document lists
	6	+each substantive point, what was done, and what was not done and why. Every change is logged as an amendment (A-002 to
	7	+A-007) in `research/decisions.md`; analyses were rerun; the manuscript was recompiled (80 pages).
	8	+
	9	+## A. Defects acknowledged and fixed (reruns performed)
	10	+
	11	+| # | Reviewer | Point | Action |
	12	+|---|---|---|---|
	13	+| 1 | Anthropologist, Economist | Five robustness variants announced in the text (range bounds, inferred codes dropped, within-era/region standardization, ranks) had **never been computed** at the crisis level: the code skipped missing columns silently; the text cited a table that did not contain them. | Variants built at the crisis level (`build_indices.py`), estimated, tabulated (Table 10): all positive, 0.13–0.27. Rule-1 violation logged (A-002). |
	14	+| 2 | Economist | The Seshat **placebo was a tautology**: `P_placebo = z(D) − z(X_prod_infra)` fitted with `X_prod_infra` as a control is an exact reparametrization of the scale-only baseline (identical coefficient 0.2572…, SE, log-likelihood). | Placebo now fitted without the production-infrastructure control, with `P` fitted on identical controls for comparison: placebo 0.19 (SE 0.17) vs P 0.29 (SE 0.08). Abstract and text changed from "performs comparably" to "not distinguishable from P" (A-006). |
	15	+| 3 | Economist, CSS | **Silently failed CV folds**: the first forward-chaining split (train ≤ 1945) failed for every model (a region absent in training); 14 Seshat model-folds failed similarly; the text reported "three splits" and "fifty folds". | Test rows with fixed-effect levels unseen in training are now mapped to the training mode (`common.map_unseen_levels`); all 3 forward splits and all 50 Seshat folds run; counts corrected (A-004). |
	16	+| 4 | Economist | Seshat k-fold not blocked by polity (16 polities with several crisis records share identical predictors). | Folds blocked by polity in all Seshat cross-validation (A-004). |
	17	+| 5 | Economist | `epidemic_v` used as a covariate is itself a CrisisDB consequence item. | Dropped from the control set; pre-registration deviation logged (A-006). |
	18	+| 6 | Economist | Baselines fitted on different samples (n = 87–127) while the text said "same observations". | All covariate specifications and baselines now fitted on a common sample (n = 93); A1 (n = 127) kept and labelled. |
	19	+| 7 | Economist | Row "A3: Y1 ~ P + D + K" was mislabelled (it conditions on the complexity component); modern "C3: P given K" is algebraically B5. | Relabelled; C3 described as the reparametrization of B5 that it is. |
	20	+| 8 | Economist | P imposes β_D = −β_K; the unrestricted model is the right object. | Wald test of β_D + β_K = 0 added in both panels: not rejected in Seshat (p = 0.22, because β_K is imprecise), rejected in the modern panel (p < 0.001). Reported in results and abstract. |
	21	+| 9 | Economist, CSS | **Negative control and H4 misread** against the written rules: the negative control passes by the letter (0.44 < 0.48); the modern H4 rule was met. | Text now states the literal verdicts (negative control passes by < 1 SE; H4 "not rejected" in the modern panel, spline 12/15, +0.003 nats) with the interpretive caveats (A-007). |
	22	+| 10 | Economist | Holm correction asserted, never computed. | Computed (`holm_primary_tests.csv`): all three adjusted p < 0.001. Reviewer's point that H1-Seshat is an "unfailable" primary is accepted and noted in the limitations. |
	23	+| 11 | Economist | Measurement overlap: settlement hierarchy (D) and administrative levels (K) both measure hierarchy depth. | Overlap checks added (D without settlement hierarchy, K without administrative levels, both): β_P = 0.25 in all three. Presence-anywhere size bias of K acknowledged in limitations; not fixable with Polaris. |
	24	+| 12 | CSS | `t_since` imputed with the panel maximum for countries with no observed prior event. | Replaced by years since panel entry plus a left-censoring indicator in every modern specification (A-004). |
	25	+| 13 | CSS | Event count "2,692 instability events" are overlapping windows. | Abstract and text now give 2,692 event windows, 901 distinct event-years in the estimation sample, 1,676 in the panel. |
	26	+| 14 | CSS | Simulation accounting: bits ignored relay hops; only the LCA administrator was charged; the capacity ratio was defined against a load that ignored withdrawn producers; federated module administrators each received the system-wide bandwidth; distribution shocks were a no-op in uniform cells; network-shock robustness of flat regimes was a floor effect. | Codeword bits charged on every hop; every administrator on the relay path charged; **equal total budget** for central and federated designs (ratio × expected relayed load of the central design, allocated to administrators in proportion to a warm-up measurement of load); shock statistics restricted to non-uniform cells; shock effects also reported by bandwidth. All 6,632 runs rerun; simulation section rewritten (A-005). Qualitative conclusions changed in one respect: central hierarchies are now expensive in bits (83 per success vs 25 for records) and federation is the only hierarchical design that survives a tight budget. |
	27	+| 15 | CSS | Theorem 1 check wrongly described: an optimal prefix code is never longer than the fixed-length code; only the (suboptimal) Shannon code used in the simulation can be. | Remark added after Theorem 1 in the paper; simulation text corrected. |
	28	+| 16 | CSS | Simulation "confirms" success effects that are parameters (eps = 0.25 vs 0.02 vs 0). | Section 8 rewritten: success rankings are explicitly described as parameter-driven (with the eps ablation showing the reversal at eps = 0.05); the measured content is the cost side. Abstract changed from "confirm them in silico" to "instantiate them in silico on the cost side". |
	29	+| 17 | Anthropologist | Severity share is not a scale (14 weakly correlated items); Hoyer et al. 2024 cited but not applied. | Item-level analysis added (Table 12): mean inter-item correlation 0.08, first eigenvalue 19%; P and D load on elite-political items (decline, assassination, deposition, successful revolution, civil war) and negatively on century-long conflict; collapse is present in 2 of 126 records. Text added. |
	30	+| 18 | Anthropologist | K saturates in the crisis sample, so P ≈ rescaled D. | Reported: crisis-sample K mean 0.71 (0.58 overall), 40% of records with K > 0.8, corr(P, D) = 0.63 and corr(P, K) = 0.02 within the sample (Table 5, last row). |
	31	+| 19 | Anthropologist | Cox result (higher K → shorter polities) dismissed in two sentences. | Periodization control (polities per NGA) added: it raises the hazard (HR 1.37) but K stays at HR 1.71 (p = 0.009). Text rewritten to report the result as running against the hypothesis, with candidate explanations left open; added to the abstract. |
	32	+| 20 | Anthropologist | Macro-region scheme pools culturally distinct areas (Sahel with North Africa; Angkor with Polynesia). | Cultural-sphere hold-out scheme added (`configs/regions.yaml`); H5 rerun: pooled coefficients 0.24–0.37 in every held-out sphere. The original geographic scheme is kept as pre-registered. |
	33	+| 21 | Anthropologist | "Pre-crisis predictors" overstated (3.6% of facts dated). | Abstract and data section reworded; dated share reported; label kept only for consistency with the pre-registration. |
	34	+| 22 | Anthropologist | Pre-registration says "13 items" but lists 14. | Typo acknowledged (A-003); the list governs. |
	35	+| 23 | Anthropologist, CSS | Equinox-2020 replication called "blocked" in the paper vs "not executed" in the audit; Appendix A claimed "no post-hoc amendments". | Both corrected: "not executed"; Appendix A now references amendments A-001 to A-007. |
	36	+| 24 | CSS | Diagnostic file `seshat_unparsed_values.csv` was empty. | Cause: a unit test called the parser on a one-row frame and overwrote the diagnostic; parser now takes `write_diag=False` in tests; file regenerated (132 rows). |
	37	+| 25 | CSS | pymc/arviz pinned but unused; `replication/` empty. | Removed from `pyproject.toml`; `replication/README.md` written. |
	38	+
	39	+## B. Points accepted but not fixable with the available data (stated in the paper)
	40	+- Historical comparability of presence codes across 13,000 years; "presence anywhere" codes mechanically increasing in
	41	+ territory; coder circularity between Polaris institutional codes and CrisisDB codes (Anthropologist, Economist).
	42	+- EA033 is itself a scalar-stress response, so H1-E replicates Carneiro (1967) rather than testing compression; EA031's
	43	+ open top category (Anthropologist). Stated in limitations.
	44	+- The modern K items correlate 0.75 with electoral democracy and are coded ex post (Economist). Stated; the democracy
	45	+ control and placebo remain the only available mitigations.
	46	+- Oster's δ = 1 is implausible when a control is a near-component of the regressor; RV lacks a benchmark (Economist).
	47	+ The sensitivity paragraph now says so and no longer uses the word "reverses" in the abstract.
	48	+- The D+K baseline nests the P model, so P cannot beat it in expectation (Economist). Stated in limitations; the Wald test
	49	+ of the restriction is now the primary object of that comparison.
	50	+- The hierarchy saturation is confirmed, not discovered (CSS). Stated.
	51	+
	52	+## C. Points not accepted, with reasons
	53	+- "The 80% rule could be gamed or is too harsh" (Economist): the rule was fixed before any result; we report fold-level
	54	+ results so that any other rule can be applied, and we note that the verdict is the same under a 50% rule against the
	55	+ scale-only baseline in Seshat (42%).
	56	+- "Transductive PCA/standardization before CV is leakage" (CSS): it is outcome-blind and absorbed by the GLM slope and
	57	+ intercept; the fold-internal check (`h2_seshat_cv_noleakage.csv`, A-001) shows the verdict unchanged. Both reviewers
	58	+ who raised it agreed it is immaterial for the logit; it would matter for variants that re-fit PC1 on the test fold,
	59	+ which none do.
	60	+- "Replace the geographic macro-regions by cultural spheres" (Anthropologist): added as an additional scheme, not as a
	61	+ replacement, because the geographic scheme was pre-registered.
	62	+
	63	+## D. What changed in the verdict
	64	+Nothing in the direction of the main conclusion: H1 holds as association; H2 fails the pre-registered out-of-sample rule
	65	+in both panels; H4 is rejected in Seshat and now reported as "not rejected, small" in the modern panel; H5 holds. Two
	66	+things became stronger against the hypothesis after the corrections: the restriction β_D + β_K = 0 is rejected in the
	67	+modern panel, and the Cox result (higher capacity, shorter polities) survives a periodization control. One thing became
	68	+weaker against it: the Seshat placebo, correctly specified, is weaker than P (0.19 vs 0.29) though not distinguishable
	69	+from it. The simulation conclusions changed in substance for the hierarchies (central designs are expensive and fragile at
	70	+equal budgets; federation is the surviving hierarchical design) and in framing for the success effects (parameters, not
	71	+findings).

added docs/reviews/review_computational_social_scientist.md +63 −0

		@@ -0,0 +1,63 @@
	1	+# Review of UQO Working Paper 35, "The Social Compression Hypothesis" (v1.0, November 2026)
	2	+
	3	+**Reviewer persona:** computational social scientist (agent-based modelling, information theory, ML validation, reproducibility).
	4	+**Examined:** `paper/main.pdf` (pp. 1–55, 66–73), `research/framework.md`, the simulation, econometrics and IRT sources, `tests/`, `Makefile`, `results/`. `.venv/bin/pytest -q` → **11 passed, 3 FutureWarnings, 1.0 s**.
	5	+
	6	+## 1. Summary
	7	+
	8	+The paper is unusually candid: it pre-registers a mismatch hypothesis, finds that the index adds nothing beyond its components, and says so. The pipeline is traceable (every table number I checked matches a file under `results/`), the theorems are correctly stated, and the EMPIRICAL/SIMULATED/CONJECTURE labelling is disciplined. My objections concern what the agent-based experiments are said to show. The simulation does not *test* Theorems 1–5; it hard-codes their premises (zero decoding error for records, a capacity parameter defined as a multiple of demand, bits that ignore relay hops, an acknowledgement round paid only by the flat regimes) and reports the consequences as findings. Several headline results are accounting artefacts or floor effects, and one theorem check in the text is wrong. On the econometric side there is a silent fold failure that contradicts the text, a one-third imputed regressor, and an "80% of folds" rule applied to two folds. None of this changes the negative verdict on H2; it changes what §8 and the abstract may claim.
	9	+
	10	+## 2. Major issues
	11	+
	12	+**M1. "Institutions raise success" is baked into the error parameters.** `model.py` lines 192–197: records/central/federated have `err = 0` by fiat; norms use `eps_code = 0.02`; "none" uses `eps·|dialect_i − dialect_j|` with `eps = 0.25`. Decode-failure shares in the main design are exactly these constants (none 0.096, norms 0.041, records 0; recomputed), and the `eps` ablation (Table 24, p. 49) shows the ranking reverses at `eps = 0.05`. Table 20's "higher success in 99.3% / 100% of pairs" (p. 45) restates the parameter order. The paper half-admits this (p. 49) yet the abstract says the experiments "confirm" the theorems. Fixable: sweep `eps_code` against `eps` and report where the ranking flips.
	13	+
	14	+**M2. Message and bit accounting differs by regime, and the differences drive the contrasts.** (a) *Acknowledgements:* `msgs_per_recip = 1 if use_records else 2` (line 200); none/norms also load the initiator's queue with k−1 acks (line 206). At `bandwidth = 2` this saturates the flat regimes: success is 0.14 (none) / 0.15 (norms) vs 0.94 (records) in bandwidth-2 cells, against 0.90 / 0.96 / 1.00 at bandwidth 8 (recomputed). Half the 864 pairs in Table 20 compare a saturated queue with an unsaturated one; that is where "records raise success by 0.43" comes from. (b) *Relay bits:* line 236 computes bits identically for records, central and federated, while messages count 2·(depth − LCA level) hops per recipient (lines 212–218). Table 26 (p. 72) shows bits/task = 15.5 for central at every N while messages go 8 → 16. "Indistinguishable from records in success and bits" (p. 45) and "bits per success are flat in N" are artefacts of not charging bits per hop. (c) Only the LCA administrator is charged load (line 224); intermediate administrators on the path carry none.
	15	+
	16	+**M3. `admin_capacity_ratio` is circular; the "phase transition" is Theorem 5(i) by construction.** Line 146 sets `admin_bandwidth = ratio × 2λN(k−1)`: capacity is defined as a multiple of expected root load. Table 23 / Figure 8a is then the deterministic Lindley recursion; the kink sits at ratio ≈ 0.75, not 1, because N/4 producers are withdrawn (lines 155–156), so the x-axis label "ρ⁻¹" is off by that factor. Federated never saturates even at ratio 0.25 (Table 23) because *every* module administrator receives the system-wide bandwidth (lines 174–175, 229) while handling one-eighth of the load; "federated routing degrades gracefully" (p. 46) is a gift of the parameterization. P. 54 concedes "confirmed, not discovered", but §8.3 and Table 25 present it as a result.
	17	+
	18	+**M4. The network-shock result is a floor effect with the wrong mechanism attached.** Table 22 and p. 46 say link removal lowers success by ≈0.08 under hierarchies and records "and by almost nothing under flat regimes" because load concentrates on surviving neighbours. Disaggregating the DiD by bandwidth (from `runs_main.parquet`): at bandwidth 8 the effect is 0.000 for every regime except federated-0.5 (−0.02); at bandwidth 2 it is −0.14/−0.15 for records/central/federated and −0.01 for none/norms — which are already at 14% success and cannot fall. The shock (line 166) also leaves the tree built from `pos` (line 150) untouched, so hierarchies are never disrupted.
	19	+
	20	+**M5. The distribution shock is a no-op in half its cells, and Table 22 juxtaposes incommensurable columns.** With `zipf = 0` the plan distribution is uniform and every cyclic shift gives D_KL = 0 (recomputed: 0.000 for zipf 0, 1.030 for zipf 1; "0.52" on p. 44 is their mean; the target of 1.0 is hit nowhere). In shocked cells the post-window holds 10 stale and 10 re-learned periods (`adapt_delay = 10`), so Δbits/task ≈ 2 recipients × ½ × KL ≈ KL: the "0.523 vs 0.515" equality is a coincidence of k−1 = 2 and a half-stale window, not a check of Theorem 3. The ablation with `adapt_delay = 1000` (1.90 vs 1.98 per recipient) is the genuine comparison, and it evaluates the identity E_q[⌈−log₂p⌉] − E_q[⌈−log₂q⌉] ≈ D_KL rather than testing anything.
	21	+
	22	+**M6. The Theorem 1 check on p. 46 is wrong.** "The Shannon code is *longer* than the six raw bits (6.24), which is the exception the theorem allows." Theorem 1(ii)–(iii) concerns the *optimal* prefix code, and L*_p ≤ ⌈log₂ s⌉ always, since the fixed-length code is prefix-free. The simulation uses Shannon lengths (line 74); `huffman_lengths` exists but is unused (line 78). Running it: s = 64, zipf 0.5 gives H = 5.78, Shannon 6.24, **Huffman 5.81 < 6**. The theorem allows no exception; the simulation's code does. Either restate Theorem 1 for Shannon codes or use Huffman in "norms".
	23	+
	24	+**M7. Theorem 4(i) is a straw man the simulation contradicts.** A k-participant task never costs N−1 messages (p. 16); flat regimes use 2–4 messages per task, hierarchies 7–16 (Table 26), so the remark "a hierarchy cuts the message cost from linear to logarithmic in N" (p. 17) is false for the simulated task process. Theorem 1(i)'s "codeless baseline" also presupposes an agreed fixed-length enumeration — itself a shared code — so "none" is a 6-bit code with noisy decoding.
	25	+
	26	+**M8. Welfare is uninformative about communication.** With `cost_per_bit = 0.002` and value 1 per successful participant, bits are ≈1% of welfare (recomputed). "Raise welfare in every pair" (p. 45) restates "raise success"; Table 21 (norms: welfare-loss 0.06, bits-loss 1.00) leaves the two metrics unreconciled.
	27	+
	28	+**M9. One of three forward-chaining splits failed silently; the text reports three.** `h2_modern_cv_folds.csv` rows 0, 3, 6, 9, 12, 15: the `train≤1945, test 1946–1975` split errored for every model ("predict requires that you use a DataFrame…" — region 4 has no training rows before 1946, so the region FE cannot be evaluated; lines 124–125 swallow the exception). Table 16 shows "Folds 2"; its caption says "three temporal splits" and p. 39 says "in any of the three temporal splits". A win share of 0.50 over two folds is one fold; the 80% rule is meaningless at n = 2 and marginal at n = 6. The Seshat 10×5-fold run lost 14 of 420 model-folds to the same error, hence "48" in Table 11.
	29	+
	30	+**M10. `t_since` is imputed for one third of the sample.** `h2_modern.py` line 198 fills "no prior observed event" with the *full-panel maximum* (2.18 = 218 years); 33% of country-years take that value (event rate 0.095 vs 0.232 elsewhere; recomputed from `modern_h2_sample.parquet`). The code comment says "flagged in the paper"; grep of `paper/sections/*.tex` finds no flag. This is a regressor-creating imputation inside the primary cubic, and the fill value uses future data in forward chaining. Use time since panel entry plus a left-censoring indicator.
	31	+
	32	+**M11. Leakage and amendment bookkeeping.** Standardizations, the D-index PCA loadings (`build_indices.py` lines 32–36, 98–99) and `P_placebo_z` (line 202) are fitted on the full sample before folds are formed. Z-scoring is prediction-invariant for a logit with intercept, so this is minor for C1/B1/B2, but it affects the PC1 and placebo variants and the composite weighting of P; the fold-internal refit (A-001) covers Seshat LORO only. Appendix A (p. 70) says "no post-hoc amendments"; `decisions.md` records A-001 as made "after seeing the main CV results". One must change. Modern LORO folds do drop region FE correctly (line 273).
	33	+
	34	+**M12. IRT.** The "mild priors" (p. 25) are log-normal sd 0.5 on a and N(0, 2²) on b with hard clips a ∈ [0.05, 8] (`irt.py` line 49); five items still sit at a = 8 (p. 71), a boundary signature of local dependence among the jointly-coded writing items. No Q3 or residual-correlation diagnostic is reported; a testlet/bifactor structure is the standard remedy. The invariance check quotes r = 0.88 but omits the mean absolute difficulty shift of 1.16 in `measurement_diagnostics.json` (Figure 9): the fits are not linked. Missingness is called MAR "by design" (docstring line 3) although coded-item count is itself a complexity proxy (Figure 1).
	35	+
	36	+**M13. Reproducibility.** Seeds are fixed and the simulation is deterministic (tested). But `make all` needs network for `fetch` and `paper` (`build_bib.py` line 55 calls Crossref); `n_jobs = 12` is hard-coded; `pymc`/`arviz` are pinned but unused; `replication/` holds only a README; `scripts/` (named in `CLAUDE.md`) does not exist; the "three adversarial reviews … included in the package" (p. 54) were absent before this one; tests cover no econometric script on data, no hierarchical accounting, and no paper-vs-results consistency.
	37	+
	38	+**M14. Number consistency.** Matching: Tables 20–24, 26 vs `simulation_summary.json`, `ablation_admin_capacity.csv`, `regime_by_N.csv`; Tables 14–16, 18–19 vs `h2_modern_*.csv`, `h4_modern_summary.json`; Table 5 vs `measurement_diagnostics.json`. **Mismatch:** the abstract and p. 37 say "2,692 instability events"; the files give 1,676 raw events (176 + 599 + 1,041, overlaps removed); 2,692 is the number of country-years whose five-year window contains an event (Table 14 note). Also p. 44's single "0.52 bits" (see M5).
	39	+
	40	+## 3. Minor issues
	41	+
	42	+- `n_admin` is capped at N/4 (line 156) while the tree has 73 (N = 200) or 585 (N = 800) internal nodes.
	43	+- `root_load_mean` is an end-of-period backlog (stock), labelled "Root load" (flow) in Table 23.
	44	+- Forward chaining scores 1976–2005 with the 1970 decade level (line 267); fair across models, but the calibration slopes are partly luck.
	45	+- H4 modern: 3 × 5 country-blocked folds re-partition the same countries; the 15 folds are not independent.
	46	+- FutureWarnings in `seshat.py` lines 86–88 will become errors in pandas 3.
	47	+
	48	+## 4. Questions to the authors
	49	+
	50	+1. Why do only the flat regimes pay an acknowledgement round, and what is the historical analogue of a record that removes both confirmation and decoding error?
	51	+2. What are the success and bit rankings when bits are charged per hop and load per intermediate administrator?
	52	+3. `h2_modern.py` line 259 calls the decade-FE substitution "pre-registered deviation A-001", but A-001 in `decisions.md` is the leakage check. Which is it, and where is the decade substitution logged?
	53	+4. How many of the 2,692 event-windows are a single event counted up to five times, and does country clustering absorb this mechanical autocorrelation?
	54	+
	55	+## 5. Recommendation
	56	+
	57	+**Major revision.** The empirical negative result is sound and valuable; the simulation section overstates what it shows. Mandatory changes:
	58	+
	59	+1. Rewrite §8 and the abstract so no simulation result is described as "confirming" or "testing" a theorem whose premise is a model parameter (M1, M3, M5); present the `eps`/`eps_code` and capacity-ratio sweeps as the actual content.
	60	+2. Fix the accounting — bits per hop, load per intermediate administrator, acknowledgement round as a regime-independent switch — re-run the main design, and report success and bits by bandwidth cell so floor effects (M2, M4) are visible.
	61	+3. Correct the Theorem 1 statement on p. 46 (M6): use Huffman lengths in "norms" or restate the theorem for Shannon codes.
	62	+4. Repair the failed forward-chaining split, report the number of folds actually run, and replace the `t_since` max-fill with a left-censored spell plus indicator (M9, M10); re-report Tables 14–16.
	63	+5. Reconcile Appendix A with `decisions.md`, correct "2,692 instability events" to event-windows, and add a test asserting that `paper/tables/*.tex` equal the `results/` files.

added docs/reviews/review_historical_anthropologist.md +60 −0

		@@ -0,0 +1,60 @@
	1	+# Review of UQO Working Paper 35, "The Social Compression Hypothesis" (v1.0, November 2026)
	2	+
	3	+**Reviewer persona:** historical anthropologist / archaeologist (Seshat, Ethnographic Atlas, comparative method, collapse literature).
	4	+**Materials read:** `paper/main.pdf` (abstract, §1–§2, §3 skimmed, §4–§7, §9–§11, appendices); `research/preregistration.md`; `research/measurement.md`; `data/metadata/data_dictionary.md`; `results/diagnostics/coverage_summary.json` and `seshat_coverage_by_variable.csv`; `results/empirical/h2_seshat_y2_y3.csv` and `h2_seshat_robustness.csv`; `src/wp35/harmonization/seshat.py`; `configs/regions.yaml`; spot checks on `data/processed/*.parquet`.
	5	+
	6	+## 1. Summary
	7	+
	8	+The paper reads institutions as compression devices, derives five textbook results (Shannon, Kraft, Cover–Thomas, Lindley/Loynes), and pre-registers a test of whether a demand–capacity "mismatch" index built from Seshat Polaris-2026 (and separately from V-Dem/Maddison) predicts instability. The honest headline is negative: the index adds nothing beyond polity scale in Seshat or beyond state capacity in the modern panel (§6.2, §7.1, §9.2). I commend the frozen pre-registration, the refusal to treat SU as absent, and the self-critical §5.5 and §10. But the problem is deeper than "the residual carries no signal." The capacity side is built from presence codes that saturate for exactly the polities that have CrisisDB records; the outcome is a share of fourteen non-commensurable, documentation-sensitive items whose very heterogeneity the paper's own cited source (Hoyer et al. 2024) establishes; "pre-crisis" is, by the authors' admission, mostly "whole-polity"; and the one secondary result that is strongly significant (capacity predicts *shorter* polities, HR 1.75, p = 0.005) is explained away in two sentences. The paper is salvageable as a measurement paper about what Seshat can support, but the historical part needs reworking.
	9	+
	10	+## 2. Major issues
	11	+
	12	+**M1. Comparability of "capacity" across 13,000 years and 42 regions.** `K` is the mean of four shares of presence codes (§4.3, p. 24). A `postal_station` coded P for Achaemenid Persia, the Roman *cursus publicus* and the Qing *yizhan* are the same unit of capacity, although their reach differs by orders of magnitude, and `road` P for a Mississippian chiefdom and for the Inca *qhapaq ñan* are the same. §10 concedes "existence, not reach" but does not draw the consequence: presence codes saturate. In the crisis sample `K_mean` averages 0.71 with 46% of records above 0.8 (my check on `seshat_h2_sample.parquet`), so within the sample that matters, `P = z(D) − z(K)` is approximately `z(D)` minus a near-constant. The main finding ("P predicts what D predicts", p. 34) is then close to a mechanical consequence of ceiling effects, not a discovery about mismatch. The IRT "invariance" check (Figure 9, p. 71; mean absolute difficulty difference 1.16 logits, almost every New-World point above the diagonal) shows non-invariance, not invariance; r = 0.88 of difficulty orderings is not measurement equivalence. Fixable in part: report K's variance within the crisis sample; the graded reach variables (`largest_communication_distance`, `fastest_individual_communication`: 27 and 11 polities) are too few, which is itself the point.
	13	+
	14	+**M2. Undated facts and the "pre-crisis predictors" claim.** The abstract says "127 CrisisDB crisis records with pre-crisis predictors." Only 2,503 of 56,274 facts are dated (Table 4); in the pre-crisis aggregates used for H2, 4.1% of facts are dated (my check on `seshat_precrisis_agg.parquet`), and `aggregate_polity(before_year=...)` in `seshat.py` keeps *every* undated fact. A fact dated 1500–1700 is dropped for a crisis in 1600 (`year_to < crisis_start`) while an undated "paper currency: P" that may post-date the crisis is kept. Two crises begin before the polity's `start_year` (`years_since_start = −28`) and twenty within 25 years. The strict dated-only test "was not estimable" (p. 53; audit line 45). Temporal precedence is not weak, it is absent for K; the abstract and §1 should say "whole-polity capacity." Fixable only in the text.
	15	+
	16	+**M3. CrisisDB is crisis-selected and the "severity share" is not a scale.** The selection caveat is stated (pp. 21, 53); what is never tested is whether the 14 items form one dimension. On `seshat_crisis.parquet` the mean inter-item correlation is 0.08 (range −0.22 to 0.80). `collapse` is present in 7 of 165 records and `century_plus` in 10 of 163; `civil_war` in 84 of 159, `uprising` 77 of 158, `depose` 70 of 146. The "severity share" is therefore a count of elite-political events (deposition, assassination, civil war, uprising) with a few rare catastrophic items thrown in at equal weight. Item-by-item, D correlates +0.26 with `assassination`, +0.25 with `depose`, +0.22 with `civil_war`, and −0.22 with `century_plus`, −0.06 with `collapse`. Large polities have emperors to assassinate, capitals to lose and elites to demote; a chiefdom that is conquered has nothing else to lose. "Decline" is not scale-independent either: it is coded SU for 42 records, the most of any item, so the denominator moves with documentation. Y1b has a 77% base rate and discriminates nothing (AUC 0.62–0.64 for every model, Table 11). Fixable: item-level models, a catastrophic vs. elite-political split, a dimensionality diagnostic; drop the word "severity."
	17	+
	18	+**M4. Coder circularity between Polaris institutional codes and CrisisDB.** §5.5 names it ("same project, overlapping historiography") but underestimates it. 26% of the facts behind the K items are inferred (IP/IA; 29% for the communication block; `seshat_coverage_by_variable.csv`: `full_time_bureaucrat` 147/434, `judge` 144/377, `courier` 91/311, `postal_station` 80/254). Inference in Seshat typically runs "a polity of this size/era must have had X," i.e. from scale to capacity, which contaminates H1 directly; it is the communication block, the most inferred, that carries the entire H1 association (Table 7, H1-S6: 1.09 vs. 0.15/−0.29/0.22). Worse, p. 32 claims the H2 coefficient is "stable under ... lower and upper range bounds, inferred codes dropped, within-era and within-region standardization, ranks" and points to Table 10, but Table 10 and `h2_seshat_robustness.csv` contain none of these rows (eleven variants, none of them). The labels exist in `h2_seshat.py` line 74, so the runs appear not to have produced output. Under the repository's own rule 1, that sentence must be removed or the rows produced.
	19	+
	20	+**M5. The polity as unit, chained polities, and the Cox result.** Seshat polities are periodization decisions (Qing 1/2 split at 1796, Tang 1/2 at 763, Habsburg 2 1649–1918 with three crisis records sharing one whole-polity P). Y2 and Y3 are pre-registered secondary outcomes and cannot be disowned after the fact; yet the one strong result in `h2_seshat_y2_y3.csv`, K_mean_z = 0.559 (SE 0.20, HR 1.75, p = 0.005; D null at 0.128), is dismissed in two sentences as "periodization" (p. 37). Either polity duration is an outcome, in which case capacity shortening polities contradicts the framework and agrees with Tainter, or it is not, in which case it should not have been pre-registered. Y3 is thin: `relationship_to_preceding` is None for 286 of 626 polities, and the 37 "discontinuities" include Yuan→Ming ("cultural assimilation") and Shang→Zhou ("elite migration"), which are not instability codes. Fixable: re-run Y2 without chained/quasi-polities and discuss it, or move Y2/Y3 to an appendix with a logged amendment.
	21	+
	22	+**M6. Galton's problem and `configs/regions.yaml`.** Ten macro-regions were fixed as "geographic only." Southwest_Asia pools Anatolia-Caucasus, Mesopotamia, Levant, Iran and Arabia: Hittites, Achaemenids, Umayyads and Hadramawt kingdoms share one fixed effect and one hold-out fold. Southeast_Asia_Oceania pools Angkor and Pagan with Polynesia and New Guinea; Central_Asia pools Siberia with Turkestan; Sahel sits in North_Africa while the Mali empire (`West Africa`) sits in Sub_Saharan_Africa. The real non-independence is cultural-historical: after c. 700 CE, Southwest Asia, Central Asia, North Africa and South Asia share Islamic administrative templates (coinage, qadi courts, barid); leave-one-macro-region-out does not remove that diffusion. Regional fixed effects with 42 clusters and 127 records (Europe 31, East Asia 26, Southwest Asia 21 = 61% of the sample; Sub-Saharan Africa 111 polities but 5 crisis records) are anti-conservative, as §10 admits. Fixable: add cultural-sphere (Islamic, Sinitic, Roman/Christian, Indic, Mesoamerican, Andean) blocks as a second hold-out scheme, and use NGA random effects as primary rather than alternative.
	23	+
	24	+**M7. Ethnographic Atlas.** EA031 "open categories use the bound" (data dictionary B), so everything above 50,000 is coded 50,000; and the Atlas includes archaic societies with focal years to −2000. EA033 (jurisdictional levels) is treated as "capacity" (p. 24), but for Johnson and Carneiro it is the *response* to scalar stress, a scale measure; H1-E reproduces Carneiro (1967) and Currie et al. (2010) and says nothing about compression. Glottolog families are a weak Galton control (Austronesian and Indo-European span oceans; many isolates are singleton clusters). Present EA as replication of a known regularity, not a third test.
	25	+
	26	+**M8. Engagement with Hoyer et al. (2024).** Cited twice (pp. 7, 10) for "no typical societal crisis," but the implication is never drawn: if consequences are heterogeneous and mutually uncorrelated, a summed share is not a severity measure (M3), and the paper is predicting an average of unrelated things. Hoyer et al. also find that administrative complexity does not predict consequences one at a time — which is the K-only null in Table 9 restated, and should be acknowledged as a replication, not a novelty.
	27	+
	28	+**M9. Alternative mechanisms for "larger polities have more severe crises."** Before any coordination mechanism a historian would offer: (i) documentation — more historiography means more items coded P rather than SU (`y1_n_coded` correlates 0.24 with D; coded SC facts 0.31); (ii) item relevance (M3); (iii) composition — multi-ethnic empires fragment because they have parts; (iv) frontier exposure scaling with territory (Turchin 2009); (v) elite overproduction scaling with elite size; (vi) crisis identification — CrisisDB episodes are delimited *because* consequences occurred. None is in §9.2, which stops at "Johnson's reading."
	29	+
	30	+**M10. Does "institutions as compression" add anything beyond scalar stress?** The paper concedes (§3.8, §9.1, p. 51) that no Shannon quantity is measured historically and that the simplest reading is Johnson's. Its one distinctive prediction — the stale-code penalty (Theorem 3): a fiscal or legal code optimized for one task distribution becoming costly after a shock (the Qing land-tax quota frozen in 1712, the Ottoman timar after the price revolution) — is never confronted with history. Without that, "compression" relabels Wright & Johnson (1975).
	31	+
	32	+## 3. Minor issues
	33	+
	34	+- Pre-registration §3.5 says "13 instability items" and lists 14; code and paper use 14. Log as an amendment.
	35	+- Table 4 reports 132 unparsed values but `results/diagnostics/seshat_unparsed_values.csv` is empty (header only); reconcile.
	36	+- `code_presence` maps "SA"/"SP" to missing without documenting them; `crisis_table()` has a no-op `epidemic_v` line.
	37	+- A polity's value for a presence variable with several facts is the mean across facts (A then P → 0.5), i.e. a time average, not status at the crisis; say so.
	38	+- Era blocks use the polity midpoint: Habsburg 2 is "post-1800" for a 1700 crisis.
	39	+- Table 12: "pooled" is the sample minus 5–13 records, so stability is near-guaranteed; the within-block column is the informative one and is mostly uncomputable.
	40	+- The Equinox-2020 replication (pre-registered robustness item 8) was simply not run (audit line 28); the paper calls it "blocked," which it was not.
	41	+- The 120 quasi-polities (p. 21, from −13,600) deserve a sentence, not only a sensitivity row.
	42	+
	43	+## 4. Questions to the authors
	44	+
	45	+1. What is the SD of K within the 127 crisis records, and what share of P's variance in that sample is D alone?
	46	+2. Where are the inferred-dropped, bounds, within-era, within-region and rank rows claimed on p. 32?
	47	+3. Which items does D predict when the 14 are modelled separately, and does the result survive excluding `depose`, `assassination`, `civil_war`, `uprising`?
	48	+4. How do you reconcile HR 1.75 for K with a framework in which capacity lowers breakdown probability?
	49	+5. Why is polity-level documentation (number of coded facts, or share SU) not a covariate given that it correlates with both D and the outcome denominator?
	50	+6. Would the H1 communication-block result survive dropping the 29% inferred codes in that block?
	51	+
	52	+## 5. Recommendation
	53	+
	54	+Major revision. The negative result is real and worth publishing, but the historical analysis currently overstates what was measured. Mandatory changes:
	55	+
	56	+1. Remove or substantiate the robustness claims on p. 32 that have no rows in Table 10 / `h2_seshat_robustness.csv` (inferred codes, bounds, standardization windows, ranks); produce the runs or strike the sentence.
	57	+2. Replace "pre-crisis predictors" (abstract, §1) with "whole-polity predictors" and report the dated-fact share in the H2 sample.
	58	+3. Report the outcome item by item and a dimensionality diagnostic; retire the single "severity share" as primary or justify it against Hoyer et al. (2024).
	59	+4. Treat the Cox result for K (HR 1.75) as a finding to be explained, with a chained-polity/quasi-polity exclusion, or formally amend the pre-registration to drop Y2/Y3.
	60	+5. Add a documentation covariate and a cultural-sphere hold-out scheme alongside the geographic macro-regions, and state explicitly that K is a saturated presence index within the crisis sample.

added docs/reviews/review_institutional_economist.md +60 −0

		@@ -0,0 +1,60 @@
	1	+# Referee report — UQO WP 35, "The Social Compression Hypothesis" (v1.0, Nov. 2026)
	2	+
	3	+**Reviewer persona:** senior institutional economist (transaction costs, state capacity, political economy of institutions, applied econometrics).
	4	+**Materials read:** `paper/main.pdf` (abstract, §1, §3–§7, §9–§11); `research/preregistration.md`, `measurement.md`, `identification.md`, `decisions.md`; the five result files named in the brief plus `h2_modern_cv_folds.csv`, `h2_seshat_cv_folds.csv`, `h2_seshat_cv_noleakage.csv`; `src/wp35/measurement/build_indices.py`, `src/wp35/econometrics/common.py`, `h2_seshat.py`, `h2_modern.py`, `leakage_check.py`.
	5	+
	6	+## 1. Summary
	7	+
	8	+The paper reads institutions as devices that compress the communication cost of coordination, formalises this in five standard information-theoretic results plus three conjectures (§3), and pre-registers a test of the empirical corollary: a mismatch index P = z(log D) − z(K) between coordination demand (scale) and coordination capacity (administration, rules/money, communication infrastructure, records) should predict instability beyond scale-only, capacity-only, complexity-only and persistence baselines, judged out of sample by an 80%-of-folds rule. On Seshat Polaris-2026 (127 CrisisDB records) and a V-Dem–Maddison panel (14,637 country-years), P is positively associated with crisis severity (0.20/SD, Table 9) and with five-year instability onsets (0.40/SD, Table 14), but fails the out-of-sample rule against scale alone (Seshat) and capacity alone (modern); a placebo performs comparably, the negative control loads as strongly as the forward test, and an Oster adjustment crosses zero. The paper concludes the universal framing is unsupported and keeps only "scale matters, capacity matters." The candour and pre-registration are welcome; the problems below concern whether the tests test what they claim.
	9	+
	10	+## 2. Major issues
	11	+
	12	+**M1. D and K are not substantively disjoint, and K is size-biased by construction.** The sets are *nominally* disjoint (prereg §3.1–3.2), but `settlement_hierarchy` (in D) and `administrative_level` (in K) are both hierarchy depth, and most K items are polity-level *presence-anywhere* codes (a road, a court, a coin "exists somewhere"). The probability that at least one instance exists rises mechanically with territory and population, so K is a saturating function of D before any economics enters. This, not only co-evolution, explains much of corr(D,K) = 0.84 (Table 5), makes the H1 "primary test" unfailable, and contaminates P at the measurement stage. The paper notes presence ≠ reach (p. 53) but not this implication. Fix: normalise presence items by log territory (item-level logit with a scale offset) or restrict K to intensity items, and re-run H1 with hierarchy items removed from both sides. K–Y overlap is also real in the modern panel: `v2svstterr` (territorial control) is close to a contemporaneous measure of civil war, and V-Dem coders rate K knowing what followed (acknowledged p. 53, not addressed).
	13	+
	14	+**M2. The Seshat placebo is a tautology.** `h2_seshat.py` line 37 defines `P_placebo = z(D) − z(X_prod_infra)`; line 54 fits it with `XS = X_warfare + X_prod_infra + epidemic_v` as controls. With `X_prod_infra` both inside the placebo and a free regressor, the model is an exact reparametrisation of B1 (D only + XS). `h2_seshat_primary.csv` confirms it: both coefficients equal 0.25723724…, with identical SE, n = 94 and log-likelihood −37.6295. The sentence "the placebo … is associated with severity at 0.26 — the same as scale alone" (p. 33) and the abstract's "a placebo index performs comparably" carry zero evidential content. Conceptually it is also poorly chosen: irrigation, markets and food storage are coordination institutions (Ostrom). The modern placebo (polyarchy for K, run without the polyarchy control, line 72) is valid.
	15	+
	16	+**M3. P is not a coherent economic object, and the paper already contains the better test.** ρ = D/K needs common units (Def. 4); P imposes the exchange rate 1 SD of D = 1 SD of K, which has no theoretical basis (the "Limitation" box, p. 13, concedes this). The empirical content of P is the restriction β_D = −β_K in z-units. B5 reports the unrestricted coefficients: Seshat 0.34 (SE 0.16) and −0.21 (SE 0.19); modern 0.17 (SE 0.07) and −0.71 (SE 0.14). In Seshat the restriction is *not* rejected and capacity has the predicted sign; in the modern panel it is rejected because capacity dominates. A Wald test of β_D + β_K = 0, or a capacity-shortfall index K − E[K | D] (orthogonal to scale by construction), would replace the fold-counting apparatus with one interpretable number. Also, C3 "P given K" is algebraically B5 (same log-likelihood −5627.075 in `h2_modern_primary.csv`) yet p. 38 presents it as a separate finding, and Table 9's row "A3: Y1 ~ P + D + K + X + FE" is mislabelled — the code (`h2_seshat.py` line 48) conditions on `complexity_pc1_z`; P + D + K would be perfectly collinear.
	17	+
	18	+**M4. Rival mechanisms are weakly represented.** The "warfare technology index" is a technology-presence share, not military competition (wars fought, neighbours' power) in the Tilly/Gennaioli–Voth sense; elite overproduction is absent (acknowledged); fiscal capacity is absent from Seshat K but inside modern K (`v2stfisccap`), so that rival is inseparable in both panels. Worse, `epidemic_v` is itself a CrisisDB consequence item coded with the outcome (`h2_seshat.py` line 26), so A2 conditions on a component of the outcome vector. Adding X drops 40 of 127 records (A2 n = 87) while B1–B4 use n = 93–100, so p. 28's "fitted on the same observations" holds in CV but not in Table 9.
	19	+
	20	+**M5. K–democracy collinearity, negative control and Oster: wrong lessons drawn.** corr(K, polyarchy) = 0.75; `v2elembcap`, `v2cltrnslw`, `v2clrspct` are components of V-Dem's liberal/electoral indices, so "controlling for polyarchy" partials out a near-component of K (C2 0.72 → C1 0.40). The negative control was pre-registered as "must not show a *stronger* association than the forward spec" (prereg §5.4). It did not: 0.44 (SE 0.10) past vs 0.48 (SE 0.10) future. By the authors' own rule it *passed*; p. 39's claim that "the pre-registration said in advance that a negative control of this kind would be read that way" is inaccurate — amend the rule (logged) or report the pass. The control is also uninformative on its own terms: instability destroys capacity (§3.8), so a backward association is expected under the hypothesis and under confounding alike. The Oster exercise treats the drop from adding polyarchy/GDP as proportional selection (δ = 1), implausible when the key observable is quasi-collinear with the regressor; the abstract's "an Oster adjustment reverses its sign" overclaims an assumption-laden point estimate. The Cinelli–Hazlett RV = 0.072 is twelve times P's partial R² of 0.006 and should be benchmarked against observed covariates.
	21	+
	22	+**M6. The 80% rule is a sign test with granularity problems, and the reported fold counts are wrong.** The rule ignores magnitude (mean Δ log-lik per observation +0.002 to +0.011 in Seshat, Table 11). With 10 LORO folds it means 8/10; with 6 modern regions 5/6 passes and 4/6 fails; with forward chaining it means "win both" — and the first forward split (train ≤ 1945) **failed for every model** with a statsmodels predict error (`h2_modern_cv_folds.csv`; `folds = 2` in `h2_modern_cv_wins.csv`), yet p. 39 says "in any of the three temporal splits" and Table 16's caption says "three temporal splits". Seshat's "fifty cross-validation folds" (p. 33) are 48 (rep0_fold1 and rep9_fold1 errored). Further: B5 (D + K) *nests* P, so requiring P to beat it out of sample asks a one-parameter restriction to beat its unrestricted version — P winning 75% of folds against B5 is mild evidence *for* the restriction; and random 5-fold CV is not blocked by polity although 16 polities contribute multiple records with identical undated whole-polity P (`load_crisis`), so LORO is the only clean scheme. Use paired fold-difference tests plus the Wald test (M3), repair the failed folds, and correct the text.
	23	+
	24	+**M7. Holm and the "three primary tests".** H1-Seshat cannot realistically fail (collinearity 0.84 was known from Turchin et al. 2017), so one of the three Holm-corrected primaries is a free pass. More seriously, the Holm correction is never computed: `common.py` defines `holm()` but no script calls it and no result file contains it; p. 32's "survives the Holm correction" is asserted, not produced. Since the binding criterion is the out-of-sample rule (no p-value), the multiple-testing apparatus is decorative. H4-modern met the pre-registered rule (hinge 12/15 = 0.80, spline 0.87, `h4_modern_summary.json`) but pp. 42–43 reinterpret it as "negligible" — a post-hoc deviation not logged in `decisions.md` (only A-001 is).
	25	+
	26	+**M8. Small-sample inference.** A2 fits about 20 parameters (10 macro-regions, 6 eras, 3 covariates) on 87 fractional observations with 31 clusters; no wild-cluster bootstrap or randomisation inference; Y1b needed era effects dropped for separation; the fractional logit does not weight by the number of coded items (3-item and 14-item records count equally); H5 within-block estimates on 5–13 records (Table 12) are noise.
	27	+
	28	+**M9. Leakage in cross-validation: real but immaterial.** `build_indices.py` fits the D loadings (`pc1_pairwise`, line 65), the z-scores of D and K (lines 88–89), the IRT item parameters (line 80) and the complexity PC1 (line 123) on the full polity sample before any fold is drawn; `cv_compare` in `common.py` then uses these fixed columns. This is transductive preprocessing, but it is outcome-blind, an affine rescaling of one regressor is absorbed exactly by the GLM slope and intercept, it is shared by every compared model, and the authors' own fold-internal check (`leakage_check.py`; `h2_seshat_cv_noleakage.csv`: P beats D-only in 2/10 vs 3/10) confirms the verdict is unchanged. The same logic covers modern forward chaining; say so once. The CV problems that matter are in M6.
	29	+
	30	+**M10. Over- and under-claiming.** Overclaims: the placebo sentence (M2), "reverses its sign" (M5), "three temporal splits"/"fifty folds" (M6), "survives the Holm correction" (M7), and "reproduces every number … with one command" (p. 55) while two folds and one split silently error. Underclaim: the interesting narrow result — in Seshat capacity given scale has the predicted sign and the restriction survives; in the modern panel capacity dominates and it fails — is buried under a fold-counting verdict.
	31	+
	32	+## 3. Minor issues
	33	+
	34	+1. Table 9 row A3 label vs code (M3); Table 16 caption "three temporal splits"; p. 33 "fifty folds".
	35	+2. `t_since` for countries with no observed prior event is set to the sample maximum (`h2_modern.py` line 52); the comment says "flagged in the paper" — I could not find the flag.
	36	+3. `K_modern` averages whichever of six items are coded (≥ 4); item composition changes over time, so K is not era-comparable without an item-availability control.
	37	+4. The Seshat persistence baseline (`prior_crisis_nga`) is built from the crisis-selected sample itself.
	38	+5. Whole-polity territory and population are means over facts, so fragmentation or conquest (Y1 items) pulls D down — circularity in the other direction.
	39	+6. Conjecture 1's falsifier promises the share of polities whose P flips sign under alternative K weightings; it is not reported.
	40	+7. Bibliography: ALL-CAPS author names (WRIGHT and JOHNSON, STASAVAGE, FEARON and LAITIN, CEDERMAN); Table 3 cells truncated; p. 3 blank.
	41	+8. The "Process" paragraph (p. 54: PI has not reviewed code or text) belongs on the title page.
	42	+
	43	+## 4. Questions to the authors
	44	+
	45	+1. Re-run the Seshat placebo without `X_prod_infra` in the controls: coefficient and LORO win share?
	46	+2. Wald p-value of β_D + β_K = 0 in B5, each panel, pooled and under LORO?
	47	+3. Why did the train ≤ 1945 forward split fail, and what are the results once fixed?
	48	+4. Where is the Holm computation, and what are the three adjusted p-values?
	49	+5. With K items normalised by log territory (or intensity-only), what is corr(D, K) and does H1 still pass?
	50	+6. Why is H4-modern reported as rejected when the pre-registered rule was met?
	51	+
	52	+## 5. Recommendation: **major revision**
	53	+
	54	+The project is well-conceived and unusually honest, and the negative conclusion may well survive, but several headline "falsifications" are implementation artefacts (M2, M6, M7) or misreadings of the paper's own rules (M5, M7), and P is tested through a restriction the paper never states as such (M3). Mandatory changes:
	55	+
	56	+1. Remove or correctly re-estimate the Seshat placebo and rewrite every sentence that relies on it (abstract, pp. 33, 51–52).
	57	+2. Make the unrestricted D + K model the primary object, with a Wald test of β_D + β_K = 0 and paired fold-difference tests; repair the failed CV folds and the forward split, block k-fold by polity, and correct all fold counts.
	58	+3. Compute and report the Holm correction or delete the claim; report the negative control and H4-modern against the pre-registered rules as written, logging any reinterpretation as a numbered amendment.
	59	+4. Address the size bias of presence-anywhere K items (M1) with at least one normalised or intensity-only K, and discuss the D/K hierarchy overlap.
	60	+5. Fix Table 9's A3 label, present C3 as the reparametrisation of B5 it is, fit the Table 9 baselines on the common sample, and remove `epidemic_v` from the controls (or justify conditioning on a consequence item).

modified paper/appendix/appendix.tex +1 −1

		@@ -3,7 +3,7 @@
3	3	\section{Pre-registration summary}
4	4	\label{app:prereg}
5	5
6		−The analysis plan (\texttt{research/preregistration.md}, SHA-256 \texttt{ecda8804e17f9317\ldots}) was frozen on 2026-11-01 before any association between a predictor and an outcome was computed. It fixes: the five hypotheses and their falsifiers; the units (polity, society, country-year) and the reasons for them (93.7\% of Polaris facts undated; CrisisDB crisis-selected); the disjoint indicator sets for $D$ and $K$ listed in Section~\ref{sec:data-measurement}; the outcomes $Y_1$ (severity share), $Y_{1b}$, $Y_2$ (duration), $Y_3$ (succession), $Y_4$--$Y_6$ (modern events) and the composite; the five-year horizon; the models of Section~\ref{sec:methods}; the baselines and the 80\% fold rule; the Holm correction over three primary tests; the a-priori power computation and its decision rule; the robustness set; and the rule for narrowing the claim. Seven decisions and no post-hoc amendments are recorded in \texttt{research/decisions.md}; the single deviation from the mandate's candidate formula ($K$ standardized rather than logged, because it is a bounded share that can be zero) was recorded before analysis. The use of decade rather than year effects in the forward-chaining folds, so that test years have a fitted level, is noted in the code and in Section~\ref{sec:methods-modern}.
	6	+The analysis plan (\texttt{research/preregistration.md}, SHA-256 \texttt{ecda8804e17f9317\ldots}) was frozen on 2026-11-01 before any association between a predictor and an outcome was computed. It fixes: the five hypotheses and their falsifiers; the units (polity, society, country-year) and the reasons for them (93.7\% of Polaris facts undated; CrisisDB crisis-selected); the disjoint indicator sets for $D$ and $K$ listed in Section~\ref{sec:data-measurement}; the outcomes $Y_1$ (severity share), $Y_{1b}$, $Y_2$ (duration), $Y_3$ (succession), $Y_4$--$Y_6$ (modern events) and the composite; the five-year horizon; the models of Section~\ref{sec:methods}; the baselines and the 80\% fold rule; the Holm correction over three primary tests; the a-priori power computation and its decision rule; the robustness set; and the rule for narrowing the claim. Seven pre-analysis decisions (D-001 to D-007) and seven post-analysis amendments (A-001 to A-007: fold-internal pre-processing check; corrections and additions made after the adversarial reviews, listed in Section~\ref{sec:limits} and in the response-to-reviewers document) are recorded in \texttt{research/decisions.md}. The single deviation from the mandate's candidate formula ($K$ standardized rather than logged, because it is a bounded share that can be zero) was recorded before analysis. The use of decade rather than year effects in the forward-chaining folds, so that test years have a fitted level, is noted in the code and in Section~\ref{sec:methods-modern}.
7	7
8	8	\section{Data dictionary (abridged)}
9	9	\label{app:dictionary}

modified paper/figures/fig_calibration_modern.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_dplace_focal_years.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_h2_coefplot.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_h4_curve.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_h5_forest.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_irt_invariance.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_measurement_DK.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_modern_coverage.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_seshat_missingness.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_seshat_temporal_coverage.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_sim_breakeven.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_sim_phase.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_sim_scaling.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_sim_shocks.pdf +0 −0

Binary file not shown.

modified paper/figures/fig_trajectories.pdf +0 −0

Binary file not shown.

modified paper/main.pdf +0 −0

Binary file not shown.

modified paper/main.tex +1 −1

		@@ -64,7 +64,7 @@
64	64	\newcommand{\WPdate}{November 2026}
65	65	\newcommand{\WPversion}{1.0}
66	66	\newcommand{\WPabstract}{%
67		−Institutions, standards, records and administrative hierarchies can be read as devices that lower the communication cost of coordinating many people: a form of compression. This paper states that reading formally, derives what it does and does not imply, and tests its empirical corollary---that a mismatch between a society's coordination demand (population, territory, settlement structure) and its coordination capacity (administration, standardized rules and money, communication infrastructure, record-keeping) predicts subsequent instability---under a frozen pre-registration. We build demand, capacity and mismatch indices from the Seshat Polaris-2026 databank (626 polities; 127 CrisisDB crisis records with pre-crisis predictors), the Ethnographic Atlas (1,291 societies) and a V-Dem--Maddison country-year panel (14,637 country-years, 2,692 instability events, 1820--2025), with a transparent block index and a latent-variable alternative, uncertainty propagation and out-of-sample validation. Capacity and scale are positively associated in all three datasets (\EMP). The mismatch index is positively associated with crisis severity in Seshat ($0.20$ log-odds per SD, stable across regions and eras) and with instability onsets in modern states ($0.40$ per SD), but by the pre-registered criterion it fails: it does not beat the scale-only baseline out of sample in Seshat, nor the capacity-only baseline in the modern panel; a placebo index performs comparably; the modern association is as strong with past as with future events; and an Oster adjustment reverses its sign. No threshold is found in Seshat and only a negligible nonlinearity in the modern panel. Five theorems make the compression reading precise (coding gain, break-even scale, stale-code penalty, hierarchical routing, backlog dynamics), and factorial agent-based experiments (6,048 runs, five imposed institutional regimes) confirm them in silico while locating the regimes where institutions hurt: unpaid set-up costs, saturated hierarchies, and weakly structured task environments (\SIM). We conclude that the compression framework is coherent and simulable, that scale and capacity each matter, but that their mismatch, as measurable in existing historical data, carries no identifiable predictive content of its own. The universal theory is not supported; the narrower results are stated with their scope.%
	67	+Institutions, standards, records and administrative hierarchies can be read as devices that lower the communication cost of coordinating many people: a form of compression. This paper states that reading formally, derives what it does and does not imply, and tests its empirical corollary---that a mismatch between a society's coordination demand (population, territory, settlement structure) and its coordination capacity (administration, standardized rules and money, communication infrastructure, record-keeping) predicts subsequent instability---under a frozen pre-registration. We build demand, capacity and mismatch indices from the Seshat Polaris-2026 databank (626 polities; 127 CrisisDB crisis records with polity-level predictors, dated facts preceding the crisis where they exist), the Ethnographic Atlas (1,291 societies) and a V-Dem--Maddison country-year panel (14,637 country-years, 901 distinct event-years, 1820--2025), with a transparent block index and a latent-variable alternative, uncertainty propagation and out-of-sample validation. Capacity and scale are positively associated in all three datasets (\EMP). The mismatch index is positively associated with crisis severity in Seshat ($0.20$ log-odds per SD, stable across regions, eras and cultural spheres) and with instability onsets in modern states ($0.39$ per SD), but by the pre-registered criterion it fails: it does not beat the scale-only baseline out of sample in Seshat, nor the capacity-only baseline in the modern panel; the restriction it imposes on scale and capacity is rejected in the modern panel; a placebo index is not distinguishable from it in Seshat; the modern association is nearly as strong with past as with future events; and it is not robust to an Oster-type adjustment. No threshold is found in Seshat and only a small nonlinearity in the modern panel. Among Seshat polities with a crisis record, higher capacity is also associated with a \emph{shorter} polity duration, against the hypothesis. Five theorems make the compression reading precise (coding gain, break-even scale, stale-code penalty, hierarchical routing, backlog dynamics), and factorial agent-based experiments (6,048 runs, five imposed institutional regimes, equal administrative budgets) instantiate them in silico on the cost side---bits, messages, queues---while locating the regimes where institutions hurt: unpaid set-up costs, saturated hierarchies, and weakly structured task environments (\SIM). We conclude that the compression framework is coherent and simulable, that scale and capacity each matter, but that their mismatch, as measurable in existing historical data, carries no identifiable predictive content of its own. The universal theory is not supported; the narrower results are stated with their scope.%
68	68	}
69	69	\newcommand{\WPkeywords}{institutions, coordination, state capacity, social complexity, Seshat, pre-registration, agent-based simulation, information theory}
70	70	\newcommand{\WPjel}{D02, D83, H11, N40, P48, C63}

modified paper/sections/data.tex +2 −2

		@@ -49,7 +49,7 @@ Quantity & Value & Detail \\
49	49	\subsection{Coding rules and uncertainty}
50	50	\label{sec:data-coding}
51	51
52		−Presence codes are mapped to 1 (P, IP), 0 (A, IA) and 0.5 (disputed codes of the form A$\sim$P), with inferred codes flagged so that a sensitivity run can drop them; unknown (SU), blank and unparseable values are missing (132 of 56,274 facts were unparseable and are logged). Numeric ranges are summarized by their geometric midpoint for population, territory and largest-settlement population and by the arithmetic midpoint for level counts, and the bounds are kept: the paper re-estimates its main model at the lower and upper bounds. Dates are kept as intervals; a crisis is dated by the start of its coded interval; nothing is interpolated. When several facts exist for a polity and a variable, the polity value is the mean over facts (the mean log for logged variables); for each crisis record, ``pre-crisis'' aggregates use dated facts ending before the crisis start together with the undated whole-polity facts, which, given the date coverage, means that temporal precedence is weak and is acknowledged as such.
	52	+Presence codes are mapped to 1 (P, IP), 0 (A, IA) and 0.5 (disputed codes of the form A$\sim$P), with inferred codes flagged so that a sensitivity run can drop them; unknown (SU), blank and unparseable values are missing (132 of 56,274 facts were unparseable and are logged). Numeric ranges are summarized by their geometric midpoint for population, territory and largest-settlement population and by the arithmetic midpoint for level counts, and the bounds are kept: the paper re-estimates its main model at the lower and upper bounds. Dates are kept as intervals; a crisis is dated by the start of its coded interval; nothing is interpolated. When several facts exist for a polity and a variable, the polity value is the mean over facts (the mean log for logged variables); for each crisis record, ``pre-crisis'' aggregates use dated facts ending before the crisis start together with the undated whole-polity facts; only 3.6\% of the facts entering these aggregates are dated, so ``pre-crisis'' means, in practice, ``whole-polity'', and temporal precedence is weak. We keep the label for consistency with the pre-registration and flag it wherever it is used.
53	53
54	54	\subsection{Demand, capacity, and the mismatch index}
55	55	\label{sec:data-measurement}
		@@ -71,7 +71,7 @@ Index & Construction and diagnostics \\
71	71
72	72	Two alternatives to the block mean were pre-registered and are reported side by side: the first principal component of the four blocks, which correlates 0.999 with the mean, and a two-parameter logistic item-response model estimated by marginal maximum likelihood with mild priors on the thirty binary items, which handles item missingness through the likelihood rather than by imputation, is defined for 579 polities, and correlates 0.96 with the block mean. The latent score carries a median posterior standard deviation of 0.34 on a unit-variance scale, which is the honest measure of how uncertain an individual polity's capacity is. Item parameters refitted separately on the Old World and on the Americas and Oceania have difficulties correlated 0.88, so the ordering of institutions by rarity is similar across the two worlds, although the New-World sample (94 polities) is too small for a strict invariance test and the leave-region-out analyses are the operational safeguard. The block structure is stable: inter-block correlations are 0.75--0.79 overall, Cronbach's $\alpha$ across the four blocks is between 0.83 and 0.97 in every macro-region, and dropping any block leaves $K$ correlated at least 0.98 with the full index.
73	73
74		−The central measurement fact of the paper is collinearity (Figure~\ref{fig:measurement}, Table~\ref{tab:measurement}). Scale and capacity correlate 0.84 across polities, as the single-dimension result of \citet{turchin2018} implies, so the mismatch index retains only 17\% of the variance of $z(D)+z(K)$; it correlates 0.39 with $D$, $-0.18$ with $K$ and $-0.04$ with a Turchin-style first principal component of 57 social-complexity indicators. $P$ is a residual, nearly orthogonal to overall complexity. That is what would make it informative if it carried signal, and what makes it fragile: measurement error in either index leaks into it, and any association with $P$ can be a disguised association with one of its components. The pre-registered baselines (scale only, capacity only, complexity only, scale and capacity entered separately) exist to detect exactly this.
	74	+The central measurement fact of the paper is collinearity (Figure~\ref{fig:measurement}, Table~\ref{tab:measurement}). Scale and capacity correlate 0.84 across polities, as the single-dimension result of \citet{turchin2018} implies, so the mismatch index retains only 17\% of the variance of $z(D)+z(K)$; it correlates 0.39 with $D$, $-0.18$ with $K$ and $-0.04$ with a Turchin-style first principal component of 57 social-complexity indicators. $P$ is a residual, nearly orthogonal to overall complexity. That is what would make it informative if it carried signal, and what makes it fragile: measurement error in either index leaks into it, and any association with $P$ can be a disguised association with one of its components. The pre-registered baselines (scale only, capacity only, complexity only, scale and capacity entered separately) exist to detect exactly this. The problem is sharper in the crisis sample: polities with a CrisisDB record are better documented and more complex than the average polity, their mean capacity share is 0.71 against 0.58 overall, and 40\% of records have $K$ above 0.8, so within that sample $P$ correlates 0.63 with $D$ and 0.02 with $K$---the mismatch index is, among crisis cases, close to a rescaled measure of scale (last row of Table~\ref{tab:measurement}).
75	75
76	76	\begin{figure}[H]
77	77	\centering

modified paper/sections/introduction.tex +2 −2

		@@ -9,8 +9,8 @@ We proceed in the order a skeptical reader would want. First, we make the compre
9	9
10	10	Second, we pre-registered the empirical tests before computing any association between predictors and outcomes (the frozen document and its hash are in the replication package). The pre-registration fixes the datasets, the unit of analysis, disjoint indicator sets for demand $D$ and capacity $K$, the outcomes, the horizon, the models, the baselines that $P$ must beat out of sample, the multiple-testing correction, a power calculation, the robustness set and the decision rule for abandoning the universal framing. We then built the indices from three real sources with full provenance: the Seshat Polaris-2026 release (626 polities; 56,274 facts, of which only 2,503 are dated; a CrisisDB table of crisis consequences for 150 polities) \citep{turchin2015,hoyer2023}, the D-PLACE Ethnographic Atlas v3.3 \citep{kirby2016,murdock1967}, and a V-Dem v16 country-year panel \citep{coppedge2020} merged with Maddison 2023 population \citep{bolt2024}. Measurement is done twice, with a transparent block index and a two-parameter latent-variable model, and uncertainty in ranges and dates is carried through.
11	11
12		−Third, we report what the pre-registered tests found (Sections~\ref{sec:hist} and~\ref{sec:modern}). Capacity and scale are positively associated in every dataset, as the framework and much prior work predict, although the direction is not identified. The mismatch index $P$ is positively associated with how badly a recorded crisis unfolds in Seshat (about $0.20$ log-odds per standard deviation, stable when any macro-region or era is held out) and with the probability of a civil war, coup or irregular regime end in the following five years in modern states (about $0.40$ per standard deviation). But the pre-registration asked a harder question than whether $\beta_P$ is positive: it asked whether $P$ beats the obvious baselines out of sample. It does not. In Seshat, a scale-only model does as well, a placebo index that replaces capacity with production infrastructure does as well, and $P$ loses to scale alone when regions are held out. In the modern panel, a capacity-only model does as well or better in every forward-chaining and leave-one-region-out comparison, past instability is associated with $P$ as strongly as future instability, and an Oster-type adjustment for unobservables reverses the sign. We find no threshold in Seshat and a nonlinearity of negligible size in the modern panel. Polity duration and discontinuous succession are unrelated to $P$. Under the decision rule we wrote down in advance, the universal framing is not supported: scale and capacity each carry signal, their difference carries none that we can identify.
	12	+Third, we report what the pre-registered tests found (Sections~\ref{sec:hist} and~\ref{sec:modern}). Capacity and scale are positively associated in every dataset, as the framework and much prior work predict, although the direction is not identified. The mismatch index $P$ is positively associated with how badly a recorded crisis unfolds in Seshat (about $0.20$ log-odds per standard deviation, stable when any macro-region or era is held out) and with the probability of a civil war, coup or irregular regime end in the following five years in modern states (about $0.39$ per standard deviation). But the pre-registration asked a harder question than whether $\beta_P$ is positive: it asked whether $P$ beats the obvious baselines out of sample. It does not. In Seshat, a scale-only model does as well, a placebo index that replaces capacity with production infrastructure is not distinguishable from $P$, and $P$ loses to scale alone when regions are held out. In the modern panel, a capacity-only model does as well or better in most forward-chaining and leave-one-region-out comparisons, the equal-and-opposite restriction that $P$ imposes on scale and capacity is rejected, past instability is associated with $P$ nearly as strongly as future instability, and the coefficient is not robust to an Oster-type adjustment for unobservables. We find no threshold in Seshat and a small nonlinearity in the modern panel. Polity duration and discontinuous succession are unrelated to $P$, and higher capacity is associated with shorter, not longer, polity durations. Under the decision rule we wrote down in advance, the universal framing is not supported: scale and capacity each carry signal, their difference carries none that we can identify.
13	13
14		−Fourth, we report the constructive part. Factorial agent-based experiments with five \emph{imposed} institutional regimes (Section~\ref{sec:sim}; 6,048 main runs, 584 ablation runs, all labelled \SIM) measure messages, bits, success rates, latency, overhead, welfare and inequality directly. Shared codes and records lower the operating cost per successful task in essentially every cell and raise success by eliminating decoding errors between heterogeneous agents; the stale-code penalty after a distribution shock equals the divergence to within the rounding of Theorem~\ref{thm:stale}; hierarchies fail abruptly when relayed load exceeds administrative capacity, and federated routing keeps most of the benefit at half the capacity. The same experiments locate where institutions hurt: a learning cost of 200 bits per agent is repaid only with a large and concentrated task space and a long horizon, saturated centralized hierarchies do worse than no institution at all, and a code can be longer than raw description when the task distribution is nearly uniform. These are the cases the framework predicts, and they are the cases historians describe.
	14	+Fourth, we report the constructive part. Factorial agent-based experiments with five \emph{imposed} institutional regimes (Section~\ref{sec:sim}; 6,048 main runs, 584 ablation runs, all labelled \SIM) measure messages, bits, success rates, latency, overhead, welfare and inequality directly. Shared codes and records lower the operating cost per successful task in essentially every cell (their effect on success is a parameter of the model, not a finding); the stale-code penalty after a distribution shock equals the divergence to within the rounding of Theorem~\ref{thm:stale}; at equal administrative budgets, centralized hierarchies fail abruptly when relayed load exceeds the budget, and federated routing keeps most of the benefit at a budget on which the centralized design fails. The same experiments locate where institutions hurt: a learning cost of 200 bits per agent is repaid only with a large and concentrated task space and a long horizon, saturated centralized hierarchies do worse than no institution at all, and a code can be longer than raw description when the task distribution is nearly uniform. These are the cases the framework predicts, and they are the cases historians describe.
15	15
16	16	The contribution, then, is threefold. We give the compression reading of institutions a formal statement that distinguishes what is proved from what is conjectured. We provide the first pre-registered, out-of-sample test of the demand--capacity mismatch conjecture on real historical and modern data, with a negative result that we think is more useful than another positive one would have been, because it shows that the mismatch, as it can be measured today, is a residual of two collinear quantities. And we provide a simulation testbed in which the mechanisms that the historical record cannot reveal---message counts, code lengths, queue saturation---can be measured. Section~\ref{sec:synth} states the scope conditions that survive, Section~\ref{sec:limits} the limitations, and Section~\ref{sec:conclusion} what a credible next test would require.

modified paper/sections/limitations.tex +3 −3

		@@ -3,12 +3,12 @@
3	3
4	4	The limitations were written down during the analysis and before this section; the full list is in the replication package. The ones that bear on how the results should be read are these.
5	5
6		−\paragraph{Data.} The CrisisDB outcome is crisis-selected (150 of 626 polities), so the Seshat test concerns the severity of recorded crises, not their occurrence. Only 6\% of Polaris facts are dated, so pre-crisis capacity is in practice whole-polity capacity and temporal precedence is weak; a dated-only test was not estimable. Seshat capacity items record the existence of institutions, not their reach, quality or cost, and the demand side is scale only because the databank has no usable indicator of division of labour or dispersion. Institutional codes and crisis codes come from the same project and overlapping historiography, which permits circularity that the inferred-code and late-crisis sensitivities can only partly address. Seshat polity boundaries are periodization decisions, so polity duration is not an instability outcome. In the modern panel, the six capacity items correlate 0.75 with electoral democracy and are rated by experts who know what came later; population begins in 1820 for most countries and there is no dispersion measure; 39 V-Dem entities without Maddison population drop out. The Ethnographic Atlas is a cross-section with coarse ordinal community sizes and no outcome.
	6	+\paragraph{Data.} The CrisisDB outcome is crisis-selected (150 of 626 polities), so the Seshat test concerns the severity of recorded crises, not their occurrence. Only 6\% of Polaris facts are dated, so pre-crisis capacity is in practice whole-polity capacity and temporal precedence is weak; a dated-only test was not estimable. Seshat capacity items record the existence of institutions, not their reach, quality or cost, and the demand side is scale only because the databank has no usable indicator of division of labour or dispersion. Institutional codes and crisis codes come from the same project and overlapping historiography, which permits circularity that the inferred-code and late-crisis sensitivities can only partly address. Seshat polity boundaries are periodization decisions, so polity duration is not an instability outcome. In the modern panel, the six capacity items correlate 0.75 with electoral democracy and are rated by experts who know what came later; population begins in 1820 for most countries and there is no dispersion measure; 39 V-Dem entities without Maddison population drop out. The Ethnographic Atlas is a cross-section with coarse ordinal community sizes (the top category is bounded at 50,000) and no outcome; jurisdictional levels are themselves a response to scale in the scalar-stress tradition \citep{johnson1982,carneiro1967}, so the H1 association there replicates a known size--hierarchy regularity rather than testing compression. The severity share aggregates fourteen weakly correlated consequences (Section~\ref{sec:hist-h2}).
7	7
8		−\paragraph{Design.} Nothing is identified causally. The negative control shows that the modern mismatch is associated with past instability as strongly as with future instability; the Oster adjustment crosses zero and the robustness value is 0.07. The Seshat binary outcome was underpowered in advance. There are 42 regions and ten macro-regions, so cluster-robust inference is anti-conservative and alternative clusterings are reported. The verdict on H2 rests on a pre-registered out-of-sample rule against baselines, which is the right test of ``beyond conventional baselines'' but is also a demanding one at $n=127$; we report the fold-level results so that readers who prefer a different rule can apply it. The Equinox-2020 replication listed in the pre-registration was not executed and is recorded as a blocked task in the audit.
	8	+\paragraph{Design.} Nothing is identified causally. The modern negative control passes the pre-registered letter (past 0.44 against future 0.48 log-odds per SD) by a margin within one standard error; the Oster adjustment under proportional selection crosses zero, with the caveat that proportional selection is a strong assumption when one control (electoral democracy) correlates 0.75 with a component of the regressor, and the robustness value is 0.07 against a partial $R^2$ of 0.006. The Seshat binary outcome was underpowered in advance. There are 42 regions and ten macro-regions, so cluster-robust inference is anti-conservative and alternative clusterings are reported. The verdict on H2 rests on a pre-registered out-of-sample rule against baselines, which is the right test of ``beyond conventional baselines'' but is also a demanding one at $n=93$--127, and one of the baselines (scale and capacity entered separately) nests the mismatch model, so that $P$ cannot do better than it in expectation; we report the fold-level results and the Wald test of the restriction so that readers who prefer a different rule can apply it. Several specification errors found by the adversarial reviews (a placebo that reparametrized the scale-only baseline, a consequence item used as a covariate, silently failed cross-validation folds, mislabelled rows) were corrected and are listed as amendments A-004 to A-007. The Equinox-2020 replication listed in the pre-registration was not executed (not blocked: the raw file is in the package, the parser was not written) and is recorded as such in the audit.
9	9
10	10	\paragraph{Theory.} The theorems describe an idealized communication system in which capacity is exogenous; historically it is built in response to demand. The conjectures that link the framework to the proxies are exactly what the measurement results call into question: with scale and capacity correlated 0.84, the proxy $P$ retains 17\% of their joint variance, and a residual of that kind cannot be assumed to track utilization.
11	11
12		−\paragraph{Simulation.} Regimes are imposed; the model says nothing about emergence. Bit and message accounting is stylized, and only within-simulation comparisons are meaningful. Decoding success does not depend on the plan drawn, so distribution shocks affect cost but not success by construction. The saturation of hierarchies follows from a free parameter swept across its range; it is confirmed, not discovered. Shannon rather than Huffman codes are used so that the stale-code penalty is comparable to the theorem.
	12	+\paragraph{Simulation.} Regimes are imposed; the model says nothing about emergence. The success ranking of regimes is the ranking of their decoding-error parameters (0.25 per unit of dialect distance without a code, 0.02 with a code, zero with records) and reverses when heterogeneity is low; only the cost side is measured rather than assumed. Bit and message accounting is stylized (one-bit acknowledgements, four-bit records, codewords repeated on every relay hop), and only within-simulation comparisons are meaningful. Decoding success does not depend on the plan drawn, so distribution shocks affect cost but not success by construction; link removal does not touch the administrative tree. The saturation of hierarchies follows from the budget multiple, a free parameter swept across its range; it is confirmed, not discovered, and its position depends on the ``capacity follows load'' allocation rule. Shannon rather than Huffman codes are used so that the stale-code penalty is comparable to the theorem; an optimal code would never be longer than fixed-length coding. Several of these choices were corrected after the computational review (amendment A-005).
13	13
14	14	\paragraph{Process.} The study was executed autonomously by a language-model agent under a human mandate in a single session: the pre-registration, code, analyses and text were produced by the agent, with the mandate's prohibitions on fabrication, imputation of missing data as zero and unapproved spending enforced by the execution log and the replication package. The principal investigator has not yet reviewed the code or the text at the time of writing; three adversarial reviews commissioned from independent reviewer personas and the responses to them are included in the package. Reference verification relied on Crossref metadata and primary-source abstracts; the chapter contents of \citet{johnson1982} and the published tables of \citet{sng2014} were not read.

modified paper/sections/methods.tex +1 −1

		@@ -16,7 +16,7 @@ For H2 the primary outcome is crisis severity $Y_1$, the share of the fourteen i
16	16	\mathbb{E}[Y_{1,c}] = \Lambda\!\left(\alpha + \beta\, P_{i(c)} + \gamma' X_{i(c)} + \mu_{r(c)} + \tau_{e(c)}\right),
17	17	\label{eq:h2a}
18	18	\end{equation}
19		−where $P_{i(c)}$ is the pre-crisis mismatch of the polity of crisis $c$, $X$ contains the warfare-technology and production-infrastructure indices and an indicator that an epidemic accompanied the crisis, and $\mu$ and $\tau$ are macro-region and era effects; standard errors are clustered by Seshat region (alternatively by natural geographic area and by polity). The binary secondary outcome $Y_{1b}$---any of collapse, fragmentation, century-long conflict, conquest or civil war coded present, defined only when all five are coded---is fitted by logit; several region--era cells are all-severe, so era effects are dropped for that outcome and heteroskedasticity-robust errors are used. Baselines are fitted on the same observations: fixed effects only, scale only, capacity only, a Turchin-style complexity component, a persistence indicator (an earlier CrisisDB crisis in the same natural geographic area), and scale and capacity entered separately. A placebo index replaces $K$ with the production-infrastructure share that was excluded from $K$ by design. Out-of-sample comparison uses leave-one-macro-region-out folds (region effects removed) and ten repetitions of five-fold cross-validation; the pre-registered rule is that $P$ ``beats'' a baseline if its mean out-of-sample log-likelihood is higher in at least 80\% of folds. For H4, a hinge model whose knot is chosen by profile likelihood on the training fold only and a natural cubic spline with three degrees of freedom are compared with the linear form on the same folds. For H5, $\beta$ is re-estimated with each macro-region and each era held out, and, where the held-out block has at least twelve crisis records, within the block.
	19	+where $P_{i(c)}$ is the pre-crisis mismatch of the polity of crisis $c$, $X$ contains the warfare-technology and production-infrastructure indices (an epidemic indicator was pre-registered as a covariate but dropped after review because it is itself a CrisisDB consequence item, amendment A-006), and $\mu$ and $\tau$ are macro-region and era effects; standard errors are clustered by Seshat region (alternatively by natural geographic area and by polity). The binary secondary outcome $Y_{1b}$---any of collapse, fragmentation, century-long conflict, conquest or civil war coded present, defined only when all five are coded---is fitted by logit; several region--era cells are all-severe, so era effects are dropped for that outcome and heteroskedasticity-robust errors are used. Baselines are fitted on a common sample on which every variable is available: fixed effects only, scale only, capacity only, a Turchin-style complexity component, a persistence indicator (an earlier CrisisDB crisis in the same natural geographic area), and scale and capacity entered separately, with a Wald test of the restriction $\beta_D+\beta_K=0$ that the mismatch index imposes. A placebo index replaces $K$ with the production-infrastructure share that was excluded from $K$ by design, fitted without that share as a control (with it, the placebo would be an exact reparametrization of the scale-only baseline). Out-of-sample comparison uses leave-one-macro-region-out folds (region effects removed) and ten repetitions of five-fold cross-validation with folds blocked by polity, test rows whose fixed-effect level is absent from the training fold being assigned the training mode; the pre-registered rule is that $P$ ``beats'' a baseline if its mean out-of-sample log-likelihood is higher in at least 80\% of folds. For H4, a hinge model whose knot is chosen by profile likelihood on the training fold only and a natural cubic spline with three degrees of freedom are compared with the linear form on the same folds. For H5, $\beta$ is re-estimated with each macro-region and each era held out, and, where the held-out block has at least twelve crisis records, within the block.
20	20
21	21	Two secondary historical outcomes use all polities. Polity duration (end year minus start year) is modelled by a Cox proportional-hazards regression stratified by macro-region, with the polity's end as the event and right-censoring for polities that reach 1900, robust errors clustered by region and a Schoenfeld test \citep{cox1972,boxsteffensmeier2004}; discontinuous succession---the successor polity's relationship to its predecessor coded as anything other than continuity---is modelled by logit. Both are reported with the caveat that Seshat polity boundaries are periodization decisions.
22	22

modified paper/sections/results_historical.tex +22 −7

		@@ -39,11 +39,11 @@ Specification & Term & Coef. & SE & $n$ & $R^2$ \\
39	39	\subsection{H2: crisis severity}
40	40	\label{sec:hist-h2}
41	41
42		−Table~\ref{tab:h2-seshat} and Figure~\ref{fig:coefplot} report the pre-registered primary test. In the primary specification, a one-standard-deviation higher pre-crisis mismatch is associated with $0.20$ higher log-odds that a given adverse consequence is recorded for the crisis (SE 0.06, $n=127$ crisis records in 117 polities and 31 Seshat regions with a defined pre-crisis index); with the rival covariates the estimate is $0.20$ (SE 0.09, $n=87$), and it is unchanged when the overall complexity component is added. The coefficient is positive and precise, survives the Holm correction over the three primary tests, and is stable under every pre-registered perturbation of the measurement model (Table~\ref{tab:h2-seshat-robust}): latent capacity ($0.15$, SE 0.09), population only ($0.27$, SE 0.11), lower and upper range bounds, inferred codes dropped, within-era and within-region standardization, ranks, quasi-polities dropped ($0.15$, SE 0.11), crises starting at least 25 years after the polity's start ($0.25$, SE 0.09), records with at least twelve coded items, no era effects, no region effects, no effects at all, and clustering by natural geographic area or by polity. Held-out macro-regions and eras (Table~\ref{tab:h5-seshat}, Figure~\ref{fig:forest}) give pooled coefficients between $0.18$ and $0.33$, every one with a confidence interval excluding zero except the block that drops the first millennium BCE ($0.20$, SE 0.17), and the within-block estimates are positive wherever they can be computed, if imprecise. By the standards of a conventional regression table, the mismatch hypothesis would be reported as confirmed.
	42	+Table~\ref{tab:h2-seshat} and Figure~\ref{fig:coefplot} report the pre-registered primary test. In the primary specification, a one-standard-deviation higher mismatch is associated with $0.20$ higher log-odds that a given adverse consequence is recorded for the crisis (SE 0.06, $n=127$ crisis records in 117 polities and 31 Seshat regions with a defined index); on the common sample used for every specification with covariates and every baseline ($n=93$), and with the warfare and production-infrastructure covariates, the estimate is $0.24$ (SE 0.09), unchanged when the overall complexity component is added. The coefficient is positive, survives the Holm correction over the three primary tests (adjusted $p<0.001$), and is stable under every pre-registered perturbation of the measurement model (Table~\ref{tab:h2-seshat-robust}): latent capacity ($0.18$, SE 0.08), population only ($0.28$, SE 0.11), capacity as a principal component, lower and upper range bounds ($0.24$ either way), inferred codes dropped ($0.13$, SE 0.06), within-era and within-region standardization, ranks ($0.19$, SE 0.08), quasi-polities dropped ($0.21$, SE 0.10), crises starting at least 25 years after the polity's start ($0.31$, SE 0.10), records with at least twelve coded items, no era effects, no region effects, no effects at all, and clustering by natural geographic area or by polity. Removing settlement hierarchy from $D$, administrative levels from $K$, or both---the two indicators that measure hierarchy depth on either side of the index---leaves the estimate at $0.25$. Held-out macro-regions, eras and---in a scheme added after review, grouping Seshat regions by cultural sphere rather than geography---held-out spheres (Table~\ref{tab:h5-seshat}, Figure~\ref{fig:forest}) give pooled coefficients between $0.24$ and $0.37$, every one with a confidence interval excluding zero except the block that drops the first millennium BCE ($0.26$, SE 0.15), and the within-block estimates are positive wherever they can be computed, if imprecise. By the standards of a conventional regression table, the mismatch hypothesis would be reported as confirmed.
43	43
44	44	\begin{table}[H]
45	45	\centering\footnotesize
46		−\caption{H2 in Seshat: crisis severity $Y_1$ (fractional logit; share of fourteen instability consequences coded present) with macro-region and era effects; standard errors clustered by Seshat region. The binary outcome $Y_{1b}$ uses region effects only (era cells are all-severe) and robust errors.}
	46	+\caption{H2 in Seshat: crisis severity $Y_1$ (fractional logit; share of fourteen instability consequences coded present) with macro-region and era effects; standard errors clustered by Seshat region. A1 uses every record with a defined index; all other rows use the common sample ($n=93$) on which every covariate and baseline is available. The binary outcome $Y_{1b}$ uses region effects only (era cells are all-severe) and robust errors.}
47	47	\label{tab:h2-seshat}
48	48	\begin{tabular}{L{8.0cm} l r r r}
49	49	\toprule
		@@ -54,11 +54,13 @@ Specification & Term & Coef. & SE & $n$ \\
54	54	\end{tabular}
55	55	\end{table}
56	56
57		−The pre-registration asked a different question, and the rest of Table~\ref{tab:h2-seshat} answers it. Scale alone is associated with severity at $0.26$ per standard deviation (SE 0.14); capacity alone at $-0.03$ (SE 0.20); the Turchin-style complexity component at $-0.01$ (SE 0.24); the persistence indicator at $0.35$ (SE 0.24). When scale and capacity are entered separately, scale carries $0.34$ (SE 0.16) and capacity $-0.21$ (SE 0.19): capacity has the sign the hypothesis predicts, but it is not distinguishable from zero. The placebo index, which replaces capacity by the share of production-infrastructure indicators that were excluded from $K$ precisely because they are not coordination capacity, is associated with severity at $0.26$ (SE 0.14)---the same as scale alone, and within sampling error of $P$. Out of sample (Table~\ref{tab:h2-seshat-cv}), the model with $P$ has a higher log-likelihood than the fixed-effects-only, capacity-only and complexity-only models in 79--85\% of the fifty cross-validation folds, but beats the scale-only model in 75\%, the persistence model in 69\% and the scale-plus-capacity model in 75\%, below the 80\% rule; with macro-regions held out it loses to scale alone in seven of ten folds and to scale plus capacity in eight. Because the principal-component loadings of $D$ and the standardization of $P$ were estimated on the full polity sample (without any outcome) before cross-validation, we repeated the leave-one-region-out comparison with loadings and standardization re-estimated inside each training fold: $P$ then beats scale alone in two of ten folds (\texttt{h2\_seshat\_cv\_noleakage.csv}), so the verdict does not depend on that pre-processing. For the binary outcome, which was declared underpowered in advance, the coefficient is negative and imprecise ($-0.18$, SE 0.20 with region effects; $-0.07$, SE 0.18 without), and no model discriminates severe from non-severe crises out of sample (areas under the curve 0.62--0.64 for every model, including the baselines). Reform consequences are unrelated to $P$. The conclusion we are obliged to draw is that what the mismatch index predicts about crisis severity is what polity scale predicts, and that nothing in the data distinguishes a capacity shortfall from a large population.
	57	+The pre-registration asked a different question, and the rest of Table~\ref{tab:h2-seshat} answers it. On the same 93 records, scale alone is associated with severity at $0.42$ per standard deviation (SE 0.17); capacity alone at $0.04$ (SE 0.25); the Turchin-style complexity component at $0.08$ (SE 0.31); the persistence indicator at $0.37$ (SE 0.20). When scale and capacity are entered separately, scale carries $0.44$ (SE 0.17) and capacity $-0.13$ (SE 0.19); the restriction that the mismatch index imposes, $\beta_D + \beta_K = 0$, is not rejected (Wald $p=0.22$), but neither is $\beta_K = 0$: capacity has the sign the hypothesis predicts and is not distinguishable from zero. The placebo index, which replaces capacity by the share of production-infrastructure indicators that were excluded from $K$ precisely because they are not coordination capacity, is associated with severity at $0.19$ (SE 0.17) against $0.29$ (SE 0.08) for $P$ with the same controls---weaker, and not distinguishable from zero, but not distinguishable from $P$ either. Out of sample (Table~\ref{tab:h2-seshat-cv}; fifty polity-blocked folds), the model with $P$ has a higher log-likelihood than the fixed-effects-only, capacity-only and complexity-only models in 84--86\% of folds, but beats the scale-only model in 42\%, the persistence model in 78\% and the scale-plus-capacity model in 44\%, below the 80\% rule; with macro-regions held out it loses to scale alone and to scale plus capacity in eight of ten folds. For the binary outcome, which was declared underpowered in advance, the coefficient is negative and imprecise ($-0.18$, SE 0.20 with region effects; $-0.07$, SE 0.18 without), and no model discriminates severe from non-severe crises out of sample much better than the fixed-effects baseline (areas under the curve 0.69--0.72 for every model). Reform consequences are unrelated to $P$. The conclusion we are obliged to draw is that what the mismatch index predicts about crisis severity is what polity scale predicts, and that nothing in the data distinguishes a capacity shortfall from a large population.
	58	+
	59	+The severity share also hides heterogeneity that \citet{hoyer2024} emphasize. The fourteen consequences are almost uncorrelated with each other (mean inter-item correlation 0.08; the first eigenvalue of their correlation matrix carries 19\% of the variance), so the share is a count of weakly related events rather than a measurement of one latent severity. Item by item (Table~\ref{tab:items}), $P$ and $D$ are associated with \emph{decline}, \emph{assassination}, \emph{deposition}, \emph{successful revolution} and \emph{civil war}---the elite-political consequences that larger polities record more often---and negatively with \emph{century-long conflict}; \emph{collapse} is present in only two of 126 coded records and its estimate is meaningless. Capacity $K$ is, if anything, positively associated with the same elite-political items. Larger and more documented polities have more recorded elite turmoil; whether that is severity or visibility, the data cannot say.
58	60
59	61	\begin{table}[H]
60	62	\centering\footnotesize
61		−\caption{Pre-registered robustness set for the Seshat primary coefficient (fractional logit of $Y_1$ on the alternative index, rival covariates, macro-region and era effects; region-clustered standard errors unless stated).}
	63	+\caption{Pre-registered robustness set for the Seshat primary coefficient (fractional logit of $Y_1$ on the alternative index, warfare and production-infrastructure covariates, macro-region and era effects; region-clustered standard errors unless stated).}
62	64	\label{tab:h2-seshat-robust}
63	65	\begin{tabular}{L{8.0cm} l r r r}
64	66	\toprule
		@@ -71,7 +73,7 @@ Variant & Term & Coef. & SE & $n$ \\
71	73
72	74	\begin{table}[H]
73	75	\centering\footnotesize
74		−\caption{Out-of-sample comparison in Seshat. Top: share of folds in which the model with $P$ (rival covariates included) has a higher out-of-sample log-likelihood than the baseline, and the mean difference per observation; the pre-registered rule requires 0.80. Bottom: out-of-sample metrics for the binary outcome $Y_{1b}$ (ten repetitions of five-fold cross-validation).}
	76	+\caption{Out-of-sample comparison in Seshat. Top: share of folds in which the model with $P$ (rival covariates included) has a higher out-of-sample log-likelihood than the baseline, and the mean difference per observation; the pre-registered rule requires 0.80. Bottom: out-of-sample metrics for the binary outcome $Y_{1b}$ (ten repetitions of five-fold cross-validation, folds blocked by polity, unseen fixed-effect levels mapped to the training mode).}
75	77	\label{tab:h2-seshat-cv}
76	78	\begin{tabular}{l l r r r}
77	79	\toprule
		@@ -109,15 +111,28 @@ Held-out block & Pooled $\beta_P$ & SE & $n$ & Held-out $n$ & Within-block (SE)
109	111	\end{tabular}
110	112	\end{table}
111	113
	114	+\begin{table}[H]
	115	+\centering\footnotesize
	116	+\caption{Item-level associations in CrisisDB: logit of each consequence (present vs absent) on $P$, on $D$ and on $K$ separately, each with era effects and robust errors; coefficients per standard deviation (SE).}
	117	+\label{tab:items}
	118	+\begin{tabular}{l r r r r r}
	119	+\toprule
	120	+Consequence & $n$ & Share present & $\beta_P$ (SE) & $\beta_D$ (SE) & $\beta_K$ (SE) \\
	121	+\midrule
	122	+\input tables/tab_items
	123	+\bottomrule
	124	+\end{tabular}
	125	+\end{table}
	126	+
112	127	\subsection{H4: no threshold}
113	128	\label{sec:hist-h4}
114	129
115		−A hinge in $P$ with its knot chosen on the training fold, and a three-degree-of-freedom spline, improve on the linear model out of sample in only 10\% and 8\% of the fifty folds (Table~\ref{tab:h4}); the in-sample hinge places its knot at $0.92$ standard deviations with a negative post-knot slope, the opposite of a tipping point. The smooth form is retained for Seshat.
	130	+A hinge in $P$ with its knot chosen on the training fold, and a three-degree-of-freedom spline, improve on the linear model out of sample in only 10\% and 8\% of the fifty polity-blocked folds (Table~\ref{tab:h4}); the in-sample hinge places its knot at $0.92$ standard deviations with a negative post-knot slope, the opposite of a tipping point. The smooth form is retained for Seshat.
116	131
117	132	\subsection{Polity duration and succession}
118	133	\label{sec:hist-y2}
119	134
120		−The mismatch index is unrelated to the hazard of a polity's end (Table~\ref{tab:y2y3}): hazard ratios of 1.03 (SE 0.06) and 1.00 (SE 0.07, Schoenfeld test $p=0.98$) with 302 and 201 events. Capacity, entered with scale, is associated with a \emph{higher} hazard (ratio 1.75, SE 0.20 on the log scale), which we read as periodization rather than fragility: well-documented complex polities are divided into more, shorter Seshat polities than poorly documented simple ones. Discontinuous succession, coded for 207 polities with 37 discontinuities, is unrelated to $P$ ($0.02$ and $0.24$ log-odds, SE 0.39 and 0.33).
	135	+The mismatch index is unrelated to the hazard of a polity's end (Table~\ref{tab:y2y3}): hazard ratios of 1.03 (SE 0.06) and 1.00 (SE 0.07, Schoenfeld test $p=0.98$) with 302 and 201 events. Capacity, entered with scale, is associated with a \emph{higher} hazard of a polity's end (ratio 1.75, SE 0.20 on the log scale). Our first reading was periodization: well-documented complex polities are divided into more, shorter Seshat polities than poorly documented simple ones, and the number of polities coded in a natural geographic area does correlate 0.29 with capacity and does raise the hazard (ratio 1.37, SE 0.14). But controlling for it leaves the capacity coefficient almost unchanged (ratio 1.71, SE 0.20), so periodization is at most part of the story. The pre-registered secondary outcome therefore yields a result that runs against the hypothesis: in Seshat, higher-capacity polities end sooner, not later, and the paper reports it as such rather than explaining it away. Candidate explanations---more frequent dynastic turnover being coded as polity change in complex states, or genuine fragility of elaborate administrations \citep{tainter1988}---are not distinguishable here. Discontinuous succession, coded for 207 polities with 37 discontinuities, is unrelated to $P$ ($0.02$ and $0.24$ log-odds, SE 0.39 and 0.33).
121	136
122	137	\begin{table}[H]
123	138	\centering\footnotesize

modified paper/sections/results_modern.tex +5 −5

		@@ -6,7 +6,7 @@ The modern panel is the form of the test that the mandate's event-risk model has
6	6	\subsection{H2: instability events in the next five years}
7	7	\label{sec:modern-h2}
8	8
9		−In the primary specification (Table~\ref{tab:h2-modern}), a one-standard-deviation higher mismatch is associated with $0.40$ higher log-odds of at least one civil-war onset, coup or irregular regime end in the following five years (SE 0.09; 14,637 country-years, 2,692 event windows), conditional on development, electoral democracy, time since the last event, region and year. The coefficient is $0.72$ without covariates, $1.31$ in a conditional logit with country and decade effects (which uses only countries that experience at least one event window), and $0.11$ in probability units in a linear model with country and year effects. It is positive for every horizon (0.33 at one year, 0.39 at three, 0.45 at ten), largest for civil-war onsets (1.01, SE 0.20), small and imprecise for coups (0.20, SE 0.14) and intermediate for irregular regime ends (0.25, SE 0.09); it does not depend on the capacity aggregation, on standardizing within years, on adding land area to demand, on dropping the nineteenth century, on removing the democracy control or on adding the persistence control (Table~\ref{tab:h2-modern-robust}). Every held-out region and period gives a pooled coefficient between $0.29$ and $0.47$, and every within-block estimate is positive (Table~\ref{tab:h5-modern}, Figure~\ref{fig:forest}). The second H2 primary test also passes the conventional bar.
	9	+In the primary specification (Table~\ref{tab:h2-modern}), a one-standard-deviation higher mismatch is associated with $0.39$ higher log-odds of at least one civil-war onset, coup or irregular regime end in the following five years (SE 0.09; 14,637 country-years, of which 2,692 have an event in their five-year window; the windows overlap, and the sample contains 901 distinct event-years out of 1,676 in the full panel), conditional on development, electoral democracy, time since the last event (with an indicator for countries whose spell is left-censored at panel entry), region and year. The coefficient is $0.72$ without covariates, $1.33$ in a conditional logit with country and decade effects (which uses only countries that experience at least one event window), and $0.11$ in probability units in a linear model with country and year effects. It is positive for every horizon (0.32 at one year, 0.38 at three, 0.45 at ten), largest for civil-war onsets (1.02, SE 0.19), small and imprecise for coups (0.19, SE 0.14) and intermediate for irregular regime ends (0.24, SE 0.09); it does not depend on the capacity aggregation, on standardizing within years, on adding land area to demand, on dropping the nineteenth century, on removing the democracy control or on adding the persistence control (Table~\ref{tab:h2-modern-robust}). Every held-out region and period gives a pooled coefficient of similar size, and every within-block estimate is positive (Table~\ref{tab:h5-modern}, Figure~\ref{fig:forest}). The second H2 primary test also passes the conventional bar.
10	10
11	11	\begin{table}[H]
12	12	\centering\footnotesize
		@@ -21,7 +21,7 @@ Specification & Term & Coef. & SE & $n$ & Events \\
21	21	\end{tabular}
22	22	\end{table}
23	23
24		−The baselines, placebo and negative control tell the rest (Table~\ref{tab:h2-modern}, Figure~\ref{fig:coefplot}). Population alone is weakly associated with instability ($0.13$, SE 0.07); capacity alone is strongly and negatively associated ($-0.63$, SE 0.13), as is development ($-0.56$ per log unit of GDP per capita). With population and capacity entered separately, population carries $0.17$ (SE 0.07) and capacity $-0.71$ (SE 0.14); conditional on capacity, the mismatch adds $0.23$ (SE 0.09). The signal in $P$ is, here, overwhelmingly the capacity component with the sign reversed. The placebo that replaces capacity with electoral democracy is associated at $0.20$ (SE 0.06) against $0.45$ (SE 0.08) for $P$ with the same controls, so the capacity items do carry information beyond regime type, which is reassuring for the measurement model but not for the hypothesis. The negative control is not reassuring: past events, in the five years before $t$, are associated with $P_t$ at $0.44$ (SE 0.10), against $0.48$ (SE 0.10) for future events with the same controls. A predictor that is as strongly related to what has already happened as to what will happen is a marker of chronically unstable countries, not a leading indicator, and the pre-registration said in advance that a negative control of this kind would be read that way.
	24	+The baselines, placebo and negative control tell the rest (Table~\ref{tab:h2-modern}, Figure~\ref{fig:coefplot}). Population alone is weakly associated with instability ($0.12$, SE 0.07); capacity alone is strongly and negatively associated ($-0.63$, SE 0.13), as is development ($-0.58$ per log unit of GDP per capita). With population and capacity entered separately, population carries $0.17$ (SE 0.07) and capacity $-0.70$ (SE 0.14), and the restriction $\beta_D + \beta_K = 0$ that the mismatch index imposes is rejected (sum $-0.54$, SE 0.15, Wald $p<0.001$): in the modern panel $P$ is a misspecified summary of two components of unequal weight, and the row that adds $K$ to $P$ (which is algebraically the same model) shows the mismatch adding $0.23$ (SE 0.09) once capacity is held fixed---that is, the population component. The signal in $P$ is, here, overwhelmingly the capacity component with the sign reversed. The placebo that replaces capacity with electoral democracy is associated at $0.20$ (SE 0.06) against $0.45$ (SE 0.08) for $P$ with the same controls, so the capacity items do carry information beyond regime type, which is reassuring for the measurement model but not for the hypothesis. The negative control, read by the letter of the pre-registration, passes: past events in the five years before $t$ are associated with $P_t$ at $0.44$ (SE 0.10), against $0.48$ (SE 0.10) for future events with the same controls, so the forward association is not weaker than the backward one. But the margin is within one standard error, and a predictor that is nearly as strongly related to what has already happened as to what will happen behaves like a marker of chronically unstable countries rather than a leading indicator; we report both the literal verdict and this reading.
25	25
26	26	\begin{table}[H]
27	27	\centering\footnotesize
		@@ -36,7 +36,7 @@ Variant & Term & Coef. & SE & $n$ & Events \\
36	36	\end{tabular}
37	37	\end{table}
38	38
39		−Out of sample (Table~\ref{tab:h2-modern-cv}, Figure~\ref{fig:calibration}) the verdict is the same as in Seshat with the roles of the components exchanged. Under forward chaining, the model with $P$ has a slightly higher area under the curve than the fixed-effects model (0.773 against 0.768) and than population alone, but a lower one than capacity alone (0.777) and than population and capacity together (0.779), and it never beats the capacity-only model on log-likelihood or discrimination in any of the three temporal splits; with regions held out it beats capacity alone in one of six folds on log-likelihood and three of six on discrimination. All models are well calibrated under forward chaining (slopes 0.95--0.99) and all are badly calibrated when a whole region is held out (slopes about 0.43), which is itself a finding: the level of instability risk is regional, and no index built from scale and capacity transports it. Under the pre-registered 80\% rule, $P$ fails against the capacity-only and the scale-plus-capacity baselines in both schemes, and the second H2 primary test is therefore not passed.
	39	+Out of sample (Table~\ref{tab:h2-modern-cv}, Figure~\ref{fig:calibration}) the verdict is the same as in Seshat with the roles of the components exchanged. Under forward chaining (three temporal splits, the first of which trains on 1820--1945 and tests on 1946--1975), the model with $P$ has the highest mean area under the curve (0.732 against 0.722 for the fixed-effects model, 0.724 for population or capacity alone and 0.726 for both), but it beats the capacity-only and the scale-plus-capacity baselines in only one of the three splits on either metric; with regions held out it beats capacity alone in one of six folds on log-likelihood and three of six on discrimination. All models are reasonably calibrated under forward chaining (slopes 0.77--0.79) and all are badly calibrated when a whole region is held out (slopes about 0.40), which is itself a finding: the level of instability risk is regional, and no index built from scale and capacity transports it. Under the pre-registered 80\% rule, $P$ fails against the capacity-only and the scale-plus-capacity baselines in both schemes (the latter nests the former, so this is the expected outcome whenever the restriction is wrong), and the second H2 primary test is therefore not passed.
40	40
41	41	\begin{table}[H]
42	42	\centering\footnotesize
		@@ -88,7 +88,7 @@ Held-out block & Pooled $\beta_P$ & SE & $n$ & Within $n$ & Within-block (SE) \\
88	88	\subsection{Sensitivity to unobservables}
89	89	\label{sec:modern-sens}
90	90
91		−Table~\ref{tab:sensitivity} applies Oster's and Cinelli--Hazlett's diagnostics to the linear-probability version of the primary model. Adding the controls moves the coefficient from $0.082$ to $0.038$ while raising $R^2$ from 0.127 to 0.161; the bias-adjusted coefficient at $R_{\max}=1.3R^2$ and $\delta=1$ is $-0.024$, on the other side of zero. The partial $R^2$ of $P$ is 0.006 and the robustness value is 0.072: an unobserved confounder that explained seven percent of the residual variance of both the mismatch and the outcome would suffice to remove the association. Under the conventions of the sensitivity literature this is a fragile coefficient.
	91	+Table~\ref{tab:sensitivity} applies Oster's and Cinelli--Hazlett's diagnostics to the linear-probability version of the primary model. Adding the controls moves the coefficient from $0.082$ to $0.037$ while raising $R^2$ from 0.127 to 0.161; the bias-adjusted coefficient at $R_{\max}=1.3R^2$ and $\delta=1$ is $-0.030$, on the other side of zero. The partial $R^2$ of $P$ is 0.005 and the robustness value is 0.07: an unobserved confounder that explained seven percent of the residual variance of both the mismatch and the outcome would suffice to remove the association. Two caveats apply. Proportional selection ($\delta=1$) is a strong assumption when one of the controls, electoral democracy, correlates 0.75 with a component of the regressor, so the Oster adjustment overstates the plausible bias; and the robustness value, while twelve times the partial $R^2$ of $P$, has no benchmark covariate against which to judge it. The coefficient is fragile under these conventions, not shown to be spurious.
92	92
93	93	\begin{table}[H]
94	94	\centering\footnotesize
		@@ -106,7 +106,7 @@ Quantity & Value \\
106	106	\subsection{H4: a nonlinearity of negligible size}
107	107	\label{sec:modern-h4}
108	108
109		−With country-blocked folds, the hinge improves on the linear form in 12 of 15 folds and the spline in 13 (Table~\ref{tab:h4}), which meets the pre-registered 80\% rule by one fold. The improvement is 0.001 nats per observation; the in-sample curve (Figure~\ref{fig:h4curve}) is a mild convexity around the median rather than a step, and the hinge knot is at $-0.08$ standard deviations, i.e.\ at the centre of the distribution. We report this as a marginal nonlinearity consistent with the smooth convexity of Theorem~\ref{thm:backlog} and not as a threshold; the pre-registration's falsifier for H4 was a hinge or step model that \emph{outperforms} the smooth one, and a spline is a smooth model.
	109	+With country-blocked folds, the hinge improves on the linear form in 11 of 15 folds and the spline in 12 (Table~\ref{tab:h4}); the spline meets the pre-registered 80\% rule exactly and the hinge does not. By the letter of the pre-registration, H4 is therefore not rejected in the modern panel. The improvement is 0.003 nats per observation; the in-sample curve (Figure~\ref{fig:h4curve}) is a mild convexity around the median rather than a step, and the hinge knot is at $-0.08$ standard deviations, i.e.\ at the centre of the distribution. We report this as a nonlinearity that survives the rule but is small, consistent with the smooth convexity of Theorem~\ref{thm:backlog} rather than with a tipping point; the pre-registration's falsifier for H4 was a hinge or step model that \emph{outperforms} the smooth one, and the form that passes is the smooth spline.
110	110
111	111	\begin{table}[H]
112	112	\centering\footnotesize

modified paper/sections/simulation.tex +20 −20

		@@ -1,27 +1,27 @@
1	1	\section{Agent-based experiments}
2	2	\label{sec:sim}
3	3
4		−Everything in this section is \SIM. The simulation exists to measure what the historical record cannot: messages, bits, queues, overhead and failures under institutional regimes that are \emph{imposed} by the experimenter. Nothing emerges in these experiments, and nothing in them is evidence about history; they test the theorems of Section~\ref{sec:theory} and search for the conditions under which institutions reduce efficiency or add fragility, as the pre-registration requires.
	4	+Everything in this section is \SIM. The simulation exists to measure what the historical record cannot: messages, bits, queues, overhead and failures under institutional regimes that are \emph{imposed} by the experimenter. Nothing emerges in these experiments, and nothing in them is evidence about history; they instantiate the theorems of Section~\ref{sec:theory} in a system where the quantities they speak of are defined, and they search for the conditions under which institutions reduce efficiency or add fragility, as the pre-registration requires. Two things should be said before any number. First, the error structure is a parameter, not a finding: agents without a shared code misdecode with a probability proportional to their dialect distance (0.25 per unit), agents with a code with probability 0.02, and agents who can verify against a record not at all. Any statement that institutions ``raise success'' in this model is therefore a statement about those parameters, and the ablation that varies them is the one to read. What the model measures that is not assumed is the \emph{cost} side: bits, messages, queueing delay, overhead, and how they scale and fail. Second, the regimes are compared at an equal total administrative budget, so that federation can gain only by relaying less.
5	5
6	6	\subsection{Design}
7	7	\label{sec:sim-design}
8	8
9		−$N$ agents are placed at random in the unit square and linked on a lattice-like nearest-neighbour graph, a small-world rewiring of it, or a random graph of the same degree. Each period, tasks arrive at rate $\lambda = 0.5$ per producing agent; a task is initiated by one agent and requires $k=3$ participants---the initiator and two network neighbours---to hold the same plan, drawn from a set of $s=64$ plans with a Zipf distribution of exponent 0 (uniform, entropy 6 bits) or 1 (entropy 4.86 bits). The initiator must communicate the plan. Each agent can process a bounded number of messages per period (2 or 8); excess messages queue first-in-first-out, and a task whose last message is served more than three periods after arrival fails. Agents differ in a continuous ``dialect'' that is partly spatial; without a shared code, a message is misdecoded with probability proportional to the dialect distance (0.25 per unit), and misdecoded tasks fail. Five regimes are compared. \emph{None}: raw messages of $\lceil\log_2 s\rceil$ bits per recipient, dialect errors, one acknowledgement round. \emph{Norms}: a shared Shannon code optimal for the plan distribution, lower decoding error (0.02), a one-time learning cost of 200 bits per agent, and, after a shift in the plan distribution, a stale code until the institution re-learns it ten periods later. \emph{Records}: norms plus a written record per task (four bits of administrative overhead) against which recipients verify, eliminating decoding errors and the acknowledgement round. \emph{Central}: records plus an eight-ary administrative tree through whose root every task is relayed; administrators have their own bandwidth and are withdrawn from production. \emph{Federated}: the same tree, but a task is relayed only up to the lowest common ancestor of its participants, so that local tasks never reach the root. Administrative bandwidth is set as a ratio (0.5 or 2) of the expected relayed load. Shocks at period 20 of 40 either shift the plan distribution (a cyclic permutation chosen to reach a target Kullback--Leibler divergence; the achieved divergence, 0.52 bits, is recorded) or remove half of the links. The main design crosses $N\in\{50,200,800\}$, three networks, two bandwidths, two entropies, three shock conditions and five regimes with eight seeds (6,048 runs); the ablations (584 runs) vary verification, adaptation speed, horizon, learning cost, heterogeneity, administrative capacity, branching and the plan space. Measured outcomes are task success, messages and bits per task, bits per successful task with and without set-up, latency, failure shares by cause, administrative overhead share, welfare (task value minus a cost per bit), the Gini coefficient of benefits among producers, and root load. Entropies, code lengths and divergences are computed because the plan distribution and the code are defined by the experimenter.
	9	+$N$ agents are placed at random in the unit square and linked on a lattice-like nearest-neighbour graph, a small-world rewiring of it, or a random graph of the same degree. Each period, tasks arrive at rate $\lambda = 0.5$ per producing agent; a task is initiated by one agent and requires $k=3$ participants---the initiator and two network neighbours---to hold the same plan, drawn from a set of $s=64$ plans with a Zipf distribution of exponent 0 (uniform, entropy 6 bits) or 1 (entropy 4.86 bits). The initiator must communicate the plan to its two partners (there is no broadcast). Each agent can process a bounded number of messages per period (2 or 8); excess messages queue first-in-first-out, and a task whose last message is served more than three periods after arrival fails. Five regimes are compared. \emph{None}: raw messages of $\lceil\log_2 s\rceil$ bits per recipient, dialect errors, and a one-bit acknowledgement round that doubles the messages each agent must process. \emph{Norms}: a shared Shannon code optimal for the plan distribution (Shannon rather than Huffman lengths, so that the stale-code penalty of Theorem~\ref{thm:stale} applies verbatim), lower decoding error, a one-time learning cost of 200 bits per agent, and, after a shift in the plan distribution, a stale code until the institution re-learns it ten periods later. \emph{Records}: norms plus a written record per task (four bits of administrative overhead) against which recipients verify, which removes decoding errors and replaces the acknowledgement round. \emph{Central}: records plus an eight-ary administrative tree; every task is relayed from the initiator up to the root and down to each partner, the codeword is transmitted on every hop, and every administrator on the path processes one message per hop. \emph{Federated}: the same tree, but a task is relayed only up to the lowest common ancestor of its participants, so that local tasks never reach the root. Administrators are withdrawn from production. The total administrative budget is a multiple (0.5 or 2 in the main design; 0.25 to 4 in the ablation) of the expected relayed load of the \emph{central} design, and it is allocated to administrators in proportion to the load each one carries in a five-period warm-up with unconstrained capacity (``capacity follows load''); central and federated designs therefore face the same budget, and the budget multiple is the inverse utilization of the central design in the sense of Theorem~\ref{thm:backlog}. Shocks at period 20 of 40 either shift the plan distribution (a cyclic permutation reaching a Kullback--Leibler divergence of 1.03 bits in the non-uniform cells; in uniform cells the shift is a no-op and those cells are excluded from the shock statistics) or remove half of the links among agents (the administrative tree is not touched). The main design crosses $N\in\{50,200,800\}$, three networks, two bandwidths, two entropies, three shock conditions and five regimes with eight seeds (6,048 runs); the ablations (584 runs) vary verification, adaptation speed, horizon, learning cost, heterogeneity, administrative budget, branching and the plan space. Measured outcomes are task success, messages and bits per task, bits per successful task with and without set-up, latency, failure shares by cause, administrative overhead share, welfare (task value minus a cost per bit), the Gini coefficient of benefits among producers, and root load. Shock effects are difference-in-differences against the unshocked run of the same cell and seed. Entropies, code lengths and divergences are computed because the plan distribution and the code are defined by the experimenter.
10	10
11		−\subsection{H3: institutions lower the cost of coordination---with exceptions}
	11	+\subsection{H3: the cost side}
12	12	\label{sec:sim-h3}
13	13
14		−Table~\ref{tab:sim-paired} pairs each institutional run with the no-institution run of the same cell and seed. Shared norms lower the operating bits per successful task in 99.7\% of pairs and raise success in 99.3\%, by eliminating most decoding failures between heterogeneous agents; records do so in 71\% and 100\% of pairs, and raise welfare in every pair. Once set-up costs are included, however, norms are never cheaper than nothing within the forty-period horizon: the coding gain of about one bit per recipient on a six-bit plan space does not repay 200 bits of learning per agent, exactly as Theorem~\ref{thm:breakeven} says it should not, and the ablation confirms that the code pays for itself only with a 256-plan space, a Zipf exponent of 1.5 and a 160-period horizon (Table~\ref{tab:sim-ablation}, Figure~\ref{fig:sim-phase}). Centralized administration with capacity twice the expected load is indistinguishable from records in success and bits, adds messages (relay), and lowers welfare in half the cells because administrators are withdrawn from production; with capacity at half the expected load it is worse than no institution in 99\% of pairs, success falls by 0.44 on average, and welfare by 28 units per agent (Table~\ref{tab:sim-loss}). Federated routing at the same insufficient capacity keeps success higher than no institution in 84\% of pairs, because most tasks never reach the root: the root-load bound of Theorem~\ref{thm:routing} is what separates the two hierarchies (Table~\ref{tab:sim-capacity}, Figure~\ref{fig:sim-phase}). Figure~\ref{fig:sim-scaling} shows the scaling with $N$: messages per task rise with the depth of the tree for the hierarchies and are constant for the flat regimes, bits per success are flat in $N$ for all regimes, and inequality of benefits is slightly higher under the hierarchies.
	14	+Table~\ref{tab:sim-paired} pairs each institutional run with the no-institution run of the same cell and seed, and Table~\ref{tab:sim-loss} counts the design cells in which an institution does worse than nothing. Shared norms lower the operating bits per successful task in 99.7\% of pairs; once the 200-bit learning cost is included they are never cheaper within the forty-period horizon, because the coding gain of a six-bit plan space is about one bit per recipient (Table~\ref{tab:sim-ablation}: the Shannon code averages 5.44 bits at Zipf exponent 1 against 6 raw bits), exactly the break-even arithmetic of Theorem~\ref{thm:breakeven}; the ablation finds the code repaying its learning cost only with a 256-plan space, a Zipf exponent of 1.5 and a 160-period horizon (Figure~\ref{fig:sim-phase}). Records lower operating cost in 71\% of pairs and raise welfare in every pair, the acknowledgement round they replace being what saturates agents' queues at low bandwidth in the flat regimes (at bandwidth 2, flat regimes complete about half of their tasks and records 96\%; at bandwidth 8 all complete more than 90\%). The hierarchies are expensive in bits because the codeword travels on every hop: at the reference budget, central administration costs 83 bits per successful task against 25 for records and 15 for nothing, federated administration 53, and both lower welfare below the flat institutional regimes in half the cells because administrators are withdrawn from production. With a budget at half the central design's load, central administration fails almost every task (success down by 0.49 relative to nothing; welfare loss in all 108 cells), and federation, which relays fewer hops, does only somewhat better (success down by 0.29; welfare loss in 80\% of cells). Figure~\ref{fig:sim-scaling} shows the scaling with $N$: messages and bits per task grow with the depth of the tree for the hierarchies and are flat for the flat regimes; inequality of benefits is slightly higher under the hierarchies.
15	15
16	16	\begin{table}[H]
17	17	\centering\footnotesize
18		−\caption{\SIM\ Paired comparison of each institutional regime with the no-institution run of the same cell and seed (main design, 864 pairs per row). Shares of pairs in which the regime has lower bits per successful task (operating; including set-up), higher success and higher welfare, and mean differences.}
	18	+\caption{\SIM\ Paired comparison of each institutional regime with the no-institution run of the same cell and seed (main design, 864 pairs per row). Shares of pairs in which the regime has lower bits per successful task (operating; including set-up), higher success and higher welfare, and mean differences. The budget multiple is relative to the central design's expected relayed load.}
19	19	\label{tab:sim-paired}
20	20	\resizebox{\textwidth}{!}{\begin{tabular}{l r r r r r r r r r}
21	21	\toprule
22		−Regime & Cap.\ ratio & Pairs & Lower bits & Lower incl.\ set-up & Higher success & Higher welfare & $\Delta$ bits/success & $\Delta$ success & $\Delta$ welfare \\
	22	+Regime & Budget & Pairs & Lower bits & Lower incl.\ set-up & Higher success & Higher welfare & $\Delta$ bits/success & $\Delta$ success & $\Delta$ welfare \\
23	23	\midrule
24		−\input tables/tab_sim_paired
	24	+\input tables/tab_sim_paired
25	25	\bottomrule
26	26	\end{tabular}}
27	27	\end{table}
		@@ -32,9 +32,9 @@ Regime & Cap.\ ratio & Pairs & Lower bits & Lower incl.\ set-up & Higher success
32	32	\label{tab:sim-loss}
33	33	\begin{tabular}{l r r r r r}
34	34	\toprule
35		−Regime & Cap.\ ratio & Cells & Welfare loss & Bits loss & Success loss \\
	35	+Regime & Budget & Cells & Welfare loss & Bits loss & Success loss \\
36	36	\midrule
37		−\input tables/tab_sim_loss
	37	+\input tables/tab_sim_loss
38	38	\bottomrule
39	39	\end{tabular}
40	40	\end{table}
		@@ -42,37 +42,37 @@ Regime & Cap.\ ratio & Cells & Welfare loss & Bits loss & Success loss \\
42	42	\begin{figure}[H]
43	43	\centering
44	44	\includegraphics[width=\textwidth]{figures/fig_sim_scaling.pdf}
45		−\caption{\SIM\ Scaling by regime without shocks (administrative capacity ratio 2): bits per successful task including set-up (log scale), success rate and messages per task against $N$; means and 95\% confidence intervals over cells and seeds.}
	45	+\caption{\SIM\ Scaling by regime without shocks (budget multiple 2): bits per successful task including set-up (log scale), success rate and messages per task against $N$; means and 95\% confidence intervals over cells and seeds.}
46	46	\label{fig:sim-scaling}
47	47	\end{figure}
48	48
49	49	\subsection{Theorems in silico}
50	50	\label{sec:sim-theorems}
51	51
52		−The coding-gain sweep reproduces Theorem~\ref{thm:coding} cell by cell: the expected code length lies in $[H(p), H(p)+1)$ for every plan space and exponent, and the code is shorter than raw description whenever $H(p) \le \log_2 s - 1$; in the one cell where the distribution is nearly uniform (exponent 0.5, entropy 5.78 bits) the Shannon code is \emph{longer} than the six raw bits (6.24), which is the exception the theorem allows and a reminder that standardization is not free when there is little regularity to exploit. After a distribution shock, a non-adapting code costs 1.90 extra bits per recipient against a divergence of 1.98 bits, within the one-bit rounding of Theorem~\ref{thm:stale}; a code that re-learns after ten periods pays half of that, and an instantly adapting code almost nothing (Table~\ref{tab:sim-ablation}). Because decoding success does not depend on which plan was drawn, distribution shocks raise cost without lowering success in this model (Table~\ref{tab:sim-shocks}); link removal, by contrast, concentrates load on surviving neighbours and lowers success by about 0.08 under the hierarchies and records and by almost nothing under flat regimes (difference-in-differences against the unshocked run of the same cell). The capacity sweep traces the saturation of Theorem~\ref{thm:backlog}: central success collapses as the capacity ratio falls below one, while federated routing degrades gracefully (Figure~\ref{fig:sim-phase}).
	52	+The coding-gain sweep reproduces Theorem~\ref{thm:coding}: the expected code length lies in $[H(p), H(p)+1)$ for every plan space and exponent, and it is below the raw length whenever $H(p) \le \log_2 s - 1$. In the one cell where the distribution is nearly uniform (exponent 0.5, entropy 5.78 bits) the Shannon code is longer than the six raw bits (6.24). As the remark after Theorem~\ref{thm:coding} notes, this is a property of the suboptimal Shannon code, not of coding: a Huffman code is never longer than the fixed-length code. We keep the Shannon code because it is the one Theorem~\ref{thm:stale} is stated for, and the sweep then reproduces that theorem too: after a shift in the plan distribution, a non-adapting code costs 1.90 extra bits per recipient against a divergence of 1.98 bits, within the one-bit rounding; a code that re-learns after ten periods pays half, an instantly adapting code almost nothing. Because decoding success does not depend on which plan was drawn, distribution shocks raise cost without lowering success in this model (Table~\ref{tab:sim-shocks}), which is a modelling choice and not a finding. Link removal lowers success by 0.07--0.08 under records and the hierarchies at the reference budget and by almost nothing under the flat regimes; the latter is partly a floor effect, since at bandwidth 2 the flat regimes are already failing half of their tasks before the shock (the replication package reports the decomposition by bandwidth). The budget sweep traces the saturation of Theorem~\ref{thm:backlog} and the routing bound of Theorem~\ref{thm:routing} together (Table~\ref{tab:sim-capacity}, Figure~\ref{fig:sim-phase}): central success is 0.03 at a budget of half the load, 0.48 at a budget equal to the load, and 0.94 at one and a half times; federated success, with the same budgets, is 0.29, 0.99 and 1.00, because its root carries a tenth of the central root's load.
53	53
54	54	\begin{table}[H]
55	55	\centering\footnotesize
56		−\caption{\SIM\ Shock responses by regime: difference-in-differences of success and of bits per task (post minus pre, shocked minus unshocked run of the same cell and seed); standard error of the mean in parentheses; achieved divergence for distribution shocks.}
	56	+\caption{\SIM\ Shock responses by regime: difference-in-differences of success and of bits per task (post minus pre, shocked minus unshocked run of the same cell and seed); standard error of the mean in parentheses; achieved divergence for distribution shocks (non-uniform cells only).}
57	57	\label{tab:sim-shocks}
58	58	\begin{tabular}{l l r r r r}
59	59	\toprule
60		−Shock & Regime & Cap.\ ratio & $\Delta$ success (SE) & $\Delta$ bits/task & $D_{KL}(q\|p)$ \\
	60	+Shock & Regime & Budget & $\Delta$ success (SE) & $\Delta$ bits/task & $D_{KL}(q\|p)$ \\
61	61	\midrule
62		−\input tables/tab_sim_shocks
	62	+\input tables/tab_sim_shocks
63	63	\bottomrule
64	64	\end{tabular}
65	65	\end{table}
66	66
67	67	\begin{table}[H]
68	68	\centering\footnotesize
69		−\caption{\SIM\ Hierarchy saturation ($N=200$, lattice): success, latency failures, mean latency and root load against the administrative capacity ratio.}
	69	+\caption{\SIM\ Hierarchy saturation ($N=200$, lattice): success, latency failures, mean latency and root load against the administrative budget multiple, at equal total budgets for the two designs.}
70	70	\label{tab:sim-capacity}
71	71	\begin{tabular}{l r r r r r}
72	72	\toprule
73		−Regime & Cap.\ ratio & Success & Latency failures & Latency & Root load \\
	73	+Regime & Budget & Success & Latency failures & Latency & Root load \\
74	74	\midrule
75		−\input tables/tab_sim_capacity
	75	+\input tables/tab_sim_capacity
76	76	\bottomrule
77	77	\end{tabular}
78	78	\end{table}
		@@ -82,7 +82,7 @@ Regime & Cap.\ ratio & Success & Latency failures & Latency & Root load \\
82	82	\begin{subfigure}[t]{0.48\textwidth}\centering\includegraphics[width=\textwidth]{figures/fig_sim_phase.pdf}\caption{Saturation of hierarchies.}\end{subfigure}\hfill
83	83	\begin{subfigure}[t]{0.48\textwidth}\centering\includegraphics[width=\textwidth]{figures/fig_sim_breakeven.pdf}\caption{Break-even of a shared code.}\end{subfigure}\\[2mm]
84	84	\includegraphics[width=0.75\textwidth]{figures/fig_sim_shocks.pdf}
85		−\caption{\SIM\ Phase diagram of success against administrative capacity (left), ratio of bits per success with and without a shared code against the horizon (right), and difference-in-differences shock responses (bottom).}
	85	+\caption{\SIM\ Success against the administrative budget multiple (left), ratio of bits per success with and without a shared code against the horizon (right), and difference-in-differences shock responses (bottom).}
86	86	\label{fig:sim-phase}
87	87	\end{figure}
88	88
		@@ -94,7 +94,7 @@ Regime & Cap.\ ratio & Success & Latency failures & Latency & Root load \\
94	94	\toprule
95	95	Ablation & Result \\
96	96	\midrule
97		−\input tables/tab_sim_ablation
	97	+\input tables/tab_sim_ablation
98	98	\bottomrule
99	99	\end{tabular}
100	100	\end{table}
		@@ -102,4 +102,4 @@ Ablation & Result \\
102	102	\subsection{What the experiments say about the hypothesis}
103	103	\label{sec:sim-verdict}
104	104
105		−H3 is supported in the operating sense and refuted in the total-cost sense within short horizons, which is the content of Theorem~\ref{thm:breakeven}; the experiments also show that the benefit of institutions in this model comes at least as much from eliminating misunderstanding between heterogeneous agents as from shortening messages, since the coding gain on a six-bit plan space is small while the decoding gain is large (with heterogeneity at 0.6, success is 0.81 without a code and 0.96 with one; with heterogeneity at 0.05 the ranking reverses, 0.98 against 0.96, because the code's own error rate is then the larger one). The fragility search found what the framework predicts: institutions hurt when their fixed cost is not amortized, when a centralized hierarchy is asked to relay more than it can process, and when there is too little regularity for a code to compress. None of these cases is exotic. All of them are described in the historical literature on over-extended administrations and rigid bureaucracies, and none of them is measurable with the indicators available in Seshat.
	105	+On the cost side, which is the side the model actually measures, H3 is supported for operating cost and refuted for total cost within short horizons: a code saves bits per task in proportion to the regularity of the task environment, and it repays its learning cost only when that regularity is large and the horizon long (Theorem~\ref{thm:breakeven}). Records are the regime that wins most often, and they win by replacing a round of acknowledgements with a lookup, not by compressing messages. Hierarchies buy routing at the price of relaying, and they fail abruptly when the budget falls below the load they relay; federation halves that load and is the only hierarchical design that survives a tight budget. On the success side, the ranking of regimes is the ranking of their error parameters: with heterogeneity at 0.6 the no-code regime succeeds in 81\% of tasks against 96\% with a code, with heterogeneity at 0.05 the ranking reverses (98\% against 96\%) because the code's own error rate is then the larger one. The fragility search therefore finds what the framework predicts and nothing else: institutions hurt when their fixed cost is not amortized, when a hierarchy is asked to relay more than it can process, and when there is too little regularity for a code to compress. These are the cases the historical literature on over-extended administrations and rigid bureaucracies describes, and none of them is measurable with the indicators available in Seshat.

modified paper/sections/synthesis.tex +5 −5

		@@ -14,10 +14,10 @@ Table~\ref{tab:observable} sets the framework's mechanisms against the two kinds
14	14	\toprule
15	15	Mechanism & Historical record (\EMP) & Simulation (\SIM) \\
16	16	\midrule
17		−Coding gain (Thm.~\ref{thm:coding}) & Not observable: presence of codes, not message lengths & Reproduced in every cell; one cell where the code is longer than raw \\
	17	+Coding gain (Thm.~\ref{thm:coding}) & Not observable: presence of codes, not message lengths & Reproduced in every cell; Shannon code longer than raw in one near-uniform cell \\
18	18	Break-even (Thm.~\ref{thm:breakeven}) & Indirect: capacity rises with scale (H1) & Code repaid only with large concentrated plan space and long horizon \\
19	19	Stale code (Thm.~\ref{thm:stale}) & Not observable & Extra bits $1.90$ vs divergence $1.98$ \\
20		−Routing and root load (Thm.~\ref{thm:routing}) & Partially: administrative levels coded, loads not & Central collapses below capacity ratio 1; federated degrades gracefully \\
	20	+Routing and root load (Thm.~\ref{thm:routing}) & Partially: administrative levels coded, loads not & At equal budgets, central collapses below budget multiple 1; federated keeps most of its success \\
21	21	Backlog (Thm.~\ref{thm:backlog}) & Not observable; $P$ is a proxy under C1 & By construction \\
22	22	Overload $\to$ breakdown (C2) & Tested through $P$: positive but not beyond components & Not modelled \\
23	23	Smooth vs threshold (C3) & No threshold in Seshat; negligible nonlinearity in modern panel & Convex latency in utilization \\
		@@ -28,16 +28,16 @@ Smooth vs threshold (C3) & No threshold in Seshat; negligible nonlinearity in mo
28	28	\subsection{The model against simpler rivals}
29	29	\label{sec:synth-rivals}
30	30
31		−The pre-registration named the rivals and the results rank them. In Seshat, a scale-only model explains crisis severity as well as the mismatch model, in and out of sample; the placebo built from production infrastructure does as well; capacity alone does nothing. The simplest reading is \citeauthor{johnson1982}'s: larger polities face more that can go wrong when a crisis strikes, and whether their administrations were ``adequate'' is not something the presence codes can tell us. In the modern panel, a capacity-only model explains instability onsets as well as or better than the mismatch model; so does development; population adds little. The simplest reading is the state-capacity literature's \citep{besley2009,hanson2021,goldstone2010}: states with impartial administrations, predictable law and fiscal and territorial reach experience fewer civil wars, coups and irregular transitions, and the mismatch index inherits that association with its sign reversed. In neither panel does the \emph{difference} between scale and capacity carry information that its components do not.
	31	+The pre-registration named the rivals and the results rank them. In Seshat, a scale-only model explains crisis severity as well as the mismatch model, in and out of sample; the placebo built from production infrastructure is not distinguishable from $P$; capacity alone does nothing, and the equal-and-opposite restriction is not rejected only because $\beta_K$ is imprecise. The simplest reading is \citeauthor{johnson1982}'s: larger polities face more that can go wrong when a crisis strikes, and whether their administrations were ``adequate'' is not something the presence codes can tell us. In the modern panel, a capacity-only model explains instability onsets as well as or better than the mismatch model; so does development; population adds little. The simplest reading is the state-capacity literature's \citep{besley2009,hanson2021,goldstone2010}: states with impartial administrations, predictable law and fiscal and territorial reach experience fewer civil wars, coups and irregular transitions, and the mismatch index inherits that association with its sign reversed. In neither panel does the \emph{difference} between scale and capacity carry information that its components do not.
32	32
33	33	The rival mechanisms named in the mandate fare as follows. Military competition: the warfare-technology index is included as a covariate in every specification and does not displace the scale association; it is not itself associated with severity once scale is in the model. Environmental shocks: an epidemic accompanying the crisis is a covariate; the climate-mediated pathway of \citet{hoyer2023} is not testable with Polaris alone. Geography: absorbed by region effects, and the leave-one-region-out results show no region driving the estimates. Resource base: the production-infrastructure index is a covariate and, used as a placebo, mimics capacity. Inequality and elite conflict: not measurable at the polity level in Polaris and only through regime-type proxies in V-Dem; the structural-demographic mechanism \citep{turchin2003,turchin2009} remains an open rival that this paper cannot distinguish from scale. Cultural diffusion and selection: addressed by region and family controls for inference and by the crisis-selection caveat, not for identification. Coercion: the military professionalization index is a covariate in H1 and was not associated with severity.
34	34
35	35	\subsection{Consistency with simulation-derived predictions}
36	36	\label{sec:synth-consistency}
37	37
38		−The simulation predicts that institutions raise success and lower operating cost per successful task, which agrees with the positive association of capacity with scale and, in the modern panel, with the negative association of capacity with instability; agreement of this kind is weak evidence because the data cannot distinguish capacity enabling scale from scale demanding capacity. The simulation also predicts a sharp utilization effect---hierarchies fail when relayed load exceeds administrative capacity---whose historical analogue would be a threshold in the mismatch index. The historical tests find no threshold in Seshat and a negligible nonlinearity in the modern panel. Either there is no utilization threshold in history, or the proxy does not measure utilization; the collinearity of the proxy's components makes the second reading at least as likely, and the paper cannot tell them apart. Disagreement is more informative than agreement here: it marks the precise place where better data would be decisive.
	38	+The simulation predicts that institutions lower operating cost per successful task (and raise success by the construction of their error parameters), which agrees with the positive association of capacity with scale and, in the modern panel, with the negative association of capacity with instability; agreement of this kind is weak evidence because the data cannot distinguish capacity enabling scale from scale demanding capacity. The simulation also predicts a sharp utilization effect---hierarchies fail when relayed load exceeds administrative capacity---whose historical analogue would be a threshold in the mismatch index. The historical tests find no threshold in Seshat and, in the modern panel, a nonlinearity that meets the pre-registered rule but is negligible in size. Either there is no utilization threshold in history, or the proxy does not measure utilization; the collinearity of the proxy's components makes the second reading at least as likely, and the paper cannot tell them apart. Disagreement is more informative than agreement here: it marks the precise place where better data would be decisive.
39	39
40	40	\subsection{Scope conditions and failed predictions}
41	41	\label{sec:synth-scope}
42	42
43		−Three statements survive, each with its scope. First, capacity and scale are positively associated across 317 historical polities, 579 ethnographic societies and 14,647 modern country-years, with the sign stable within regions and eras and over fifty-year differences; the direction is not identified. Second, among the 127 historical crises with a defined pre-crisis index, a larger polity's crisis is recorded with more adverse consequences, at about 0.26--0.34 log-odds per standard deviation of scale; this is a statement about scale, not about mismatch. Third, modern states with higher administrative, legal, fiscal and territorial capacity experience fewer instability onsets in the following five years, at about $-0.63$ log-odds per standard deviation; this replicates a known result and is as strongly related to past as to future instability. Four predictions failed. The mismatch index does not outperform its components out of sample in either panel. The placebo index performs comparably in Seshat. The negative control in the modern panel is as strong as the forward association. There is no threshold. By the rule written before the data were examined, the universal theory of coordination, institutions and civilizational complexity is not supported by the historical and modern evidence available in 2026, and the paper is written as the test that did not confirm it.
	43	+Three statements survive, each with its scope. First, capacity and scale are positively associated across 317 historical polities, 579 ethnographic societies and 14,647 modern country-years, with the sign stable within regions and eras and over fifty-year differences; the direction is not identified. Second, among the 127 historical crises with a defined pre-crisis index, a larger polity's crisis is recorded with more adverse consequences, at about 0.42--0.44 log-odds per standard deviation of scale on the common sample; this is a statement about scale, not about mismatch. Third, modern states with higher administrative, legal, fiscal and territorial capacity experience fewer instability onsets in the following five years, at about $-0.63$ log-odds per standard deviation; this replicates a known result and is as strongly related to past as to future instability. Five predictions failed or fell short. The mismatch index does not outperform its components out of sample in either panel. The restriction it imposes is rejected in the modern panel. The placebo index is not distinguishable from it in Seshat. The negative control in the modern panel passes by the letter but by less than one standard error. There is no threshold in Seshat and only a small nonlinearity in the modern panel; and the secondary polity-duration outcome runs against the hypothesis. By the rule written before the data were examined, the universal theory of coordination, institutions and civilizational complexity is not supported by the historical and modern evidence available in 2026, and the paper is written as the test that did not confirm it.

modified paper/sections/theory.tex +3 −0

		@@ -109,6 +109,9 @@ $\sum_x p(x)\ell(x) < \sum_x p(x)(-\log_2 p(x) + 1) = H(p) + 1$. The optimal cod
109	109	(iii) Subtract (ii) from (i).
110	110	\end{proof}
111	111
	112	+\noindent\emph{Remark.} Because the fixed-length code of $\lceil\log_2 s\rceil$ bits is itself prefix-free, the optimal code satisfies $L_p^* \le \lceil\log_2 s\rceil$ and the saving $G(p)$ is never negative; (iii) is only a lower bound. A suboptimal prefix code with Shannon lengths $\lceil -\log_2 p(x)\rceil$ also satisfies $H(p) \le L < H(p)+1$ but can exceed $\lceil\log_2 s\rceil$ when $p$ is nearly uniform; the simulations of Section~\ref{sec:sim} use Shannon lengths so that Theorem~\ref{thm:stale} applies verbatim, and this is why one cell there shows a code longer than raw description.
	113	+
	114	+
112	115	\paragraph{Remark.}
113	116	Theorem~\ref{thm:coding} is the source-coding theorem of \citet{shannon1948} applied to the plan set. Its content
114	117	for the paper is only this: \emph{a shared code converts a worst-case identification cost into an average-case

modified paper/tables/tab_h2_modern.tex +13 −12

		@@ -1,15 +1,16 @@
1		−C1 primary: Y456 $\sim$ P + X + region FE + year FE & P\_modern\_z & 0.398$^{***}$ & 0.089 & 14637 & 2692.00 \\
	1	+C1 primary: Y456 $\sim$ P + X + region FE + year FE & P\_modern\_z & 0.392$^{***}$ & 0.089 & 14637 & 2692.00 \\
2	2	C2: Y456 $\sim$ P + FE only & P\_modern\_z & 0.723$^{***}$ & 0.080 & 15687 & 2981.00 \\
3		−C3: Y456 $\sim$ P + K + X + FE (P given K) & P\_modern\_z & 0.229$^{**}$ & 0.094 & 14637 & 2692.00 \\
4		−B1 baseline population only & D\_modern\_z & 0.126$^{*}$ & 0.071 & 14754 & 2710.00 \\
5		−B2 baseline capacity only & K\_modern\_z & -0.632$^{***}$ & 0.126 & 15675 & 2957.00 \\
6		−B3 baseline development (gdppc) only & log\_gdppc & -0.557$^{***}$ & 0.095 & 15862 & 2980.00 \\
7		−B4 baseline persistence & y456\_b5 & 0.136 & 0.133 & 15806 & 2968.00 \\
8		−B5 D and K separately & D\_modern\_z & 0.170$^{**}$ & 0.070 & 14637 & 2692.00 \\
9		−B5 D and K separately & K\_modern\_z & -0.707$^{***}$ & 0.137 & 14637 & -- \\
10		−Placebo: P with polyarchy in place of K (polyarchy not separately controlled) & P\_placebo\_z & 0.197$^{***}$ & 0.059 & 14754 & 2710.00 \\
11		−Placebo comparison: P\_main with the same controls as the placebo & P\_modern\_z & 0.446$^{***}$ & 0.083 & 14676 & 2706.00 \\
	3	+C3: Y456 $\sim$ P + K + X + FE (reparametrization of B5: beta\_P = beta\_D) & P\_modern\_z & 0.225$^{**}$ & 0.094 & 14637 & 2692.00 \\
	4	+B1 baseline population only & D\_modern\_z & 0.122$^{*}$ & 0.071 & 14754 & 2710.00 \\
	5	+B2 baseline capacity only & K\_modern\_z & -0.631$^{***}$ & 0.131 & 15675 & 2957.00 \\
	6	+B3 baseline development (gdppc) only & log\_gdppc & -0.582$^{***}$ & 0.096 & 15862 & 2980.00 \\
	7	+B4 baseline persistence & y456\_b5 & 0.246$^{*}$ & 0.138 & 15806 & 2968.00 \\
	8	+B5 D and K separately & D\_modern\_z & 0.166$^{**}$ & 0.070 & 14637 & 2692.00 \\
	9	+B5 D and K separately & K\_modern\_z & -0.701$^{***}$ & 0.139 & 14637 & -- \\
	10	+B5 Wald test of beta\_D + beta\_K = 0 (restriction implied by P) & D+K & -0.535$^{***}$ & 0.150 & 14637 & -- \\
	11	+Placebo: P with polyarchy in place of K (polyarchy not separately controlled) & P\_placebo\_z & 0.202$^{***}$ & 0.058 & 14754 & 2710.00 \\
	12	+Placebo comparison: P\_main with the same controls as the placebo & P\_modern\_z & 0.448$^{***}$ & 0.082 & 14676 & 2706.00 \\
12	13	Negative control: past events (t-5..t-1) $\sim$ P\_t + gdppc + polyarchy + FE (no duration terms) & P\_modern\_z & 0.436$^{***}$ & 0.098 & 14608 & 2757.00 \\
13	14	Forward comparison for the negative control: future events $\sim$ P\_t + gdppc + polyarchy + FE & P\_modern\_z & 0.475$^{***}$ & 0.099 & 14637 & 2692.00 \\
14		−C4: conditional logit, country FE + decade FE & P\_modern\_z & 1.314$^{***}$ & 0.098 & 14637 & 2692.00 \\
15		−C5: linear probability, country + year FE & P\_modern\_z\_dm & 0.111$^{***}$ & 0.031 & 14637 & 2692.00 \\
	15	+C4: conditional logit, country FE + decade FE & P\_modern\_z & 1.329$^{***}$ & 0.099 & 14637 & 2692.00 \\
	16	+C5: linear probability, country + year FE & P\_modern\_z\_dm & 0.109$^{***}$ & 0.032 & 14637 & 2692.00 \\

modified paper/tables/tab_h2_modern_cv.tex +20 −20

		@@ -1,20 +1,20 @@
1		−B0 FE+X & forward & oos\_loglik & 2 & 0.50 & -0.0002 \\
2		−B0 FE+X & forward & auc & 2 & 1.00 & +0.0047 \\
3		−B1 D & forward & oos\_loglik & 2 & 0.50 & +0.0002 \\
4		−B1 D & forward & auc & 2 & 1.00 & +0.0044 \\
5		−B2 K & forward & oos\_loglik & 2 & 0.00 & -0.0029 \\
6		−B2 K & forward & auc & 2 & 0.00 & -0.0041 \\
7		−B4 persistence & forward & oos\_loglik & 2 & 1.00 & +0.0016 \\
8		−B4 persistence & forward & auc & 2 & 1.00 & +0.0096 \\
9		−B5 D+K & forward & oos\_loglik & 2 & 0.00 & -0.0019 \\
10		−B5 D+K & forward & auc & 2 & 0.00 & -0.0059 \\
11		−B0 FE+X & loro & oos\_loglik & 6 & 0.83 & -0.0004 \\
12		−B0 FE+X & loro & auc & 6 & 0.83 & +0.0077 \\
13		−B1 D & loro & oos\_loglik & 6 & 0.83 & +0.0007 \\
14		−B1 D & loro & auc & 6 & 1.00 & +0.0074 \\
15		−B2 K & loro & oos\_loglik & 6 & 0.17 & -0.0093 \\
16		−B2 K & loro & auc & 6 & 0.50 & +0.0018 \\
17		−B4 persistence & loro & oos\_loglik & 6 & 0.83 & -0.0008 \\
18		−B4 persistence & loro & auc & 6 & 0.67 & +0.0075 \\
19		−B5 D+K & loro & oos\_loglik & 6 & 0.17 & -0.0084 \\
20		−B5 D+K & loro & auc & 6 & 0.50 & -0.0012 \\
	1	+B0 FE+X & forward & oos\_loglik & 3 & 0.67 & +0.0031 \\
	2	+B0 FE+X & forward & auc & 3 & 1.00 & +0.0093 \\
	3	+B1 D & forward & oos\_loglik & 3 & 0.67 & +0.0022 \\
	4	+B1 D & forward & auc & 3 & 1.00 & +0.0080 \\
	5	+B2 K & forward & oos\_loglik & 3 & 0.33 & +0.0221 \\
	6	+B2 K & forward & auc & 3 & 0.33 & +0.0073 \\
	7	+B4 persistence & forward & oos\_loglik & 3 & 0.67 & +0.0049 \\
	8	+B4 persistence & forward & auc & 3 & 1.00 & +0.0136 \\
	9	+B5 D+K & forward & oos\_loglik & 3 & 0.33 & +0.0215 \\
	10	+B5 D+K & forward & auc & 3 & 0.33 & +0.0057 \\
	11	+B0 FE+X & loro & oos\_loglik & 6 & 0.67 & -0.0028 \\
	12	+B0 FE+X & loro & auc & 6 & 0.83 & +0.0071 \\
	13	+B1 D & loro & oos\_loglik & 6 & 0.67 & -0.0014 \\
	14	+B1 D & loro & auc & 6 & 1.00 & +0.0070 \\
	15	+B2 K & loro & oos\_loglik & 6 & 0.17 & -0.0128 \\
	16	+B2 K & loro & auc & 6 & 0.50 & +0.0014 \\
	17	+B4 persistence & loro & oos\_loglik & 6 & 0.50 & -0.0047 \\
	18	+B4 persistence & loro & auc & 6 & 0.83 & +0.0073 \\
	19	+B5 D+K & loro & oos\_loglik & 6 & 0.17 & -0.0108 \\
	20	+B5 D+K & loro & auc & 6 & 0.50 & -0.0011 \\

modified paper/tables/tab_h2_modern_cv_summary.tex +12 −12

		@@ -1,12 +1,12 @@
1		−forward & B0 FE+X & -0.334 & 0.768 & 0.102 & 0.97 \\
2		−forward & B1 D & -0.334 & 0.769 & 0.103 & 0.96 \\
3		−forward & B2 K & -0.331 & 0.777 & 0.101 & 0.99 \\
4		−forward & B4 persistence & -0.336 & 0.763 & 0.103 & 0.95 \\
5		−forward & B5 D+K & -0.332 & 0.779 & 0.102 & 0.97 \\
6		−forward & C1 P & -0.334 & 0.773 & 0.103 & 0.95 \\
7		−loro & B0 FE+X & -0.518 & 0.674 & 0.142 & 0.44 \\
8		−loro & B1 D & -0.519 & 0.674 & 0.142 & 0.44 \\
9		−loro & B2 K & -0.509 & 0.680 & 0.139 & 0.43 \\
10		−loro & B4 persistence & -0.517 & 0.674 & 0.141 & 0.44 \\
11		−loro & B5 D+K & -0.510 & 0.683 & 0.140 & 0.43 \\
12		−loro & C1 P & -0.518 & 0.681 & 0.142 & 0.44 \\
	1	+forward & B0 FE+X & -0.425 & 0.722 & 0.137 & 0.79 \\
	2	+forward & B1 D & -0.424 & 0.724 & 0.137 & 0.79 \\
	3	+forward & B2 K & -0.444 & 0.724 & 0.146 & 0.77 \\
	4	+forward & B4 persistence & -0.427 & 0.718 & 0.138 & 0.78 \\
	5	+forward & B5 D+K & -0.443 & 0.726 & 0.146 & 0.77 \\
	6	+forward & C1 P & -0.422 & 0.732 & 0.136 & 0.79 \\
	7	+loro & B0 FE+X & -0.527 & 0.671 & 0.142 & 0.40 \\
	8	+loro & B1 D & -0.528 & 0.671 & 0.143 & 0.40 \\
	9	+loro & B2 K & -0.517 & 0.677 & 0.140 & 0.39 \\
	10	+loro & B4 persistence & -0.525 & 0.671 & 0.142 & 0.40 \\
	11	+loro & B5 D+K & -0.519 & 0.679 & 0.141 & 0.39 \\
	12	+loro & C1 P & -0.530 & 0.678 & 0.143 & 0.39 \\

modified paper/tables/tab_h2_modern_robust.tex +12 −12

		@@ -1,12 +1,12 @@
1		−R: horizon h=1 & P\_modern\_z & 0.326$^{***}$ & 0.091 & 14635 & 902.00 \\
2		−R: horizon h=3 & P\_modern\_z & 0.386$^{***}$ & 0.087 & 14636 & 1911.00 \\
3		−R: horizon h=10 & P\_modern\_z & 0.453$^{***}$ & 0.101 & 14639 & 4143.00 \\
4		−R: component y4\_f5 & P\_modern\_z & 1.011$^{***}$ & 0.197 & 10279 & 566.00 \\
5		−R: component y5\_f5 & P\_modern\_z & 0.196 & 0.137 & 10853 & 1460.00 \\
6		−R: component y6\_f5 & P\_modern\_z & 0.245$^{***}$ & 0.094 & 14636 & 2068.00 \\
7		−R: K = PC1 & P\_modern\_pc1\_z & 0.397$^{***}$ & 0.090 & 14637 & 2692.00 \\
8		−R: standardized within year & P\_modern\_within\_year\_z & 0.409$^{***}$ & 0.091 & 14637 & 2692.00 \\
9		−R: D = (log pop + log area)/2 & P\_modern\_area\_z & 0.352$^{***}$ & 0.098 & 11735 & 2359.00 \\
10		−R: years $\geq$ 1900 & P\_modern\_z & 0.394$^{***}$ & 0.092 & 13083 & 2475.00 \\
11		−R: polyarchy not controlled & P\_modern\_z & 0.446$^{***}$ & 0.083 & 14676 & 2706.00 \\
12		−R: + persistence control & P\_modern\_z & 0.390$^{***}$ & 0.089 & 14598 & 2682.00 \\
	1	+R: horizon h=1 & P\_modern\_z & 0.321$^{***}$ & 0.091 & 14635 & 902.00 \\
	2	+R: horizon h=3 & P\_modern\_z & 0.380$^{***}$ & 0.087 & 14636 & 1911.00 \\
	3	+R: horizon h=10 & P\_modern\_z & 0.448$^{***}$ & 0.099 & 14639 & 4143.00 \\
	4	+R: component y4\_f5 & P\_modern\_z & 1.016$^{***}$ & 0.193 & 10279 & 566.00 \\
	5	+R: component y5\_f5 & P\_modern\_z & 0.193 & 0.137 & 10853 & 1460.00 \\
	6	+R: component y6\_f5 & P\_modern\_z & 0.240$^{**}$ & 0.094 & 14636 & 2068.00 \\
	7	+R: K = PC1 & P\_modern\_pc1\_z & 0.391$^{***}$ & 0.090 & 14637 & 2692.00 \\
	8	+R: standardized within year & P\_modern\_within\_year\_z & 0.404$^{***}$ & 0.091 & 14637 & 2692.00 \\
	9	+R: D = (log pop + log area)/2 & P\_modern\_area\_z & 0.346$^{***}$ & 0.098 & 11735 & 2359.00 \\
	10	+R: years $\geq$ 1900 & P\_modern\_z & 0.388$^{***}$ & 0.091 & 13083 & 2475.00 \\
	11	+R: polyarchy not controlled & P\_modern\_z & 0.448$^{***}$ & 0.082 & 14676 & 2706.00 \\
	12	+R: + persistence control & P\_modern\_z & 0.389$^{***}$ & 0.088 & 14598 & 2682.00 \\

modified paper/tables/tab_h2_seshat.tex +13 −11

		@@ -1,14 +1,16 @@
1	1	A1 primary: Y1 $\sim$ P + FE & P\_main\_z & 0.196$^{***}$ & 0.056 & 127 \\
2		−A2: Y1 $\sim$ P + X + FE & P\_main\_z & 0.204$^{**}$ & 0.085 & 87 \\
3		−A3: Y1 $\sim$ P + D + K + X + FE (P given components) & P\_main\_z & 0.204$^{**}$ & 0.083 & 87 \\
4		−B1 baseline D only & D\_pc1\_z & 0.257$^{*}$ & 0.138 & 94 \\
5		−B2 baseline K only & K\_mean\_z & -0.034 & 0.195 & 93 \\
6		−B3 baseline complexity PC1 & complexity\_pc1\_z & -0.010 & 0.240 & 100 \\
7		−B4 baseline persistence & prior\_crisis\_nga & 0.345 & 0.242 & 100 \\
8		−B5 D and K separately & D\_pc1\_z & 0.341$^{**}$ & 0.164 & 87 \\
9		−B5 D and K separately & K\_mean\_z & -0.213 & 0.193 & 87 \\
10		−Placebo: P built with production infrastructure instead of K & P\_placebo & 0.257$^{*}$ & 0.138 & 94 \\
	2	+A2: Y1 $\sim$ P + X + FE & P\_main\_z & 0.243$^{***}$ & 0.086 & 93 \\
	3	+A3: Y1 $\sim$ P + complexity PC1 + X + FE & P\_main\_z & 0.243$^{***}$ & 0.087 & 93 \\
	4	+B1 baseline D only & D\_pc1\_z & 0.416$^{**}$ & 0.173 & 93 \\
	5	+B2 baseline K only & K\_mean\_z & 0.041 & 0.248 & 93 \\
	6	+B3 baseline complexity PC1 & complexity\_pc1\_z & 0.078 & 0.313 & 93 \\
	7	+B4 baseline persistence & prior\_crisis\_nga & 0.373$^{*}$ & 0.202 & 93 \\
	8	+B5 D and K separately & D\_pc1\_z & 0.440$^{***}$ & 0.170 & 93 \\
	9	+B5 D and K separately & K\_mean\_z & -0.126 & 0.188 & 93 \\
	10	+B5 Wald test of beta\_D + beta\_K = 0 (restriction implied by P) & D+K & 0.314 & 0.257 & 93 \\
	11	+Placebo: P built with production infrastructure instead of K (X = warfare only) & P\_placebo & 0.192 & 0.166 & 93 \\
	12	+Placebo comparison: P with the same controls (X = warfare only) & P\_main\_z & 0.290$^{***}$ & 0.079 & 93 \\
11	13	A1b: Y1b $\sim$ P + region FE (logit, HC1; era FE dropped: separation) & P\_main\_z & -0.178 & 0.203 & 123 \\
12		−A2b: Y1b $\sim$ P + X + region FE (logit, HC1) & P\_main\_z & 0.184 & 0.324 & 86 \\
	14	+A2b: Y1b $\sim$ P + X + region FE (logit, HC1) & P\_main\_z & 0.254 & 0.316 & 91 \\
13	15	A3b: Y1b $\sim$ P (logit, HC1, no FE) & P\_main\_z & -0.068 & 0.180 & 123 \\
14		−Exploratory: reform consequences $\sim$ P + X + FE & P\_main\_z & -0.015 & 0.181 & 86 \\
	16	+Exploratory: reform consequences $\sim$ P + X + FE & P\_main\_z & -0.013 & 0.164 & 91 \\

modified paper/tables/tab_h2_seshat_cv.tex +12 −12

		@@ -1,12 +1,12 @@
1		−B0 FE only & loro & 10 & 0.80 & +0.0104 \\
2		−B1 D only & loro & 10 & 0.30 & -0.0027 \\
3		−B2 K only & loro & 10 & 0.70 & +0.0104 \\
4		−B3 complexity & loro & 10 & 0.80 & +0.0109 \\
5		−B4 persistence & loro & 10 & 0.70 & +0.0030 \\
6		−B5 D+K & loro & 10 & 0.20 & -0.0028 \\
7		−B0 FE only & kfold & 48 & 0.79 & +0.0036 \\
8		−B1 D only & kfold & 48 & 0.75 & +0.0018 \\
9		−B2 K only & kfold & 48 & 0.81 & +0.0050 \\
10		−B3 complexity & kfold & 48 & 0.85 & +0.0051 \\
11		−B4 persistence & kfold & 48 & 0.69 & +0.0019 \\
12		−B5 D+K & kfold & 48 & 0.75 & +0.0018 \\
	1	+B0 FE only & loro & 10 & 1.00 & +0.0161 \\
	2	+B1 D only & loro & 10 & 0.20 & -0.0038 \\
	3	+B2 K only & loro & 10 & 1.00 & +0.0167 \\
	4	+B3 complexity & loro & 10 & 1.00 & +0.0166 \\
	5	+B4 persistence & loro & 10 & 0.70 & +0.0110 \\
	6	+B5 D+K & loro & 10 & 0.20 & -0.0039 \\
	7	+B0 FE only & kfold & 50 & 0.86 & +0.0081 \\
	8	+B1 D only & kfold & 50 & 0.42 & -0.0005 \\
	9	+B2 K only & kfold & 50 & 0.84 & +0.0106 \\
	10	+B3 complexity & kfold & 50 & 0.84 & +0.0105 \\
	11	+B4 persistence & kfold & 50 & 0.78 & +0.0051 \\
	12	+B5 D+K & kfold & 50 & 0.44 & +0.0008 \\

modified paper/tables/tab_h2_seshat_cv_binary.tex +7 −7

		@@ -1,7 +1,7 @@
1		−A2 P & -1.350 & 0.633 & 0.258 \\
2		−B0 FE only & -1.278 & 0.641 & 0.249 \\
3		−B1 D only & -1.339 & 0.634 & 0.257 \\
4		−B2 K only & -1.318 & 0.630 & 0.259 \\
5		−B3 complexity & -1.322 & 0.629 & 0.261 \\
6		−B4 persistence & -1.341 & 0.624 & 0.263 \\
7		−B5 D+K & -1.396 & 0.621 & 0.268 \\
	1	+A2 P & -1.192 & 0.702 & 0.219 \\
	2	+B0 FE only & -1.172 & 0.716 & 0.214 \\
	3	+B1 D only & -1.196 & 0.703 & 0.218 \\
	4	+B2 K only & -1.191 & 0.708 & 0.218 \\
	5	+B3 complexity & -1.194 & 0.703 & 0.220 \\
	6	+B4 persistence & -1.206 & 0.702 & 0.224 \\
	7	+B5 D+K & -1.216 & 0.693 & 0.223 \\

modified paper/tables/tab_h2_seshat_robust.tex +20 −11

		@@ -1,11 +1,20 @@
1		−R: K = IRT latent & \_P & 0.146$^{*}$ & 0.088 & 94 \\
2		−R: D = log population only & \_P & 0.265$^{**}$ & 0.105 & 70 \\
3		−R: K = PC1 of blocks & \_P & 0.206$^{**}$ & 0.084 & 87 \\
4		−R: quasi-polities dropped & P\_main\_z & 0.153 & 0.107 & 83 \\
5		−R: crises starting $\geq$25 years after polity start & P\_main\_z & 0.250$^{***}$ & 0.091 & 76 \\
6		−R: $\geq$12 of 14 items coded & P\_main\_z & 0.184$^{**}$ & 0.091 & 84 \\
7		−R: no era FE & P\_main\_z & 0.252$^{**}$ & 0.105 & 87 \\
8		−R: no region FE & P\_main\_z & 0.232$^{***}$ & 0.084 & 87 \\
9		−R: no FE & P\_main\_z & 0.240$^{**}$ & 0.097 & 87 \\
10		−R: SE clustered by NGA & P\_main\_z & 0.204$^{**}$ & 0.085 & 87 \\
11		−R: SE clustered by polity & P\_main\_z & 0.204$^{**}$ & 0.096 & 87 \\
	1	+R: K = IRT latent & \_P & 0.175$^{**}$ & 0.082 & 100 \\
	2	+R: D = log population only & \_P & 0.279$^{**}$ & 0.109 & 76 \\
	3	+R: K = PC1 of blocks & \_P & 0.247$^{***}$ & 0.084 & 93 \\
	4	+R: D at lower bounds & \_P & 0.237$^{***}$ & 0.087 & 93 \\
	5	+R: D at upper bounds & \_P & 0.244$^{***}$ & 0.085 & 93 \\
	6	+R: inferred codes dropped & \_P & 0.131$^{**}$ & 0.060 & 77 \\
	7	+R: standardized within era & \_P & 0.265$^{***}$ & 0.091 & 93 \\
	8	+R: standardized within macro-region & \_P & 0.224$^{***}$ & 0.085 & 93 \\
	9	+R: rank-based P & \_P & 0.185$^{**}$ & 0.076 & 93 \\
	10	+R: D without settlement hierarchy & \_P & 0.249$^{***}$ & 0.074 & 84 \\
	11	+R: K without administrative levels & \_P & 0.250$^{***}$ & 0.088 & 94 \\
	12	+R: both hierarchy indicators removed & \_P & 0.247$^{***}$ & 0.068 & 85 \\
	13	+R: quasi-polities dropped & P\_main\_z & 0.205$^{**}$ & 0.101 & 87 \\
	14	+R: crises starting $\geq$25 years after polity start & P\_main\_z & 0.305$^{***}$ & 0.095 & 79 \\
	15	+R: $\geq$12 of 14 items coded & P\_main\_z & 0.250$^{***}$ & 0.091 & 88 \\
	16	+R: no era FE & P\_main\_z & 0.296$^{***}$ & 0.111 & 93 \\
	17	+R: no region FE & P\_main\_z & 0.277$^{***}$ & 0.079 & 93 \\
	18	+R: no FE & P\_main\_z & 0.288$^{***}$ & 0.095 & 93 \\
	19	+R: SE clustered by NGA & P\_main\_z & 0.243$^{***}$ & 0.074 & 93 \\
	20	+R: SE clustered by polity & P\_main\_z & 0.243$^{***}$ & 0.088 & 93 \\

modified paper/tables/tab_h2_seshat_y2y3.tex +3 −0

		@@ -3,5 +3,8 @@ Y2-2 Cox: P + X + midpoint + strata & P\_z & 0.000 & 0.068 & 208 & 1.00 & 201 \\
3	3	Y2-3 Cox: D + K + X + midpoint + strata & D\_pc1\_z & 0.128 & 0.133 & 208 & 1.14 & 201 \\
4	4	Y2-3 Cox: D + K + X + midpoint + strata & K\_mean\_z & 0.559$^{***}$ & 0.200 & 208 & 1.75 & 201 \\
5	5	Y2-4 Cox: complexity PC1 + X + strata & complexity\_pc1\_z & 0.399 & 0.291 & 208 & 1.49 & 201 \\
	6	+Y2-5 Cox: D + K + X + midpoint + polities per NGA (periodization) + strata & D\_pc1\_z & 0.082 & 0.135 & 206 & 1.09 & 200 \\
	7	+Y2-5 Cox: D + K + X + midpoint + polities per NGA (periodization) + strata & K\_mean\_z & 0.535$^{***}$ & 0.203 & 206 & 1.71 & 200 \\
	8	+Y2-5 Cox: D + K + X + midpoint + polities per NGA (periodization) + strata & n\_polities\_in\_nga\_z & 0.317$^{**}$ & 0.140 & 206 & 1.37 & 200 \\
6	9	Y3-1 logit: discontinuous succession $\sim$ P + FE & P\_z & 0.018 & 0.393 & 145 & -- & 24 \\
7	10	Y3-2 logit: + X & P\_z & 0.239 & 0.325 & 117 & -- & 18 \\

modified paper/tables/tab_h4.tex +2 −2

		@@ -1,2 +1,2 @@
1		−Seshat Y1 (10$\times$5-fold) & 48 & 0.10 & 0.08 & -0.6036 & -0.6092 & -0.6058 \\
2		−Modern Y456 (3$\times$5 country-blocked folds) & 15 & 0.80 & 0.87 & -0.4096 & -0.4087 & -0.4064 \\
	1	+Seshat Y1 (10$\times$5-fold) & 50 & 0.10 & 0.08 & -0.5998 & -0.6046 & -0.6015 \\
	2	+Modern Y456 (3$\times$5 country-blocked folds) & 15 & 0.73 & 0.80 & -0.4114 & -0.4108 & -0.4088 \\

modified paper/tables/tab_h5_modern_blocks.tex +10 −10

		@@ -1,10 +1,10 @@
1		−leave-region-out: 1.0 & 0.293$^{***}$ & 0.094 & 13252 & 1385 & 1.067 (0.432) \\
2		−leave-region-out: 2.0 & 0.328$^{***}$ & 0.099 & 12017 & 2620 & 0.542 (0.292) \\
3		−leave-region-out: 3.0 & 0.303$^{***}$ & 0.097 & 13175 & 1462 & 0.449 (0.350) \\
4		−leave-region-out: 4.0 & 0.312$^{***}$ & 0.117 & 11363 & 3274 & 0.391 (0.167) \\
5		−leave-region-out: 5.0 & 0.294$^{***}$ & 0.097 & 10717 & 3920 & 0.781 (0.285) \\
6		−leave-region-out: 6.0 & 0.470$^{***}$ & 0.093 & 12661 & 1976 & 0.234 (0.206) \\
7		−leave-period-out: 1900-1945 & 0.415$^{***}$ & 0.099 & 12616 & 2021 & 0.298 (0.187) \\
8		−leave-period-out: 1946-1989 & 0.361$^{***}$ & 0.113 & 8927 & 5710 & 0.435 (0.122) \\
9		−leave-period-out: 1990+ & 0.417$^{***}$ & 0.106 & 9285 & 5352 & 0.348 (0.161) \\
10		−leave-period-out: pre-1900 & 0.394$^{***}$ & 0.092 & 13083 & 1554 & 0.665 (0.277) \\
	1	+leave-region-out: 1.0 & 0.284$^{***}$ & 0.094 & 13252 & 1385 & 1.034 (0.436) \\
	2	+leave-region-out: 2.0 & 0.318$^{***}$ & 0.100 & 12017 & 2620 & 0.517 (0.001) \\
	3	+leave-region-out: 3.0 & 0.311$^{***}$ & 0.097 & 13175 & 1462 & 0.441 (0.354) \\
	4	+leave-region-out: 4.0 & 0.310$^{***}$ & 0.117 & 11363 & 3274 & 0.407 (0.163) \\
	5	+leave-region-out: 5.0 & 0.279$^{***}$ & 0.096 & 10717 & 3920 & 0.804 (0.314) \\
	6	+leave-region-out: 6.0 & 0.462$^{***}$ & 0.096 & 12661 & 1976 & 0.246 (0.201) \\
	7	+leave-period-out: 1900-1945 & 0.407$^{***}$ & 0.101 & 12616 & 2021 & 0.256 (0.213) \\
	8	+leave-period-out: 1946-1989 & 0.367$^{***}$ & 0.115 & 8927 & 5710 & 0.436 (0.124) \\
	9	+leave-period-out: 1990+ & 0.409$^{***}$ & 0.108 & 9285 & 5352 & 0.346 (0.161) \\
	10	+leave-period-out: pre-1900 & 0.388$^{***}$ & 0.091 & 13083 & 1554 & 0.621 (0.287) \\

modified paper/tables/tab_h5_seshat_blocks.tex +26 −16

		@@ -1,16 +1,26 @@
1		−leave-region-out: Central\_Asia & 0.225$^{***}$ & 0.086 & 84 & 13 & -- \\
2		−leave-region-out: East\_Asia & 0.229$^{**}$ & 0.107 & 64 & 30 & 0.189 (0.180) \\
3		−leave-region-out: Europe & 0.327$^{***}$ & 0.124 & 72 & 39 & 0.196 (0.161) \\
4		−leave-region-out: North\_Africa & 0.230$^{***}$ & 0.082 & 78 & 12 & -- \\
5		−leave-region-out: North\_America & 0.210$^{**}$ & 0.090 & 85 & 13 & -- \\
6		−leave-region-out: South\_America & 0.261$^{***}$ & 0.089 & 84 & 6 & -- \\
7		−leave-region-out: South\_Asia & 0.225$^{***}$ & 0.084 & 82 & 5 & -- \\
8		−leave-region-out: Southeast\_Asia\_Oceania & 0.214$^{**}$ & 0.093 & 81 & 13 & -- \\
9		−leave-region-out: Southwest\_Asia & 0.235$^{***}$ & 0.090 & 70 & 29 & 0.437 (0.234) \\
10		−leave-region-out: Sub\_Saharan\_Africa & 0.177$^{***}$ & 0.061 & 83 & 6 & -- \\
11		−leave-era-out: 1-1000 & 0.319$^{**}$ & 0.137 & 66 & 40 & 0.127 (0.177) \\
12		−leave-era-out: 1000BCE-0 & 0.204 & 0.174 & 69 & 28 & 0.340 (0.081) \\
13		−leave-era-out: 1001-1500 & 0.263$^{***}$ & 0.100 & 65 & 43 & 0.205 (0.091) \\
14		−leave-era-out: 1501-1800 & 0.224$^{***}$ & 0.073 & 73 & 31 & 0.300 (0.524) \\
15		−leave-era-out: post-1800 & 0.224$^{**}$ & 0.108 & 81 & 16 & 0.279 (0.157) \\
16		−leave-era-out: pre-1000BCE & 0.242$^{**}$ & 0.107 & 81 & 8 & -- \\
	1	+leave-region-out: Central\_Asia & 0.275$^{***}$ & 0.080 & 89 & 13 & -- \\
	2	+leave-region-out: East\_Asia & 0.269$^{***}$ & 0.103 & 70 & 30 & 0.362 (0.083) \\
	3	+leave-region-out: Europe & 0.366$^{***}$ & 0.096 & 78 & 39 & 0.187 (0.164) \\
	4	+leave-region-out: North\_Africa & 0.281$^{***}$ & 0.080 & 83 & 12 & -- \\
	5	+leave-region-out: North\_America & 0.266$^{***}$ & 0.088 & 89 & 13 & -- \\
	6	+leave-region-out: South\_America & 0.285$^{***}$ & 0.085 & 90 & 6 & -- \\
	7	+leave-region-out: South\_Asia & 0.269$^{***}$ & 0.078 & 88 & 5 & -- \\
	8	+leave-region-out: Southeast\_Asia\_Oceania & 0.255$^{***}$ & 0.090 & 86 & 13 & -- \\
	9	+leave-region-out: Southwest\_Asia & 0.264$^{***}$ & 0.087 & 75 & 29 & 0.437 (0.234) \\
	10	+leave-region-out: Sub\_Saharan\_Africa & 0.236$^{***}$ & 0.065 & 89 & 6 & -- \\
	11	+leave-era-out: 1-1000 & 0.336$^{**}$ & 0.130 & 71 & 40 & 0.193 (0.182) \\
	12	+leave-era-out: 1000BCE-0 & 0.255$^{*}$ & 0.146 & 73 & 28 & 0.390 (0.145) \\
	13	+leave-era-out: 1001-1500 & 0.310$^{***}$ & 0.112 & 70 & 43 & 0.316 (0.152) \\
	14	+leave-era-out: 1501-1800 & 0.267$^{***}$ & 0.084 & 79 & 31 & 0.325 (0.483) \\
	15	+leave-era-out: post-1800 & 0.287$^{**}$ & 0.121 & 86 & 16 & 0.208 (0.264) \\
	16	+leave-era-out: pre-1000BCE & 0.293$^{**}$ & 0.114 & 86 & 8 & -- \\
	17	+leave-sphere-out: Andes\_Amazonia & 0.285$^{***}$ & 0.085 & 90 & 6 & -- \\
	18	+leave-sphere-out: East\_Asia & 0.269$^{***}$ & 0.103 & 70 & 30 & 0.362 (0.083) \\
	19	+leave-sphere-out: Europe & 0.366$^{***}$ & 0.096 & 78 & 38 & 0.187 (0.164) \\
	20	+leave-sphere-out: Mesoamerica\_North\_America & 0.266$^{***}$ & 0.088 & 89 & 13 & -- \\
	21	+leave-sphere-out: Oceania & 0.278$^{***}$ & 0.091 & 91 & 6 & -- \\
	22	+leave-sphere-out: South\_Asia & 0.269$^{***}$ & 0.078 & 88 & 5 & -- \\
	23	+leave-sphere-out: Southeast\_Asia & 0.272$^{***}$ & 0.077 & 88 & 7 & -- \\
	24	+leave-sphere-out: Steppe\_Central\_Asia & 0.279$^{***}$ & 0.079 & 91 & 12 & -- \\
	25	+leave-sphere-out: Sub\_Saharan\_Africa & 0.236$^{***}$ & 0.065 & 89 & 6 & -- \\
	26	+leave-sphere-out: West\_Asia\_North\_Africa & 0.274$^{***}$ & 0.089 & 63 & 43 & 0.401 (0.393) \\

added paper/tables/tab_items.tex +16 −0

		@@ -0,0 +1,16 @@
	1	+decline & 93 & 0.41 & 1.11 (0.36) & 0.59 (0.29) & 0.10 (0.26) \\
	2	+collapse & 126 & 0.02 & 5.70 (2.60) & -0.49 (0.48) & -0.99 (0.39) \\
	3	+downward\_mobility & 123 & 0.45 & 0.28 (0.18) & 0.20 (0.19) & 0.05 (0.20) \\
	4	+extermination & 122 & 0.14 & 0.32 (0.34) & 0.54 (0.28) & 0.46 (0.24) \\
	5	+uprising & 123 & 0.46 & 0.15 (0.19) & 0.47 (0.23) & 0.52 (0.24) \\
	6	+revolution & 117 & 0.17 & 0.50 (0.25) & 0.66 (0.41) & 0.57 (0.38) \\
	7	+successful\_revolution & 121 & 0.10 & 0.59 (0.27) & 1.10 (0.58) & 1.44 (0.94) \\
	8	+civil\_war & 124 & 0.54 & 0.31 (0.21) & 0.45 (0.19) & 0.35 (0.20) \\
	9	+century\_plus & 126 & 0.06 & -0.73 (0.36) & -0.87 (0.38) & -0.52 (0.32) \\
	10	+fragmentation & 127 & 0.35 & 0.21 (0.19) & 0.12 (0.20) & -0.00 (0.19) \\
	11	+capital & 116 & 0.18 & 0.06 (0.23) & 0.12 (0.23) & 0.13 (0.23) \\
	12	+conquest & 124 & 0.40 & -0.12 (0.19) & -0.05 (0.19) & 0.03 (0.19) \\
	13	+assassination & 115 & 0.40 & 0.52 (0.21) & 0.66 (0.24) & 0.39 (0.25) \\
	14	+depose & 113 & 0.50 & 0.22 (0.20) & 0.54 (0.21) & 0.56 (0.23) \\
	15	+\midrule
	16	+\multicolumn{6}{l}{Mean inter-item correlation 0.08 (median 0.07); first eigenvalue share 0.19 of 14 items} \\

modified paper/tables/tab_measurement.tex +1 −0

		@@ -5,3 +5,4 @@ corr$(z(D), z(K))$ & 0.838 (IRT: 0.800) \\
5	5	P = $z(D)-z(K)$ & variance share of $z(D)+z(K)$ retained 0.166; corr. with D 0.39, with K -0.18, with complexity PC1 -0.04; $n=317$ polities, 127 crisis records \\
6	6	IRT invariance (Old World vs Americas/Oceania fits) & difficulty correlation 0.88 over 30 items \\
7	7	Modern K (6 V-Dem items) & PC1 share 0.605; corr$(D,K)$ 0.062; corr$(K,\text{polyarchy})$ 0.754; $n=15,697$ \\
	8	+Crisis sample (127 records) & K mean 0.71 (all polities 0.58); share of records with K $>$ 0.8: 0.40; corr$(P,D)$ 0.63, corr$(P,K)$ 0.02; dated share of pre-crisis facts 0.036 \\

modified paper/tables/tab_sensitivity.tex +5 −5

		@@ -1,6 +1,6 @@
1		−LPM $\beta_P$ with controls & 0.0383 \\
	1	+LPM $\beta_P$ with controls & 0.0367 \\
2	2	LPM $\beta_P$ without controls & 0.0819 \\
3		−$R^2$ with / without controls & 0.161 / 0.127 \\
4		−Oster bias-adjusted $\beta_P$ ($R_{\max}=1.3R^2$, $\delta=1$) & -0.0238 \\
5		−Partial $R^2$ of $P$ & 0.0056 \\
6		−Cinelli--Hazlett robustness value $RV_{q=1}$ & 0.072 \\
	3	+$R^2$ with / without controls & 0.159 / 0.127 \\
	4	+Oster bias-adjusted $\beta_P$ ($R_{\max}=1.3R^2$, $\delta=1$) & -0.0303 \\
	5	+Partial $R^2$ of $P$ & 0.0051 \\
	6	+Cinelli--Hazlett robustness value $RV_{q=1}$ & 0.069 \\

modified paper/tables/tab_sim_ablation.tex +2 −2

		@@ -1,8 +1,8 @@
1		−Reference (N=200, lattice, no shock): success / bits per success / welfare per agent & central: 1.000 / 28.1 / 44.4; federated: 1.000 / 28.1 / 44.4; none: 0.918 / 15.3 / 55.0; norms: 0.960 / 23.7 / 57.3; records: 1.000 / 24.8 / 59.3 \\
	1	+Reference (N=200, lattice, no shock): success / bits per success / welfare per agent & central: 0.997 / 82.8 / 42.7; federated: 1.000 / 53.3 / 43.7; none: 0.918 / 15.3 / 55.0; norms: 0.960 / 23.7 / 57.3; records: 1.000 / 24.8 / 59.3 \\
2	2	Records without verification & success 0.960, bits per success 25.8, welfare 57.2 \\
3	3	Norms, extra bits per task after a distribution shock by adaptation delay & delay 0: +0.03; delay 10: +1.92; delay 1000: +3.79 (KL $= 1.98$ bits) \\
4	4	Break-even (norms cheaper than none incl. set-up) & s=16,zipf=0.0,T=40,learn=200.0: no; s=64,zipf=0.0,T=40,learn=200.0: no; s=64,zipf=0.5,T=40,learn=200.0: no; s=64,zipf=1.0,T=10,learn=0.0: yes; s=64,zipf=1.0,T=10,learn=200.0: no; s=64,zipf=1.0,T=10,learn=1000.0: no; s=64,zipf=1.0,T=40,learn=0.0: yes; s=64,zipf=1.0,T=40,learn=200.0: no; s=64,zipf=1.0,T=40,learn=1000.0: no; s=64,zipf=1.0,T=160,learn=0.0: yes; s=64,zipf=1.0,T=160,learn=200.0: no; s=64,zipf=1.0,T=160,learn=1000.0: no; s=64,zipf=1.5,T=40,learn=200.0: no; s=64,zipf=2.5,T=40,learn=200.0: no; s=256,zipf=0.0,T=40,learn=200.0: no; s=256,zipf=1.5,T=10,learn=200.0: no; s=256,zipf=1.5,T=10,learn=1000.0: no; s=256,zipf=1.5,T=40,learn=200.0: no; s=256,zipf=1.5,T=40,learn=1000.0: no; s=256,zipf=1.5,T=160,learn=200.0: yes; s=256,zipf=1.5,T=160,learn=1000.0: no \\
5	5	Heterogeneity $\epsilon$: success none vs norms & eps=0.05,none: 0.984; eps=0.05,norms: 0.960; eps=0.25,none: 0.918; eps=0.25,norms: 0.960; eps=0.6,none: 0.808; eps=0.6,norms: 0.960 \\
6		−Federated branching $b$: success / messages per task / root load & b=4: 1.00 / 8.86 / 48; b=8: 1.00 / 6.61 / 42; b=16: 1.00 / 6.27 / 208 \\
	6	+Federated branching $b$: success / messages per task / root load & b=4: 0.93 / 8.86 / 48; b=8: 0.99 / 6.61 / 42; b=16: 1.00 / 6.27 / 209 \\
7	7	Coding gain (norms): $H(p)$, $L_{code}$, $L_{raw}$ & s=16,zipf=0.0: 4.00, 4.00, 4; s=64,zipf=0.0: 6.00, 6.00, 6; s=64,zipf=0.5: 5.78, 6.24, 6; s=64,zipf=1.0: 4.86, 5.44, 6; s=64,zipf=1.5: 3.44, 4.00, 6; s=64,zipf=2.5: 1.45, 1.96, 6; s=256,zipf=0.0: 8.00, 8.00, 8; s=256,zipf=1.5: 3.92, 4.44, 8 \\
8	8	Theorem checks & T1 ($H\le L<H+1$): True; code shorter than raw whenever $H\le\log_2 s-1$: True; T3 extra bits per recipient 1.90 vs KL 1.98 \\

modified paper/tables/tab_sim_capacity.tex +11 −11

		@@ -1,14 +1,14 @@
1		−central & 0.25 & 0.011 & 0.989 & 1.50 & 4261 \\
2		−central & 0.50 & 0.080 & 0.920 & 2.57 & 2311 \\
3		−central & 0.75 & 1.000 & 0.000 & 1.99 & 418 \\
4		−central & 1.00 & 1.000 & 0.000 & 1.01 & 302 \\
5		−central & 1.50 & 1.000 & 0.000 & 1.00 & 302 \\
6		−central & 2.00 & 1.000 & 0.000 & 1.00 & 302 \\
	1	+central & 0.25 & 0.004 & 0.996 & 2.98 & 4748 \\
	2	+central & 0.50 & 0.028 & 0.972 & 2.62 & 3286 \\
	3	+central & 0.75 & 0.088 & 0.912 & 2.49 & 1823 \\
	4	+central & 1.00 & 0.483 & 0.517 & 2.21 & 418 \\
	5	+central & 1.50 & 0.938 & 0.062 & 1.33 & 302 \\
	6	+central & 2.00 & 0.997 & 0.003 & 1.09 & 302 \\
7	7	central & 4.00 & 1.000 & 0.000 & 1.00 & 302 \\
8		−federated & 0.25 & 1.000 & 0.000 & 1.00 & 42 \\
9		−federated & 0.50 & 1.000 & 0.000 & 1.00 & 42 \\
10		−federated & 0.75 & 1.000 & 0.000 & 1.00 & 42 \\
11		−federated & 1.00 & 1.000 & 0.000 & 1.00 & 42 \\
12		−federated & 1.50 & 1.000 & 0.000 & 1.00 & 42 \\
	8	+federated & 0.25 & 0.026 & 0.974 & 2.59 & 464 \\
	9	+federated & 0.50 & 0.293 & 0.707 & 2.27 & 119 \\
	10	+federated & 0.75 & 0.886 & 0.114 & 1.46 & 44 \\
	11	+federated & 1.00 & 0.987 & 0.013 & 1.16 & 42 \\
	12	+federated & 1.50 & 1.000 & 0.000 & 1.01 & 42 \\
13	13	federated & 2.00 & 1.000 & 0.000 & 1.00 & 42 \\
14	14	federated & 4.00 & 1.000 & 0.000 & 1.00 & 42 \\

modified paper/tables/tab_sim_loss.tex +3 −3

		@@ -1,6 +1,6 @@
1	1	central & 0.5 & 108 & 1.00 & 1.00 & 1.00 \\
2		−central & 2.0 & 108 & 0.50 & 0.50 & 0.00 \\
3		−federated & 0.5 & 108 & 0.56 & 0.56 & 0.22 \\
4		−federated & 2.0 & 108 & 0.50 & 0.50 & 0.00 \\
	2	+central & 2.0 & 108 & 0.50 & 0.75 & 0.00 \\
	3	+federated & 0.5 & 108 & 0.80 & 0.95 & 0.78 \\
	4	+federated & 2.0 & 108 & 0.50 & 0.54 & 0.00 \\
5	5	norms & 2.0 & 108 & 0.06 & 1.00 & 0.00 \\
6	6	records & 2.0 & 108 & 0.00 & 0.50 & 0.00 \\

modified paper/tables/tab_sim_paired.tex +4 −4

		@@ -1,6 +1,6 @@
1	1	norms & 2.0 & 864 & 1.00 & 0.00 & 0.99 & 0.90 & -5.5 & +0.032 & +1.5 \\
2	2	records & 2.0 & 864 & 0.71 & 0.50 & 1.00 & 1.00 & -43.6 & +0.434 & +25.7 \\
3		−central & 0.5 & 864 & 0.00 & 0.00 & 0.01 & 0.00 & +145.9 & -0.442 & -28.0 \\
4		−central & 2.0 & 864 & 0.72 & 0.50 & 1.00 & 0.50 & -43.1 & +0.414 & +11.6 \\
5		−federated & 0.5 & 864 & 0.61 & 0.44 & 0.83 & 0.47 & -28.6 & +0.272 & +4.7 \\
6		−federated & 2.0 & 864 & 0.72 & 0.50 & 1.00 & 0.50 & -43.1 & +0.414 & +11.6 \\
	3	+central & 0.5 & 864 & 0.00 & 0.00 & 0.00 & 0.00 & +2450.9 & -0.493 & -32.1 \\
	4	+central & 2.0 & 864 & 0.33 & 0.25 & 1.00 & 0.50 & +19.5 & +0.409 & +9.7 \\
	5	+federated & 0.5 & 864 & 0.09 & 0.04 & 0.22 & 0.18 & +692.9 & -0.293 & -22.4 \\
	6	+federated & 2.0 & 864 & 0.50 & 0.46 & 1.00 & 0.50 & -4.6 & +0.412 & +10.5 \\

modified paper/tables/tab_sim_scaling.tex +6 −6

		@@ -1,9 +1,9 @@
1		−central & 50 & 0.939 & 8.00 & 15.5 & 30.1 & 45.0 & 0.145 \\
2		−central & 200 & 0.936 & 12.00 & 15.5 & 31.1 & 41.3 & 0.157 \\
3		−central & 800 & 0.937 & 16.00 & 15.5 & 31.2 & 41.2 & 0.158 \\
4		−federated & 50 & 0.939 & 6.81 & 15.5 & 30.1 & 45.0 & 0.145 \\
5		−federated & 200 & 0.936 & 8.06 & 15.5 & 31.1 & 41.3 & 0.157 \\
6		−federated & 800 & 0.937 & 9.34 & 15.5 & 31.2 & 41.2 & 0.158 \\
	1	+central & 50 & 0.932 & 8.00 & 50.1 & 68.0 & 43.5 & 0.149 \\
	2	+central & 200 & 0.932 & 12.00 & 73.1 & 93.7 & 39.4 & 0.158 \\
	3	+central & 800 & 0.932 & 16.00 & 96.3 & 118.9 & 38.6 & 0.160 \\
	4	+federated & 50 & 0.934 & 6.81 & 43.2 & 60.5 & 43.9 & 0.147 \\
	5	+federated & 200 & 0.936 & 8.06 & 50.4 & 69.4 & 40.3 & 0.157 \\
	6	+federated & 800 & 0.936 & 9.34 & 57.9 & 78.0 & 39.9 & 0.158 \\
7	7	none & 50 & 0.520 & 4.00 & 14.0 & 62.4 & 30.7 & 0.224 \\
8	8	none & 200 & 0.524 & 4.00 & 14.0 & 59.1 & 31.0 & 0.224 \\
9	9	none & 800 & 0.525 & 4.00 & 14.0 & 58.1 & 30.9 & 0.222 \\

modified paper/tables/tab_sim_shocks.tex +11 −11

		@@ -1,14 +1,14 @@
1		−distribution & central & 0.5 & +0.000 (0.000) & +0.523 & 0.515 \\
2		−distribution & central & 2.0 & +0.000 (0.000) & +0.523 & 0.515 \\
3		−distribution & federated & 0.5 & +0.000 (0.000) & +0.523 & 0.515 \\
4		−distribution & federated & 2.0 & +0.000 (0.000) & +0.523 & 0.515 \\
5		−distribution & none & 2.0 & +0.000 (0.000) & +0.000 & 0.515 \\
6		−distribution & norms & 2.0 & +0.000 (0.000) & +0.515 & 0.515 \\
7		−distribution & records & 2.0 & +0.000 (0.000) & +0.524 & 0.515 \\
8		−network & central & 0.5 & +0.000 (0.000) & +0.012 & 0.000 \\
9		−network & central & 2.0 & -0.076 (0.006) & +0.012 & 0.000 \\
10		−network & federated & 0.5 & -0.080 (0.012) & +0.012 & 0.000 \\
11		−network & federated & 2.0 & -0.076 (0.006) & +0.012 & 0.000 \\
	1	+distribution & central & 0.5 & +0.000 (0.000) & +6.286 & 1.030 \\
	2	+distribution & central & 2.0 & +0.000 (0.000) & +6.286 & 1.030 \\
	3	+distribution & federated & 0.5 & +0.000 (0.000) & +4.224 & 1.030 \\
	4	+distribution & federated & 2.0 & +0.000 (0.000) & +4.224 & 1.030 \\
	5	+distribution & none & 2.0 & +0.000 (0.000) & +0.000 & 1.030 \\
	6	+distribution & norms & 2.0 & +0.000 (0.000) & +1.031 & 1.030 \\
	7	+distribution & records & 2.0 & +0.000 (0.000) & +1.049 & 1.030 \\
	8	+network & central & 0.5 & +0.000 (0.000) & +0.031 & 0.000 \\
	9	+network & central & 2.0 & -0.073 (0.006) & +0.031 & 0.000 \\
	10	+network & federated & 0.5 & -0.019 (0.003) & +0.565 & 0.000 \\
	11	+network & federated & 2.0 & -0.075 (0.006) & +0.565 & 0.000 \\
12	12	network & none & 2.0 & -0.004 (0.001) & +0.000 & 0.000 \\
13	13	network & norms & 2.0 & -0.004 (0.001) & +0.001 & 0.000 \\
14	14	network & records & 2.0 & -0.070 (0.005) & -0.003 & 0.000 \\

modified pyproject.toml +0 −2

		@@ -21,8 +21,6 @@ dependencies = [
21	21	"pyarrow==21.0.0",
22	22	"pytest==8.4.2",
23	23	"tqdm==4.67.1",
24		− "pymc==5.26.1",
25		− "arviz==0.22.0",
26	24	]
27	25	[tool.setuptools.packages.find]
28	26	where = ["src"]

modified research/decisions.md +6 −0

		@@ -10,3 +10,9 @@
10	10	| D-006 | 2026-11-01 | 1 | Macro-regions and era blocks fixed in `configs/regions.yaml` (10 macro-regions, 6 eras) | Needed for FE and leave-region-out; 42 regions too many for N≈150 | before |
11	11	| D-007 | 2026-11-01 | 1 | No paid LLM API calls; literature verified via Crossref/public pages | No keys configured; mandate forbids unapproved spending | — |
12	12	| A-001 | 2026-11-01 | 4 | Added a fold-internal re-estimation of the $D$ loadings and of the standardization of $P$ for the Seshat leave-one-region-out comparison (`econometrics/leakage_check.py`) | Pre-registration §(mandate rule) requires preprocessing fitted on training folds for predictive tests; the main CV used full-sample (outcome-free) preprocessing | after seeing the main CV results, before the adversarial reviews; conclusion unchanged (P beats scale-only in 2/10 folds vs 3/10) |
	13	+| A-002 | 2026-11-01 | 9 | After the historical-anthropologist review: (i) the five robustness variants announced in §6 (range bounds, inferred codes dropped, within-era/region standardization, ranks) had not been computed at the crisis level because the crisis-level file lacked those columns and the code skipped them silently — now computed (`build_indices.py`) and reported; (ii) item-level CrisisDB analysis and dimensionality diagnostic added; (iii) cultural-sphere hold-out scheme added to H5 (`configs/regions.yaml`); (iv) periodization control (polities per NGA) added to the Cox model; (v) crisis-sample saturation of K reported; (vi) "pre-crisis" wording qualified (3.6% dated facts) | Reviewer-identified text–results mismatch and interpretive gaps | after results, after review |
	14	+| A-003 | 2026-11-01 | 9 | Pre-registration §3.5 says "13 instability items" but lists 14 (and the code uses 14) — typo in the frozen text; the list is authoritative | reviewer | after |
	15	+| A-004 | 2026-11-01 | 9 | Cross-validation repairs after the economist and computational reviews: test rows with fixed-effect levels unseen in the training fold are mapped to the training mode instead of being dropped (the first forward-chaining split and 14 Seshat model-folds had failed silently); Seshat k-fold now blocked by polity; `t_since` for countries with no observed prior event = years since panel entry plus an indicator (was: panel maximum) | silent fold failures misreported fold counts; polities with several crisis records shared identical predictors across folds | after |
	16	+| A-005 | 2026-11-01 | 9 | Simulation accounting corrected after the computational review: codeword bits counted on every relay hop; every administrator on the relay path is charged; the SAME total administrative budget for central and federated designs (ratio x expected relayed load of the central design), allocated to administrators in proportion to a warm-up measurement of their load; distribution-shock statistics restricted to non-uniform plan cells (KL = 0 otherwise); shock effects reported by bandwidth | bits ignored relay hops; only the LCA administrator was charged; the capacity ratio was defined against a load that ignored withdrawn producers; the uniform cells made the shock a no-op | after |
	17	+| A-006 | 2026-11-01 | 9 | Seshat specification repairs after the economist review: placebo index fitted without the production-infrastructure control (the previous placebo was an exact reparametrization of the scale-only baseline — its reported coefficient carried no information); `epidemic_v` dropped from the control set (it is a CrisisDB consequence item); all covariate specifications and baselines fitted on a common sample (n = 93); A3 relabelled (it conditions on the complexity component, not on D and K); C3 (modern) labelled as the reparametrization of B5 that it is; Wald test of beta_D + beta_K = 0 added; D without settlement hierarchy and K without administrative levels added as overlap checks | reviewer-identified tautology and mislabels | after |
	18	+| A-007 | 2026-11-01 | 9 | Interpretation per the written rules: the modern negative control PASSES the pre-registered literal rule (past 0.44 < future 0.48) and is reported as such with the caveat that the gap is within one SE; the modern H4 rule IS met (hinge 12/15, spline 13/15) and H4 is therefore "not rejected" in the modern panel, with the size of the improvement reported; Theorem 1 remark added (optimal code never longer than fixed-length; the simulation's Shannon code can be) | reviewer-identified misreadings of the pre-registration | after |

modified results/diagnostics/coverage_summary.json +1 −1

		@@ -1,5 +1,5 @@
1	1	{
2		− "built_at_utc": "2026-11-01T08:13:28+00:00",
	2	+ "built_at_utc": "2026-11-01T09:08:03+00:00",
3	3	"seshat": {
4	4	"polities": 626,
5	5	"regions": 42,

modified results/diagnostics/seshat_unparsed_values.csv +132 −0

		@@ -1 +1,133 @@
1	1	sheet,polity_id,variable,value_from
	2	+Social complexity,fr_carolingian_emp_2,canal,IU
	3	+Social complexity,gb_england_plantagenet,source_of_support,state salary
	4	+Social complexity,gb_england_plantagenet,source_of_support,land
	5	+Social complexity,us_united_states_of_america_reconstruction,source_of_support,salary
	6	+Social complexity,us_antebellum,source_of_support,salary
	7	+Social complexity,gb_tudor_stuart,source_of_support,state salary
	8	+Social complexity,gb_tudor_stuart,source_of_support,pensions
	9	+Social complexity,gb_tudor_stuart,source_of_support,land
	10	+Social complexity,at_habsburg_2,source_of_support,salary
	11	+Social complexity,at_habsburg_2,source_of_support,enoblement
	12	+Social complexity,tm_khwarezmid_emp,source_of_support,salary
	13	+Social complexity,mo_alawi_dyn_1,source_of_support,salary
	14	+Social complexity,gb_british_emp_1,source_of_support,salary
	15	+Social complexity,fr_france_napoleonic,source_of_support,salary
	16	+Social complexity,at_habsburg_2,source_of_support,salary
	17	+Social complexity,at_austro_hungarian_emp,source_of_support,enoblement
	18	+Social complexity,gb_british_emp_2,source_of_support,salary
	19	+Social complexity,ru_romanov_dyn_2,source_of_support,enoblement
	20	+Social complexity,ru_romanov_dyn_2,source_of_support,state salary
	21	+Social complexity,ru_romanov_dyn_1,source_of_support,governed population
	22	+Social complexity,ru_romanov_dyn_1,source_of_support,state salary
	23	+Social complexity,bd_bengal_sultanate,source_of_support,land
	24	+Social complexity,bd_chandra_dyn,source_of_support,state salary
	25	+Social complexity,bd_chandra_dyn,source_of_support,land
	26	+Social complexity,cn_qin_emp,source_of_support,land
	27	+Social complexity,cn_qin_emp,source_of_support,state salary
	28	+Social complexity,cn_southern_song_dyn,source_of_support,state salary
	29	+Social complexity,cn_eastern_zhou_warring_states,source_of_support,land
	30	+Social complexity,cn_eastern_zhou_warring_states,source_of_support,state salary
	31	+Social complexity,gr_macedonian_emp,source_of_support,governed population
	32	+Social complexity,gr_macedonian_emp,source_of_support,land
	33	+Social complexity,pt_portuguese_emp_2,source_of_support,state salary
	34	+Social complexity,bd_sena_dyn,source_of_support,state salary
	35	+Social complexity,bd_sena_dyn,source_of_support,land
	36	+Social complexity,bd_yadava_varman_dyn,source_of_support,state salary
	37	+Social complexity,bd_yadava_varman_dyn,source_of_support,land
	38	+Social complexity,et_aksum_emp_2,source_of_support,land
	39	+Social complexity,et_aksum_emp_3,source_of_support,land
	40	+Social complexity,cn_chu_dyn_spring_autumn,source_of_support,land
	41	+Social complexity,cn_chu_k_warring_states,source_of_support,land
	42	+Social complexity,cn_chu_k_warring_states,source_of_support,state salary
	43	+Social complexity,pa_cocle_1,source_of_support,not applicable
	44	+Social complexity,af_ghaznavid_emp,source_of_support,state salary
	45	+Social complexity,af_ghaznavid_emp,source_of_support,governed population
	46	+Social complexity,ma_idrisid_dyn,source_of_support,land
	47	+Social complexity,ru_kazan_khanate,source_of_support,land
	48	+Social complexity,ao_kongo_2,source_of_support,state salary
	49	+Social complexity,ao_kongo_2,source_of_support,land
	50	+Social complexity,ge_georgia_k_2,source_of_support,state salary
	51	+Social complexity,it_sicily_k_2,source_of_support,state
	52	+Social complexity,pa_la_mula_sarigua,source_of_support,not applicable
	53	+Social complexity,pa_cocle_3,source_of_support,not applicable
	54	+Social complexity,pa_cocle_2,source_of_support,not applicable
	55	+Social complexity,pa_monagrillo,source_of_support,not applicable
	56	+Social complexity,dz_numidia,source_of_support,state salary
	57	+Social complexity,dz_numidia,source_of_support,land
	58	+Social complexity,tr_pergamon_k,source_of_support,governed population
	59	+Social complexity,tr_pergamon_k,source_of_support,state salary
	60	+Social complexity,ml_songhai_1,source_of_support,land
	61	+Social complexity,th_sukhotai,source_of_support,land
	62	+Social complexity,dz_tahert,source_of_support,governed population
	63	+Social complexity,ua_skythian_k_3,source_of_support,land
	64	+Social complexity,eg_old_k_1,source_of_support,state salary
	65	+Social complexity,eg_old_k_1,source_of_support,land
	66	+Social complexity,gb_england_norman,source_of_support,land
	67	+Social complexity,de_east_francia,source_of_support,land
	68	+Social complexity,tr_ottoman_emp_4,source_of_support,state salary
	69	+Warfare,af_kushan_emp,long_wall,60
	70	+Warfare,cn_eastern_han_dyn,long_wall,7200
	71	+Warfare,cn_hmong_1,long_wall,0
	72	+Warfare,cn_later_great_jin,long_wall,1700
	73	+Warfare,cn_ming_dyn,long_wall,6259
	74	+Warfare,cn_northern_wei_dyn,long_wall,1073
	75	+Warfare,cn_northern_wei_dyn,long_wall,681
	76	+Warfare,cn_peiligang,long_wall,0
	77	+Warfare,cn_qing_dyn_1,long_wall,6259
	78	+Warfare,cn_qing_dyn_2,long_wall,6259
	79	+Warfare,cn_sui_dyn,long_wall,2414
	80	+Warfare,cn_tang_dyn_1,long_wall,48
	81	+Warfare,cn_tang_dyn_2,long_wall,48
	82	+Warfare,cn_western_han_dyn,long_wall,7200
	83	+Warfare,gh_akan,long_wall,0
	84	+Warfare,gh_ashanti_emp,long_wall,0
	85	+Warfare,gr_crete_archaic,long_wall,0
	86	+Warfare,gr_crete_classical,long_wall,0
	87	+Warfare,gr_crete_emirate,long_wall,0
	88	+Warfare,gr_crete_post_palace_2,long_wall,0
	89	+Warfare,gr_crete_geometric,long_wall,0
	90	+Warfare,gr_crete_hellenistic,long_wall,0
	91	+Warfare,gr_crete_mono_palace,long_wall,0
	92	+Warfare,gr_crete_new_palace,long_wall,0
	93	+Warfare,gr_crete_nl,long_wall,0
	94	+Warfare,gr_crete_old_palace,long_wall,0
	95	+Warfare,gr_crete_post_palace_1,long_wall,0
	96	+Warfare,gr_crete_pre_palace,long_wall,0
	97	+Warfare,id_iban_1,long_wall,0
	98	+Warfare,id_iban_2,long_wall,0
	99	+Warfare,in_achik_1,long_wall,0
	100	+Warfare,in_achik_2,long_wall,0
	101	+Warfare,iq_neo_babylonian_emp,long_wall,43
	102	+Warfare,iq_ur_dyn_3,long_wall,190
	103	+Warfare,ir_achaemenid_emp,long_wall,60
	104	+Warfare,ir_sassanid_emp_1,long_wall,195
	105	+Warfare,ir_sassanid_emp_2,long_wall,195
	106	+Warfare,it_ostrogoth_k,long_wall,85
	107	+Warfare,it_st_peter_rep_2,long_wall,50
	108	+Warfare,it_st_peter_rep_2,long_wall,85
	109	+Warfare,it_papal_state_1,long_wall,50
	110	+Warfare,it_papal_state_1,long_wall,85
	111	+Warfare,it_roman_principate,long_wall,119
	112	+Warfare,jp_jomon_1,long_wall,0
	113	+Warfare,jp_jomon_2,long_wall,0
	114	+Warfare,jp_jomon_3,long_wall,0
	115	+Warfare,jp_jomon_4,long_wall,0
	116	+Warfare,jp_jomon_5,long_wall,0
	117	+Warfare,jp_jomon_6,long_wall,0
	118	+Warfare,jp_kamakura,long_wall,20
	119	+Warfare,jp_kofun,long_wall,0
	120	+Warfare,jp_yayoi,long_wall,0
	121	+Warfare,kh_angkor_2,long_wall,12
	122	+Warfare,kh_angkor_1,long_wall,12
	123	+Warfare,kh_angkor_3,long_wall,12
	124	+Warfare,kh_funan_1,long_wall,8
	125	+Warfare,kh_funan_2,long_wall,8
	126	+Warfare,ml_jenne_jeno_1,long_wall,0
	127	+Warfare,tr_byzantine_emp_1,long_wall,56
	128	+Warfare,tr_east_roman_emp,long_wall,56
	129	+Warfare,tr_phrygian_k,long_wall,7
	130	+Warfare,us_kamehameha_k,long_wall,0
	131	+Warfare,us_haudenosaunee_1,long_wall,0
	132	+Warfare,us_haudenosaunee_2,long_wall,0
	133	+Warfare,us_proto_haudenosaunee,long_wall,0

modified results/diagnostics/seshat_zero_scale_values.csv +2 −0

		@@ -1 +1,3 @@
1	1	polity_id,variable,value_from,value_to,year_from
	2	+ni_nok_1,polity_territory,0,1.0,
	3	+ni_nok_2,polity_territory,0,1.0,

modified results/empirical/h2_modern_cv_folds.csv +55 −61

		@@ -1,61 +1,55 @@
1		−model,fold,scheme,error,oos_loglik,brier,n,events,auc,calibration_slope,calibration_intercept
2		−C1 P,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
3		−that was ",,,,,,,
4		−C1 P,"train<=1975,test1976-2005",forward,,-0.3896968836751431,0.1237495731604952,4547.0,724.0,0.7679012461648962,1.000904248066669,-0.50049484516642
5		−C1 P,"train<=2005,test>=2006",forward,,-0.2786899335020621,0.08173934298756222,2763.0,272.0,0.7783358325265071,0.9009745183312277,-0.18451748271199148
6		−B0 FE+X,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
7		−that was ",,,,,,,
8		−B0 FE+X,"train<=1975,test1976-2005",forward,,-0.3885312365556662,0.12258455414654446,4547.0,724.0,0.7591963009582883,0.9804629576806113,-0.4201046174537736
9		−B0 FE+X,"train<=2005,test>=2006",forward,,-0.2793662975136024,0.08196035119899624,2763.0,272.0,0.7776938153824356,0.9524381319645183,-0.00552774411451371
10		−B1 D,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
11		−that was ",,,,,,,
12		−B1 D,"train<=1975,test1976-2005",forward,,-0.38953429694606506,0.12328919822681099,4547.0,724.0,0.7596226243310698,0.9760294958079441,-0.4498223888534745
13		−B1 D,"train<=2005,test>=2006",forward,,-0.2792891777917825,0.08201117617476096,2763.0,272.0,0.7778546886438237,0.938482813587797,-0.0516838253065424
14		−B2 K,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
15		−that was ",,,,,,,
16		−B2 K,"train<=1975,test1976-2005",forward,,-0.3849801776989598,0.12124847509575513,4547.0,724.0,0.7753077115394897,1.0544389335550675,-0.43796340828142305
17		−B2 K,"train<=2005,test>=2006",forward,,-0.2775428640557777,0.08100321771210951,2763.0,272.0,0.7791889035823081,0.9181391137840416,-0.15288282802850622
18		−B4 persistence,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
19		−that was ",,,,,,,
20		−B4 persistence,"train<=1975,test1976-2005",forward,,-0.3899986454519143,0.12283225551877612,4547.0,724.0,0.7536183293037344,0.9610769765154307,-0.4222670176145258
21		−B4 persistence,"train<=2005,test>=2006",forward,,-0.2814896865412362,0.08259077047553244,2763.0,272.0,0.7733605686353225,0.9417246650434868,-0.03340252334286143
22		−B5 D+K,"train<=1945,test1946-1975",forward,"predict requires that you use a DataFrame when predicting from a model
23		−that was ",,,,,,,
24		−B5 D+K,"train<=1975,test1976-2005",forward,,-0.3867779967114898,0.12229824702784269,4547.0,724.0,0.7764168748907095,1.0514028348323388,-0.48428868485365767
25		−B5 D+K,"train<=2005,test>=2006",forward,,-0.27773960516612894,0.08117405300471837,2763.0,272.0,0.7815872434883226,0.8953092857560343,-0.22666742134605794
26		−C1 P,loro_1,loro,,-0.28604106913790095,0.08037555228647814,1354.0,122.0,0.7198211624441132,0.7202717664143574,-0.9178588629164446
27		−C1 P,loro_2,loro,,-0.8545932591693702,0.20707500421599742,2619.0,773.0,0.6731820417980067,0.050258268346224065,-0.7573396795185294
28		−C1 P,loro_3,loro,,-0.4889061317680881,0.1555615740338974,1459.0,281.0,0.649233576421826,0.4881779729651372,-0.5803492981909237
29		−C1 P,loro_4,loro,,-0.5141651171538956,0.1680840150947805,3273.0,747.0,0.6687891709355236,0.6355322568001194,-0.3146956772603173
30		−C1 P,loro_5,loro,,-0.47327159597995455,0.08004066802591614,3919.0,350.0,0.6612420445903214,0.0024092654533086092,-2.3134367476854316
31		−C1 P,loro_6,loro,,-0.4910218489609021,0.1625629982222162,1974.0,409.0,0.7165345227586961,0.7162949290641591,-0.7281838669211945
32		−B0 FE+X,loro_1,loro,,-0.2905671519097122,0.08087934576529415,1354.0,122.0,0.7031682456887374,0.6853066453961756,-0.9772801577802834
33		−B0 FE+X,loro_2,loro,,-0.8557860882971616,0.20776424071294652,2619.0,773.0,0.6734203108991295,0.05092649533359051,-0.7571490808018743
34		−B0 FE+X,loro_3,loro,,-0.4940076798830174,0.1563878507489776,1459.0,281.0,0.6345002386577165,0.4625231117526357,-0.6189838429934523
35		−B0 FE+X,loro_4,loro,,-0.5184356062261626,0.17118489820040494,3273.0,747.0,0.6590012729725977,0.6120481944244622,-0.4885235844252921
36		−B0 FE+X,loro_5,loro,,-0.4792678881995545,0.08205765719681389,3919.0,350.0,0.6587795701076733,-0.0012876366789915557,-2.326687313634474
37		−B0 FE+X,loro_6,loro,,-0.46723966615119294,0.15096728971620393,1974.0,409.0,0.7134880523680449,0.8138942090943181,-0.39616669238994273
38		−B1 D,loro_1,loro,,-0.2903237601924835,0.08089028706645629,1354.0,122.0,0.7046186395571641,0.6872507761325369,-0.974762992807576
39		−B1 D,loro_2,loro,,-0.8595579531967988,0.20868132549978732,2619.0,773.0,0.671969672548176,0.049545831807584415,-0.7579122253275613
40		−B1 D,loro_3,loro,,-0.4938309024221851,0.1563553468522629,1459.0,281.0,0.6356482124839132,0.463785644278651,-0.6159683262451083
41		−B1 D,loro_4,loro,,-0.5185941178507459,0.1712825438774896,3273.0,747.0,0.6588020066542231,0.6119013272384245,-0.492656870245138
42		−B1 D,loro_5,loro,,-0.47939940869806885,0.08209700835827424,3919.0,350.0,0.658969299123404,-0.0011830694601739167,-2.326310606851259
43		−B1 D,loro_6,loro,,-0.4705049397174371,0.15262618015293009,1974.0,409.0,0.7145301014708983,0.8123269544478149,-0.4705822573588978
44		−B2 K,loro_1,loro,,-0.28786037215041194,0.07939095691384952,1354.0,122.0,0.6993958909942516,0.7031237381143722,-0.9063360306219664
45		−B2 K,loro_2,loro,,-0.8350593492367202,0.2012296840735442,2619.0,773.0,0.6803973908131844,0.05664057681676268,-0.7576213474744232
46		−B2 K,loro_3,loro,,-0.4844426314424256,0.15512259474512663,1459.0,281.0,0.6513120132439927,0.5057056470586925,-0.5972627525293253
47		−B2 K,loro_4,loro,,-0.5068305225238278,0.1660612533146381,3273.0,747.0,0.6769225224996052,0.6819985530642002,-0.3178711560433633
48		−B2 K,loro_5,loro,,-0.4605597419644665,0.0756869573523636,3919.0,350.0,0.6589837089220671,0.007510949039233791,-2.2933901575924796
49		−B2 K,loro_6,loro,,-0.4773719148809536,0.15649635102052561,1974.0,409.0,0.7109337041174218,0.6232607611125086,-0.5106993939668955
50		−B4 persistence,loro_1,loro,,-0.2902062521971979,0.08069304394519712,1354.0,122.0,0.702596071960826,0.6931620017293897,-0.9686837576803118
51		−B4 persistence,loro_2,loro,,-0.8552566856589786,0.2075386171559939,2619.0,773.0,0.6741295118707067,0.05111727735849174,-0.7568173825178399
52		−B4 persistence,loro_3,loro,,-0.4941761083947784,0.15632627192482743,1459.0,281.0,0.6347298334229559,0.4645221028429317,-0.6095866208639465
53		−B4 persistence,loro_4,loro,,-0.5182643358407806,0.17115229370049168,3273.0,747.0,0.659480360078477,0.6127907685023691,-0.4869914129692327
54		−B4 persistence,loro_5,loro,,-0.4767582775988263,0.08120807318543667,3919.0,350.0,0.6616343113317057,-0.00016604004096815379,-2.3226987765983935
55		−B4 persistence,loro_6,loro,,-0.4686287022546417,0.15125941040837595,1974.0,409.0,0.7111727348711498,0.8021524888807867,-0.4081660474947933
56		−B5 D+K,loro_1,loro,,-0.2861781738979432,0.07922107862067106,1354.0,122.0,0.7074528954651906,0.7174313272085634,-0.8843839476718729
57		−B5 D+K,loro_2,loro,,-0.8360565200635777,0.20148487899444903,2619.0,773.0,0.6799320652745211,0.055229182534372125,-0.7576113141899997
58		−B5 D+K,loro_3,loro,,-0.4834795853642856,0.15508964187203983,1459.0,281.0,0.6561002724927346,0.5094523696480574,-0.5904095126185185
59		−B5 D+K,loro_4,loro,,-0.507601825011439,0.1656841392746702,3273.0,747.0,0.6790964332388939,0.6783626229919831,-0.2535390795968614
60		−B5 D+K,loro_5,loro,,-0.46063634830115807,0.07581305103877756,3919.0,350.0,0.6595392867149662,0.008808050209616622,-2.2884452023016038
61		−B5 D+K,loro_6,loro,,-0.48370822469274244,0.16007015842608954,1974.0,409.0,0.713917682807752,0.6141781750815443,-0.630542591051736
	1	+oos_loglik,brier,n,events,auc,calibration_slope,calibration_intercept,model,fold,scheme
	2	+-0.5921505517235969,0.2036483779226404,3716,915,0.652659959460224,0.49225551859917094,-0.8324385643703355,C1 P,"train<=1945,test1946-1975",forward
	3	+-0.39369679347817754,0.12376619545078918,4547,724,0.7659275134653153,0.9740631758918905,-0.5226074803124519,C1 P,"train<=1975,test1976-2005",forward
	4	+-0.2788632949828506,0.08178100286391637,2763,272,0.7767949913807354,0.9044750828709811,-0.17760924610760379,C1 P,"train<=2005,test>=2006",forward
	5	+-0.6021664727083607,0.20776457934601772,3716,915,0.6348552332012571,0.46163822624779255,-0.8623848499534179,B0 FE+X,"train<=1945,test1946-1975",forward
	6	+-0.3920457737946821,0.12259361021418254,4547,724,0.7573324007208477,0.9537678956063382,-0.4439979765841806,B0 FE+X,"train<=1975,test1976-2005",forward
	7	+-0.27975207898273574,0.08204181373320764,2763,272,0.775170023850568,0.9545745756774403,-0.0019201388331103986,B0 FE+X,"train<=2005,test>=2006",forward
	8	+-0.5984414402054218,0.2060528445846647,3716,915,0.6381592834721401,0.4712167631442125,-0.8508560449923216,B1 D,"train<=1945,test1946-1975",forward
	9	+-0.39327807695277833,0.1233006781992915,4547,724,0.7577272917771614,0.9485928271229898,-0.4741283565847332,B1 D,"train<=1975,test1976-2005",forward
	10	+-0.2795915672717714,0.08208185482167955,2763,272,0.7754105957919097,0.941759127558064,-0.04507153988491505,B1 D,"train<=2005,test>=2006",forward
	11	+-0.6649943346578067,0.23499505676228488,3716,915,0.6208223838870973,0.35963126727863287,-0.998734268649113,B2 K,"train<=1945,test1946-1975",forward
	12	+-0.38820927667282246,0.12109768479875098,4547,724,0.7742447934354872,1.034144233863238,-0.45004327718059184,B2 K,"train<=1975,test1976-2005",forward
	13	+-0.277882002666467,0.08102524501534417,2763,272,0.778269416959879,0.9225907928687551,-0.14449392903182712,B2 K,"train<=2005,test>=2006",forward
	14	+-0.6049999633624149,0.2090481901013945,3716,915,0.6345384064629533,0.468551503874905,-0.8757122901323028,B4 persistence,"train<=1945,test1946-1975",forward
	15	+-0.3921800341904585,0.12306971263207658,4547,724,0.7500628646329356,0.9367566991032754,-0.44183852930835626,B4 persistence,"train<=1975,test1976-2005",forward
	16	+-0.28233239048581454,0.08277337889962413,2763,272,0.770014699978747,0.9366696179541988,-0.043135475858448756,B4 persistence,"train<=2005,test>=2006",forward
	17	+-0.6609982820196652,0.23340515254845845,3716,915,0.623049535392317,0.36754128861711277,-0.9928585439034735,B5 D+K,"train<=1945,test1946-1975",forward
	18	+-0.39029969175617907,0.12219059464844287,4547,724,0.7750786530493682,1.0298429214718015,-0.4981728367224605,B5 D+K,"train<=1975,test1976-2005",forward
	19	+-0.27796804125213215,0.08117758430377334,2763,272,0.7801113420076984,0.8999886894233831,-0.21762327192468875,B5 D+K,"train<=2005,test>=2006",forward
	20	+-0.28443226733793214,0.08036518874466293,1354,122,0.7273126463700235,0.8207105270664086,-0.8349781303335524,C1 P,loro_1,loro
	21	+-0.8564980157815917,0.2077842005708017,2619,773,0.6726249125762637,0.047263884983215466,-0.763200293164085,C1 P,loro_2,loro
	22	+-0.5401640164966862,0.15644587572524452,1459,281,0.6406690874816475,0.2308715168671017,-0.973895331880004,C1 P,loro_3,loro
	23	+-0.5230011227525659,0.1694405576732413,3273,747,0.663378772413486,0.5652580701496799,-0.3601211958230622,C1 P,loro_4,loro
	24	+-0.47438720110187294,0.0800192563086373,3919,350,0.6550978665492535,-0.00037668692646986116,-2.3234718713793816,C1 P,loro_5,loro
	25	+-0.49935006727531767,0.16612251855162583,1974,409,0.7083043658264137,0.6874665933725674,-0.7734014512387365,C1 P,loro_6,loro
	26	+-0.28843168423803656,0.08085573936643968,1354,122,0.7111786778794975,0.7802442018740855,-0.8949945194727328,B0 FE+X,loro_1,loro
	27	+-0.8570962032056129,0.20825419804475312,2619,773,0.673846392115255,0.04895053862870052,-0.7604004990897422,B0 FE+X,loro_2,loro
	28	+-0.537528443635443,0.15738028390514824,1459,281,0.6258964769287472,0.23361208928600305,-0.9724942504895411,B0 FE+X,loro_3,loro
	29	+-0.5252289556782352,0.1722378209124822,3273,747,0.6546831294563316,0.5433699410447786,-0.5174651157698286,B0 FE+X,loro_4,loro
	30	+-0.4789181383767503,0.08147900701205486,3919,350,0.6532662210303006,-0.003573653303254131,-2.3349174855951653,B0 FE+X,loro_5,loro
	31	+-0.4735688669267427,0.1536115543603631,1974,409,0.7057156471406141,0.7762069805792448,-0.46886137929706495,B0 FE+X,loro_6,loro
	32	+-0.2883784322146087,0.0808566148870172,1354,122,0.711491377474984,0.7808663669094341,-0.8943365711656918,B1 D,loro_1,loro
	33	+-0.8608912052369636,0.20921136238844554,2619,773,0.6720958150134763,0.047194047666228436,-0.7624289545294443,B1 D,loro_2,loro
	34	+-0.5387188735204934,0.15732286810347926,1459,281,0.6273797799515435,0.22999630967472742,-0.97774621367164,B1 D,loro_3,loro
	35	+-0.5253715109864204,0.17238647760678122,3273,747,0.654423977249722,0.5432963351261859,-0.5238713870202469,B1 D,loro_4,loro
	36	+-0.4789383465866623,0.08148765285893134,3919,350,0.6533270624024338,-0.0035381216015889674,-2.3347875433559437,B1 D,loro_5,loro
	37	+-0.4771720618228803,0.15538139715419497,1974,409,0.7068545583789653,0.7756546967611821,-0.533342074485317,B1 D,loro_6,loro
	38	+-0.2867856314143838,0.07969609945296449,1354,122,0.7067609644453907,0.7754315574328577,-0.8443404073039626,B2 K,loro_1,loro
	39	+-0.8354990333151598,0.2014493230252136,2619,773,0.6809370002480802,0.055043732575728824,-0.7593269134533571,B2 K,loro_2,loro
	40	+-0.5191219383302589,0.15613001083311295,1459,281,0.6432852594118748,0.3136293944429791,-0.8586524316292553,B2 K,loro_3,loro
	41	+-0.5131841668217264,0.16704208532462278,3273,747,0.6729265968598597,0.61970508531535,-0.3460748893835036,B2 K,loro_4,loro
	42	+-0.4638124749010821,0.07612568461978836,3919,350,0.6502537725653444,0.00221499239673688,-2.3136528580868414,B2 K,loro_5,loro
	43	+-0.482470045942264,0.15888824811478489,1974,409,0.7050454236546709,0.601203458693825,-0.5574370053206993,B2 K,loro_6,loro
	44	+-0.2878214136808872,0.08049782290295285,1354,122,0.709941185863317,0.7769920937370647,-0.8895313823310835,B4 persistence,loro_1,loro
	45	+-0.855411162869489,0.20750827297127147,2619,773,0.6759739950299869,0.05006411368101781,-0.7582824408886449,B4 persistence,loro_2,loro
	46	+-0.5294821557624868,0.15736918862444432,1459,281,0.625207692633029,0.2713922016210796,-0.9056986073767649,B4 persistence,loro_3,loro
	47	+-0.5233507896765123,0.1720321013487375,3273,747,0.6561023720111376,0.5574890675856393,-0.5093360631849099,B4 persistence,loro_4,loro
	48	+-0.48016238948324347,0.08145650881512868,3919,350,0.6502825921626706,-0.004475720834185697,-2.3380487933327934,B4 persistence,loro_5,loro
	49	+-0.4733259819232942,0.1531345408459596,1974,409,0.7058999976565612,0.7737803954294711,-0.4614666376571286,B4 persistence,loro_6,loro
	50	+-0.28534759354958034,0.07952530718174977,1354,122,0.713387534596551,0.7917924872743533,-0.8224355505821479,B5 D+K,loro_1,loro
	51	+-0.8369155251906112,0.2018390546307726,2619,773,0.6800392863700263,0.05318650025509488,-0.7606109689605635,B5 D+K,loro_2,loro
	52	+-0.521994668941056,0.15604666643927462,1459,281,0.6475448465038155,0.3033430942084132,-0.8724296720169397,B5 D+K,loro_3,loro
	53	+-0.5148417550118606,0.16681514809437822,3273,747,0.6747867691404308,0.6167862286471628,-0.2885264906176205,B5 D+K,loro_4,loro
	54	+-0.46391863460773414,0.07625625516596225,3919,350,0.6509390385462115,0.003248305632604009,-2.3097157126434085,B5 D+K,loro_5,loro
	55	+-0.48996068709465335,0.16289513018460844,1974,409,0.7073544919815338,0.591578629084927,-0.6712026424109265,B5 D+K,loro_6,loro

modified results/empirical/h2_modern_cv_predictions.parquet +0 −0

Binary file not shown.

modified results/empirical/h2_modern_cv_summary.csv +12 −12

		@@ -1,13 +1,13 @@
1	1	scheme,model,oos_loglik,auc,brier,calibration_slope
2		−forward,B0 FE+X,-0.3339487670346343,0.7684450581703619,0.10227245267277035,0.9664505448225649
3		−forward,B1 D,-0.33441173736892377,0.7687386564874468,0.10265018720078598,0.9572561546978705
4		−forward,B2 K,-0.33126152087736876,0.7772483075608989,0.10112584640393232,0.9862890236695545
5		−forward,B4 persistence,-0.33574416599657525,0.7634894489695285,0.10271151299715428,0.9514008207794588
6		−forward,B5 D+K,-0.3322588009388094,0.7790020591895161,0.10173615001628053,0.9733560602941865
7		−forward,C1 P,-0.3341934085886026,0.7731185393457016,0.10274445807402871,0.9509393831989483
8		−loro,B0 FE+X,-0.5175506801111335,0.6737262817823165,0.14154021372344017,0.43723516988703176
9		−loro,B1 D,-0.5187018470129532,0.6740896553062964,0.14198878196786674,0.4372712440741397
10		−loro,B2 K,-0.5086874220331342,0.6796575384317537,0.1389979662366746,0.4297067042009617
11		−loro,B4 persistence,-0.5172150603242006,0.6739571372559702,0.14136295172005378,0.43726309987883344
12		−loro,B5 D+K,-0.5096101128885243,0.6826731059990098,0.1395604913711162,0.4305769546123561
13		−loro,C1 P,-0.517999837028352,0.6814670864914145,0.14228330197988096,0.4354907431738843
	2	+forward,B0 FE+X,-0.4246547751619261,0.7224525525908909,0.13746666776446928,0.789993565843857
	3	+forward,B1 D,-0.4237703614766572,0.7237657236804038,0.13714512586854524,0.7871895726084222
	4	+forward,B2 K,-0.44369520466569873,0.7244455314274879,0.14570599552546,0.772122098003542
	5	+forward,B4 persistence,-0.4265041293462293,0.7182053236915453,0.1382970938776984,0.7806592736441265
	6	+forward,B5 D+K,-0.4430886716759921,0.7260798434831278,0.14559111050022489,0.7657909665040991
	7	+forward,C1 P,-0.42157021339487505,0.7317941547687582,0.13639852541244865,0.7902645924540143
	8	+loro,B0 FE+X,-0.5267953820101368,0.670764424091791,0.14230310060020687,0.3964683496849264
	9	+loro,B1 D,-0.5282450717280048,0.6709287617451875,0.14277439549980825,0.39557827242269483
	10	+loro,B2 K,-0.5168122151208124,0.6765348361975367,0.1398885752284145,0.3945380368095796
	11	+loro,B4 persistence,-0.5249256488993188,0.6705679725594504,0.14199973925141574,0.40420702520334784
	12	+loro,B5 D+K,-0.5188298107325826,0.6790086611897616,0.14056292694945763,0.3933225408504259
	13	+loro,C1 P,-0.5296387817909944,0.6778979418695146,0.1433629329290356,0.39186565091875053

modified results/empirical/h2_modern_cv_wins.csv +20 −20

		@@ -1,21 +1,21 @@
1	1	model,baseline,folds,win_share,mean_diff,scheme,metric
2		−C1 P,B0 FE+X,2,0.5,-0.00024464155396830556,forward,oos_loglik
3		−C1 P,B0 FE+X,2,1.0,0.0046734811753396666,forward,auc
4		−C1 P,B1 D,2,0.5,0.00021832878032115866,forward,oos_loglik
5		−C1 P,B1 D,2,1.0,0.004379882858254913,forward,auc
6		−C1 P,B2 K,2,0.0,-0.0029318877112338437,forward,oos_loglik
7		−C1 P,B2 K,2,0.0,-0.0041297682151972825,forward,auc
8		−C1 P,B4 persistence,2,1.0,0.0015507574079726416,forward,oos_loglik
9		−C1 P,B4 persistence,2,1.0,0.009629090376173155,forward,auc
10		−C1 P,B5 D+K,2,0.0,-0.001934607649793224,forward,oos_loglik
11		−C1 P,B5 D+K,2,0.0,-0.0058835198438144665,forward,auc
12		−C1 P,B0 FE+X,6,0.8333333333333334,-0.0004491569172183722,loro,oos_loglik
13		−C1 P,B0 FE+X,6,0.8333333333333334,0.007740804709097955,loro,auc
14		−C1 P,B1 D,6,0.8333333333333334,0.000702009984601302,loro,oos_loglik
15		−C1 P,B1 D,6,1.0,0.007377431185118059,loro,auc
16		−C1 P,B2 K,6,0.16666666666666666,-0.009312414995217658,loro,oos_loglik
17		−C1 P,B2 K,6,0.5,0.0018095480596606965,loro,auc
18		−C1 P,B4 persistence,6,0.8333333333333334,-0.0007847767041513446,loro,oos_loglik
19		−C1 P,B4 persistence,6,0.6666666666666666,0.007509949235444328,loro,auc
20		−C1 P,B5 D+K,6,0.16666666666666666,-0.00838972413982759,loro,oos_loglik
21		−C1 P,B5 D+K,6,0.5,-0.0012060195075952462,loro,auc
	2	+C1 P,B0 FE+X,3,0.6666666666666666,0.003084561767051149,forward,oos_loglik
	3	+C1 P,B0 FE+X,3,1.0,0.00934160217786728,forward,auc
	4	+C1 P,B1 D,3,0.6666666666666666,0.002200148081782168,forward,oos_loglik
	5	+C1 P,B1 D,3,1.0,0.008028431088354493,forward,auc
	6	+C1 P,B2 K,3,0.3333333333333333,0.022124991270823697,forward,oos_loglik
	7	+C1 P,B2 K,3,0.3333333333333333,0.007348623341270399,forward,auc
	8	+C1 P,B4 persistence,3,0.6666666666666666,0.004933915951354302,forward,oos_loglik
	9	+C1 P,B4 persistence,3,1.0,0.013588831077212907,forward,auc
	10	+C1 P,B5 D+K,3,0.3333333333333333,0.02151845828111711,forward,oos_loglik
	11	+C1 P,B5 D+K,3,0.3333333333333333,0.00571431128563038,forward,auc
	12	+C1 P,B0 FE+X,6,0.6666666666666666,-0.0028433997808576437,loro,oos_loglik
	13	+C1 P,B0 FE+X,6,0.8333333333333334,0.007133517777723641,loro,auc
	14	+C1 P,B1 D,6,0.6666666666666666,-0.001393710062989643,loro,oos_loglik
	15	+C1 P,B1 D,6,1.0,0.006969180124327161,loro,auc
	16	+C1 P,B2 K,6,0.16666666666666666,-0.01282656667018193,loro,oos_loglik
	17	+C1 P,B2 K,6,0.5,0.0013631056719778378,loro,auc
	18	+C1 P,B4 persistence,6,0.5,-0.004713132891675594,loro,oos_loglik
	19	+C1 P,B4 persistence,6,0.8333333333333334,0.007329969310064244,loro,auc
	20	+C1 P,B5 D+K,6,0.16666666666666666,-0.01080897105841184,loro,oos_loglik
	21	+C1 P,B5 D+K,6,0.5,-0.0011107193202468484,loro,auc

added results/empirical/h2_modern_event_counts.json +1 −0

		@@ -0,0 +1 @@
	1	+{"country_years": 14637, "event_windows": 2692, "distinct_event_years_in_sample": 901, "distinct_event_years_panel": 1676}
		\ No newline at end of file

modified results/empirical/h2_modern_primary.csv +13 −12

		@@ -1,16 +1,17 @@
1	1	model,term,coef,se,ci_lo,ci_hi,p,n,llf,events
2		−C1 primary: Y456 ~ P + X + region FE + year FE,P_modern_z,0.39790378484144,0.08918214639646332,0.2231067779043719,0.5727007917785081,8.131284893393963e-06,14637,-5672.972626457216,2692.0
	2	+C1 primary: Y456 ~ P + X + region FE + year FE,P_modern_z,0.3922687947606827,0.08921893423252436,0.21739968366493498,0.5671379058564304,1.099096174755617e-05,14637,-5678.550371306823,2692.0
3	3	C2: Y456 ~ P + FE only,P_modern_z,0.7229243246287831,0.07967857684830817,0.5667543140060991,0.8790943352514671,1.1577632379582483e-19,15687,-6528.089258015958,2981.0
4		−C3: Y456 ~ P + K + X + FE (P given K),P_modern_z,0.22924427308006917,0.09387563776277562,0.04524802306502895,0.4132405230951094,0.014606161422636878,14637,-5627.074969974541,2692.0
5		−B1 baseline population only,D_modern_z,0.12594267679623367,0.07109162227063932,-0.013396902854219406,0.26528225644668674,0.07646855355079077,14754,-5769.778570873028,2710.0
6		−B2 baseline capacity only,K_modern_z,-0.6324390571653153,0.12569164780016462,-0.878794686853638,-0.3860834274769927,4.86222381583211e-07,15675,-6150.572623841682,2957.0
7		−B3 baseline development (gdppc) only,log_gdppc,-0.5569688251338336,0.09464003919975214,-0.7424633019653478,-0.3714743483023194,3.977439983972847e-09,15862,-6313.047510446031,2980.0
8		−B4 baseline persistence,y456_b5,0.13566387709403316,0.13289892982966736,-0.12481802537211484,0.39614577956018115,0.30734688842262725,15806,-6275.443015609264,2968.0
9		−B5 D and K separately,D_modern_z,0.16972157501358381,0.06950106487910981,0.033499487850528586,0.30594366217663904,0.01460616142263262,14637,-5627.074969974539,2692.0
10		−B5 D and K separately,K_modern_z,-0.7070134445104032,0.13742936225970592,-0.9763749945394268,-0.43765189448137953,2.681501580305378e-07,14637,-5627.074969974539,
11		−Placebo: P with polyarchy in place of K (polyarchy not separately controlled),P_placebo_z,0.19689648797181336,0.05871082730792415,0.08182326644828203,0.3119697094953447,0.0007974864751106516,14754,-5777.87225067883,2710.0
12		−Placebo comparison: P_main with the same controls as the placebo,P_modern_z,0.446333634501256,0.0828013326218674,0.2840430225623959,0.6086242464401161,7.029473808128925e-08,14676,-5712.556048147622,2706.0
	4	+C3: Y456 ~ P + K + X + FE (reparametrization of B5: beta_P = beta_D),P_modern_z,0.22481431764865165,0.09444802678560853,0.03969618514885895,0.40993245014844437,0.01729870781063778,14637,-5632.895171588186,2692.0
	5	+B1 baseline population only,D_modern_z,0.12150826607689458,0.07124658532961264,-0.018135041169146215,0.26115157332293537,0.0881084971966551,14754,-5776.305342993084,2710.0
	6	+B2 baseline capacity only,K_modern_z,-0.6311311734732039,0.13124190560307603,-0.888365308455233,-0.3738970384911749,1.517516830199489e-06,15675,-6166.041966794448,2957.0
	7	+B3 baseline development (gdppc) only,log_gdppc,-0.5816251465131039,0.09637108956674491,-0.770512482063924,-0.3927378109622839,1.5870118731352746e-09,15862,-6353.052539809014,2980.0
	8	+B4 baseline persistence,y456_b5,0.2458345978414203,0.13849583638276927,-0.02561724146880745,0.5172864371516481,0.07589253637739889,15806,-6295.0200354609,2968.0
	9	+B5 D and K separately,D_modern_z,0.16644184635141004,0.06992483453394764,0.029389170664872677,0.3034945220379474,0.01729870781063917,14637,-5632.895171588186,2692.0
	10	+B5 D and K separately,K_modern_z,-0.7011839796893495,0.13893510730735503,-0.9734967900117654,-0.4288711693669337,4.4916497721890495e-07,14637,-5632.895171588186,
	11	+B5 Wald test of beta_D + beta_K = 0 (restriction implied by P),D+K,-0.5347421333379395,0.15015111743715862,,,0.0003689392857068041,14637,,
	12	+Placebo: P with polyarchy in place of K (polyarchy not separately controlled),P_placebo_z,0.2020267600829871,0.05760023821421619,0.08913029318312335,0.3149232269828508,0.0004525177068698837,14754,-5786.982957537655,2710.0
	13	+Placebo comparison: P_main with the same controls as the placebo,P_modern_z,0.4477254264122505,0.08200788224449172,0.28698997721304675,0.6084608756114542,4.7736607748738476e-08,14676,-5719.316648249614,2706.0
13	14	Negative control: past events (t-5..t-1) ~ P_t + gdppc + polyarchy + FE (no duration terms),P_modern_z,0.4359470058900018,0.09815066302576257,0.24357170635950715,0.6283223054204965,8.928811539479297e-06,14608,-5813.946618858069,2757.0
14	15	Forward comparison for the negative control: future events ~ P_t + gdppc + polyarchy + FE,P_modern_z,0.475241690571331,0.09908186802752562,0.28104122923738073,0.6694421519052812,1.6149846684551553e-06,14637,-5834.237016756066,2692.0
15		−"C4: conditional logit, country FE + decade FE",P_modern_z,1.313878132712599,0.09775172551732352,1.1222847506986449,1.5054715147265532,3.4784091628447276e-41,14637,,2692.0
16		−"C5: linear probability, country + year FE",P_modern_z_dm,0.11091972996929868,0.031351674099522706,0.049470448734234174,0.17236901120436318,0.0004032914904174095,14637,-4875.021383489231,2692.0
	16	+"C4: conditional logit, country FE + decade FE",P_modern_z,1.329137190646608,0.09869472552992661,1.135695528607952,1.5225788526852642,2.440904644595632e-41,14637,,2692.0
	17	+"C5: linear probability, country + year FE",P_modern_z_dm,0.10860863215596052,0.03153444395246181,0.04680112200913537,0.17041614230278568,0.000572906371323011,14637,-4859.202679956166,2692.0

modified results/empirical/h2_modern_robustness.csv +13 −13

		@@ -1,14 +1,14 @@
1	1	model,term,coef,se,ci_lo,ci_hi,p,n,llf,events,oster_beta_full,oster_beta_short,r2_full,r2_short,partial_r2_P,robustness_value
2		−R: horizon h=1,P_modern_z,0.3257542271113133,0.09081766270455952,0.14775160821037664,0.5037568460122499,0.0003346275971506812,14635,-2809.836558711094,902.0,,,,,,
3		−R: horizon h=3,P_modern_z,0.3859734372235748,0.08713084027910899,0.2151969902765212,0.5567498841706284,9.431429129652506e-06,14636,-4668.956267136606,1911.0,,,,,,
4		−R: horizon h=10,P_modern_z,0.453180001872445,0.1005232322313484,0.2561544666990021,0.6502055370458879,6.537635724308696e-06,14639,-6759.637175029395,4143.0,,,,,,
5		−R: component y4_f5,P_modern_z,1.0108347606887813,0.19699525159210712,0.6247240675682513,1.3969454538093113,2.878019958646695e-07,10279,-1799.0294235356334,566.0,,,,,,
6		−R: component y5_f5,P_modern_z,0.19570664298345658,0.1369247476865778,-0.07266586248223594,0.4640791484491491,0.15291781385485442,10853,-3397.20354607048,1460.0,,,,,,
7		−R: component y6_f5,P_modern_z,0.24520063567506914,0.09399153647288098,0.060977224188222406,0.42942404716191585,0.009087299826968763,14636,-5000.87591521976,2068.0,,,,,,
8		−R: K = PC1,P_modern_pc1_z,0.3972334263449546,0.08951641349789637,0.22178125588907774,0.5726855968008315,9.098922495056218e-06,14637,-5673.709489767933,2692.0,,,,,,
9		−R: standardized within year,P_modern_within_year_z,0.40912714945850015,0.09087386000252914,0.23101438385354303,0.5872399150634573,6.727156439169484e-06,14637,-5672.715544195022,2692.0,,,,,,
10		−R: D = (log pop + log area)/2,P_modern_area_z,0.35201470671308055,0.09785002058259724,0.16022866637118996,0.5438007470549712,0.00032129986428979315,11735,-4781.639744406581,2359.0,,,,,,
11		−R: years ≥ 1900,P_modern_z,0.3940733263264095,0.09183407311667917,0.21407854301771834,0.5740681096351007,1.777541709685051e-05,13083,-5223.594464430906,2475.0,,,,,,
12		−R: polyarchy not controlled,P_modern_z,0.446333634501256,0.0828013326218674,0.2840430225623959,0.6086242464401161,7.029473808128925e-08,14676,-5712.556048147622,2706.0,,,,,,
13		−R: + persistence control,P_modern_z,0.39035420906242724,0.08878195519207539,0.21634157688595948,0.564366841238895,1.098715434814722e-05,14598,-5641.175860478964,2682.0,,,,,,
14		−Sensitivity (LPM): Oster bias-adjusted β at Rmax=1.3·R² (δ=1); Cinelli–Hazlett RV,P_modern_z,-0.023790190743910956,,,,,14637,,,0.03827708995287325,0.08187222469624048,0.16093354475530663,0.12702234634493959,0.00556334755160398,0.0718580292018959
	2	+R: horizon h=1,P_modern_z,0.3213567148708202,0.09083368821472391,0.1433226859699613,0.499390743771679,0.0004033865236081567,14635,-2808.0317675913648,902.0,,,,,,
	3	+R: horizon h=3,P_modern_z,0.3801552512939208,0.08743103898059633,0.208790414891952,0.5515200876958897,1.3734812300788488e-05,14636,-4671.148530889891,1911.0,,,,,,
	4	+R: horizon h=10,P_modern_z,0.44836787666130096,0.09946197757436491,0.2534224006155458,0.6433133527070561,6.546240788346645e-06,14639,-6772.628475707487,4143.0,,,,,,
	5	+R: component y4_f5,P_modern_z,1.0156500813999165,0.19346828503420777,0.6364522427328694,1.3948479200669637,1.5234871478518989e-07,10279,-1808.887349599621,566.0,,,,,,
	6	+R: component y5_f5,P_modern_z,0.19260512197836493,0.13688401515900994,-0.07568754773329456,0.4608977916900244,0.15940721605412467,10853,-3395.11361590985,1460.0,,,,,,
	7	+R: component y6_f5,P_modern_z,0.2401324779402956,0.09374377507937355,0.056394678784723457,0.4238702770958678,0.010419628530832212,14636,-4999.715992637788,2068.0,,,,,,
	8	+R: K = PC1,P_modern_pc1_z,0.3914537472844485,0.08956840530204131,0.21589967289244752,0.5670078216764495,1.2399384937912057e-05,14637,-5679.302262784689,2692.0,,,,,,
	9	+R: standardized within year,P_modern_within_year_z,0.40368351781176914,0.09086097243313335,0.2255960118428278,0.5817710237807105,8.876667058058466e-06,14637,-5678.188330888828,2692.0,,,,,,
	10	+R: D = (log pop + log area)/2,P_modern_area_z,0.34592881078299376,0.09798058420039851,0.15388686575021268,0.5379707558157748,0.0004146412839006179,11735,-4783.260595993208,2359.0,,,,,,
	11	+R: years ≥ 1900,P_modern_z,0.38849560536553623,0.09125844722362034,0.20962904880724037,0.567362161923832,2.0710314821567656e-05,13083,-5226.436878149341,2475.0,,,,,,
	12	+R: polyarchy not controlled,P_modern_z,0.4477254264122505,0.08200788224449172,0.28698997721304675,0.6084608756114542,4.7736607748738476e-08,14676,-5719.316648249614,2706.0,,,,,,
	13	+R: + persistence control,P_modern_z,0.3885024901469061,0.08845342400145681,0.21513377910405074,0.5618712011897614,1.1222480551485459e-05,14598,-5648.480224894294,2682.0,,,,,,
	14	+Sensitivity (LPM): Oster bias-adjusted β at Rmax=1.3·R² (δ=1); Cinelli–Hazlett RV,P_modern_z,-0.03033118155565985,,,,,14637,,,0.03674837760182206,0.08187222469624048,0.15913747227692998,0.12702234634493959,0.0050953734994381135,0.06887964840435769

modified results/empirical/h2_seshat_cv_binary_summary.csv +7 −7

		@@ -1,8 +1,8 @@
1	1	model,oos_loglik,auc,brier
2		−A2 P,-1.34987025159919,0.6332183406162998,0.25810946028618476
3		−B0 FE only,-1.277565410590492,0.6410582685455134,0.24859475238814943
4		−B1 D only,-1.3389349056151325,0.6337711324956223,0.25689550126921973
5		−B2 K only,-1.3182015924857153,0.6299753251283864,0.25865358661769416
6		−B3 complexity,-1.3215637613178275,0.6290379971502421,0.2609623970447837
7		−B4 persistence,-1.3414549962341589,0.6243098808787584,0.26262025957466023
8		−B5 D+K,-1.3956401024749936,0.621057655496431,0.26819092780677006
	2	+A2 P,-1.1917453541967393,0.702196396250808,0.2187700900214412
	3	+B0 FE only,-1.171989849851237,0.7160045688788336,0.21372218939484727
	4	+B1 D only,-1.1955364336529728,0.7025330568124686,0.21753313231483934
	5	+B2 K only,-1.1914111522564088,0.7079322990734755,0.21828945568512384
	6	+B3 complexity,-1.1940069389354726,0.7025283469798176,0.22048718833428932
	7	+B4 persistence,-1.2062658112793236,0.7020088522588523,0.22370053411005777
	8	+B5 D+K,-1.2164775328144117,0.6930657580981111,0.2227231695917103

modified results/empirical/h2_seshat_cv_folds.csv +421 −435

		@@ -1,435 +1,421 @@
1		−oos_loglik,n,model,fold,n_train,scheme,error
2		−-0.4880613524495527,3.0,A2 P,loo_Central_Asia,84.0,leave-one-macro-region-out,
3		−-0.6179689032006479,23.0,A2 P,loo_East_Asia,64.0,leave-one-macro-region-out,
4		−-0.5431030904486999,15.0,A2 P,loo_Europe,72.0,leave-one-macro-region-out,
5		−-0.6351336323453667,9.0,A2 P,loo_North_Africa,78.0,leave-one-macro-region-out,
6		−-0.6808994688904817,2.0,A2 P,loo_North_America,85.0,leave-one-macro-region-out,
7		−-0.5396641871002942,3.0,A2 P,loo_South_America,84.0,leave-one-macro-region-out,
8		−-0.5556779398097061,5.0,A2 P,loo_South_Asia,82.0,leave-one-macro-region-out,
9		−-0.5444416294257195,6.0,A2 P,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
10		−-0.6043015795250417,17.0,A2 P,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
11		−-0.5829162440745959,4.0,A2 P,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
12		−-0.49648122248801574,3.0,B0 FE only,loo_Central_Asia,84.0,leave-one-macro-region-out,
13		−-0.6250604270020237,23.0,B0 FE only,loo_East_Asia,64.0,leave-one-macro-region-out,
14		−-0.5398083196722638,15.0,B0 FE only,loo_Europe,72.0,leave-one-macro-region-out,
15		−-0.6375346096250804,9.0,B0 FE only,loo_North_Africa,78.0,leave-one-macro-region-out,
16		−-0.7220199085483197,2.0,B0 FE only,loo_North_America,85.0,leave-one-macro-region-out,
17		−-0.5217824682759374,3.0,B0 FE only,loo_South_America,84.0,leave-one-macro-region-out,
18		−-0.5617711756893108,5.0,B0 FE only,loo_South_Asia,82.0,leave-one-macro-region-out,
19		−-0.5720629248539654,6.0,B0 FE only,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
20		−-0.6044692070236013,17.0,B0 FE only,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
21		−-0.6150321505709686,4.0,B0 FE only,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
22		−-0.48416191268852043,3.0,B1 D only,loo_Central_Asia,84.0,leave-one-macro-region-out,
23		−-0.6154927045326366,23.0,B1 D only,loo_East_Asia,64.0,leave-one-macro-region-out,
24		−-0.5543058313879182,15.0,B1 D only,loo_Europe,72.0,leave-one-macro-region-out,
25		−-0.6322613087171535,9.0,B1 D only,loo_North_Africa,78.0,leave-one-macro-region-out,
26		−-0.6829120580961434,2.0,B1 D only,loo_North_America,85.0,leave-one-macro-region-out,
27		−-0.5219812850414702,3.0,B1 D only,loo_South_America,84.0,leave-one-macro-region-out,
28		−-0.5572203126706954,5.0,B1 D only,loo_South_Asia,82.0,leave-one-macro-region-out,
29		−-0.5380005961660977,6.0,B1 D only,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
30		−-0.6025306317512673,17.0,B1 D only,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
31		−-0.5762049031856433,4.0,B1 D only,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
32		−-0.4962374116819824,3.0,B2 K only,loo_Central_Asia,84.0,leave-one-macro-region-out,
33		−-0.6248569672964271,23.0,B2 K only,loo_East_Asia,64.0,leave-one-macro-region-out,
34		−-0.5407854830332525,15.0,B2 K only,loo_Europe,72.0,leave-one-macro-region-out,
35		−-0.6371071925590664,9.0,B2 K only,loo_North_Africa,78.0,leave-one-macro-region-out,
36		−-0.724521093462311,2.0,B2 K only,loo_North_America,85.0,leave-one-macro-region-out,
37		−-0.5207608968080311,3.0,B2 K only,loo_South_America,84.0,leave-one-macro-region-out,
38		−-0.5617890023264124,5.0,B2 K only,loo_South_Asia,82.0,leave-one-macro-region-out,
39		−-0.5716201572503757,6.0,B2 K only,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
40		−-0.6042279843058415,17.0,B2 K only,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
41		−-0.614480512511814,4.0,B2 K only,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
42		−-0.49613569030529164,3.0,B3 complexity,loo_Central_Asia,84.0,leave-one-macro-region-out,
43		−-0.6253154805860182,23.0,B3 complexity,loo_East_Asia,64.0,leave-one-macro-region-out,
44		−-0.5408574084779093,15.0,B3 complexity,loo_Europe,72.0,leave-one-macro-region-out,
45		−-0.6379062547566421,9.0,B3 complexity,loo_North_Africa,78.0,leave-one-macro-region-out,
46		−-0.7262541192578305,2.0,B3 complexity,loo_North_America,85.0,leave-one-macro-region-out,
47		−-0.5219368338910513,3.0,B3 complexity,loo_South_America,84.0,leave-one-macro-region-out,
48		−-0.5615010883228737,5.0,B3 complexity,loo_South_Asia,82.0,leave-one-macro-region-out,
49		−-0.5717786784269936,6.0,B3 complexity,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
50		−-0.6047361088148459,17.0,B3 complexity,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
51		−-0.6144770465919185,4.0,B3 complexity,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
52		−-0.48265473851259905,3.0,B4 persistence,loo_Central_Asia,84.0,leave-one-macro-region-out,
53		−-0.6205265583512978,23.0,B4 persistence,loo_East_Asia,64.0,leave-one-macro-region-out,
54		−-0.5477744449447155,15.0,B4 persistence,loo_Europe,72.0,leave-one-macro-region-out,
55		−-0.6320692993979519,9.0,B4 persistence,loo_North_Africa,78.0,leave-one-macro-region-out,
56		−-0.6921181974675796,2.0,B4 persistence,loo_North_America,85.0,leave-one-macro-region-out,
57		−-0.518826113170134,3.0,B4 persistence,loo_South_America,84.0,leave-one-macro-region-out,
58		−-0.5597455042562001,5.0,B4 persistence,loo_South_Asia,82.0,leave-one-macro-region-out,
59		−-0.5586279807589243,6.0,B4 persistence,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
60		−-0.6088468482433093,17.0,B4 persistence,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
61		−-0.6014078618663397,4.0,B4 persistence,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
62		−-0.4841547538303477,3.0,B5 D+K,loo_Central_Asia,84.0,leave-one-macro-region-out,
63		−-0.6153225387136638,23.0,B5 D+K,loo_East_Asia,64.0,leave-one-macro-region-out,
64		−-0.5544283583261507,15.0,B5 D+K,loo_Europe,72.0,leave-one-macro-region-out,
65		−-0.6326218199475553,9.0,B5 D+K,loo_North_Africa,78.0,leave-one-macro-region-out,
66		−-0.6783564705568839,2.0,B5 D+K,loo_North_America,85.0,leave-one-macro-region-out,
67		−-0.5282965208177354,3.0,B5 D+K,loo_South_America,84.0,leave-one-macro-region-out,
68		−-0.5560823821545652,5.0,B5 D+K,loo_South_Asia,82.0,leave-one-macro-region-out,
69		−-0.5369885748781366,6.0,B5 D+K,loo_Southeast_Asia_Oceania,81.0,leave-one-macro-region-out,
70		−-0.6030949216167942,17.0,B5 D+K,loo_Southwest_Asia,70.0,leave-one-macro-region-out,
71		−-0.5744541520428678,4.0,B5 D+K,loo_Sub_Saharan_Africa,83.0,leave-one-macro-region-out,
72		−-0.6073319277431232,18.0,A2 P,rep0_fold0,69.0,10x5-fold,
73		−,,A2 P,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
74		−that was "
75		−-0.5430608999692377,17.0,A2 P,rep0_fold2,70.0,10x5-fold,
76		−-0.6066033552073161,17.0,A2 P,rep0_fold3,70.0,10x5-fold,
77		−-0.6305116194410425,17.0,A2 P,rep0_fold4,70.0,10x5-fold,
78		−-0.6195177999366908,18.0,A2 P,rep1_fold0,69.0,10x5-fold,
79		−-0.5928356372674103,18.0,A2 P,rep1_fold1,69.0,10x5-fold,
80		−-0.5755512951369498,17.0,A2 P,rep1_fold2,70.0,10x5-fold,
81		−-0.622916299479458,17.0,A2 P,rep1_fold3,70.0,10x5-fold,
82		−-0.5852119207254092,17.0,A2 P,rep1_fold4,70.0,10x5-fold,
83		−-0.616472621816557,18.0,A2 P,rep2_fold0,69.0,10x5-fold,
84		−-0.5482053008327218,18.0,A2 P,rep2_fold1,69.0,10x5-fold,
85		−-0.6821407631459,17.0,A2 P,rep2_fold2,70.0,10x5-fold,
86		−-0.6293622469961574,17.0,A2 P,rep2_fold3,70.0,10x5-fold,
87		−-0.545310589513379,17.0,A2 P,rep2_fold4,70.0,10x5-fold,
88		−-0.5540445983633765,18.0,A2 P,rep3_fold0,69.0,10x5-fold,
89		−-0.6415004544963518,18.0,A2 P,rep3_fold1,69.0,10x5-fold,
90		−-0.6258069537622215,17.0,A2 P,rep3_fold2,70.0,10x5-fold,
91		−-0.6058292956883152,17.0,A2 P,rep3_fold3,70.0,10x5-fold,
92		−-0.5859419890086269,17.0,A2 P,rep3_fold4,70.0,10x5-fold,
93		−-0.6170286166705549,18.0,A2 P,rep4_fold0,69.0,10x5-fold,
94		−-0.6898464795595827,18.0,A2 P,rep4_fold1,69.0,10x5-fold,
95		−-0.5876564900362858,17.0,A2 P,rep4_fold2,70.0,10x5-fold,
96		−-0.599407944209624,17.0,A2 P,rep4_fold3,70.0,10x5-fold,
97		−-0.5554073836595318,17.0,A2 P,rep4_fold4,70.0,10x5-fold,
98		−-0.5754837141305577,18.0,A2 P,rep5_fold0,69.0,10x5-fold,
99		−-0.6366553669723513,18.0,A2 P,rep5_fold1,69.0,10x5-fold,
100		−-0.626181965497974,17.0,A2 P,rep5_fold2,70.0,10x5-fold,
101		−-0.5724568522125653,17.0,A2 P,rep5_fold3,70.0,10x5-fold,
102		−-0.5958887681386783,17.0,A2 P,rep5_fold4,70.0,10x5-fold,
103		−-0.6467235094755063,18.0,A2 P,rep6_fold0,69.0,10x5-fold,
104		−-0.5530973471164247,18.0,A2 P,rep6_fold1,69.0,10x5-fold,
105		−-0.5729226848601316,17.0,A2 P,rep6_fold2,70.0,10x5-fold,
106		−-0.6268760451697202,17.0,A2 P,rep6_fold3,70.0,10x5-fold,
107		−-0.6215745903739501,17.0,A2 P,rep6_fold4,70.0,10x5-fold,
108		−-0.6460179014798269,18.0,A2 P,rep7_fold0,69.0,10x5-fold,
109		−-0.615112194855756,18.0,A2 P,rep7_fold1,69.0,10x5-fold,
110		−-0.5649084327318514,17.0,A2 P,rep7_fold2,70.0,10x5-fold,
111		−-0.646666369436335,17.0,A2 P,rep7_fold3,70.0,10x5-fold,
112		−-0.5661568215832,17.0,A2 P,rep7_fold4,70.0,10x5-fold,
113		−-0.5822039715456159,18.0,A2 P,rep8_fold0,69.0,10x5-fold,
114		−-0.6049258779799229,18.0,A2 P,rep8_fold1,69.0,10x5-fold,
115		−-0.6104916839226251,17.0,A2 P,rep8_fold2,70.0,10x5-fold,
116		−-0.620428349688623,17.0,A2 P,rep8_fold3,70.0,10x5-fold,
117		−-0.5804883865855972,17.0,A2 P,rep8_fold4,70.0,10x5-fold,
118		−-0.6800275101477173,18.0,A2 P,rep9_fold0,69.0,10x5-fold,
119		−,,A2 P,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
120		−that was "
121		−-0.6098519300366266,17.0,A2 P,rep9_fold2,70.0,10x5-fold,
122		−-0.5925777083749647,17.0,A2 P,rep9_fold3,70.0,10x5-fold,
123		−-0.5587977262042823,17.0,A2 P,rep9_fold4,70.0,10x5-fold,
124		−-0.6222589966701738,18.0,B0 FE only,rep0_fold0,69.0,10x5-fold,
125		−,,B0 FE only,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
126		−that was "
127		−-0.5484545915943224,17.0,B0 FE only,rep0_fold2,70.0,10x5-fold,
128		−-0.6141295916001708,17.0,B0 FE only,rep0_fold3,70.0,10x5-fold,
129		−-0.6431128539272781,17.0,B0 FE only,rep0_fold4,70.0,10x5-fold,
130		−-0.6280493673157582,18.0,B0 FE only,rep1_fold0,69.0,10x5-fold,
131		−-0.597782103219764,18.0,B0 FE only,rep1_fold1,69.0,10x5-fold,
132		−-0.5771116387983672,17.0,B0 FE only,rep1_fold2,70.0,10x5-fold,
133		−-0.6248360038428907,17.0,B0 FE only,rep1_fold3,70.0,10x5-fold,
134		−-0.5924549688245182,17.0,B0 FE only,rep1_fold4,70.0,10x5-fold,
135		−-0.6224244199300977,18.0,B0 FE only,rep2_fold0,69.0,10x5-fold,
136		−-0.5462603369963266,18.0,B0 FE only,rep2_fold1,69.0,10x5-fold,
137		−-0.6913028440869243,17.0,B0 FE only,rep2_fold2,70.0,10x5-fold,
138		−-0.6191640061924059,17.0,B0 FE only,rep2_fold3,70.0,10x5-fold,
139		−-0.5487987002674055,17.0,B0 FE only,rep2_fold4,70.0,10x5-fold,
140		−-0.554652762577848,18.0,B0 FE only,rep3_fold0,69.0,10x5-fold,
141		−-0.642908611052659,18.0,B0 FE only,rep3_fold1,69.0,10x5-fold,
142		−-0.6369788488977463,17.0,B0 FE only,rep3_fold2,70.0,10x5-fold,
143		−-0.6148872113525065,17.0,B0 FE only,rep3_fold3,70.0,10x5-fold,
144		−-0.583431865864677,17.0,B0 FE only,rep3_fold4,70.0,10x5-fold,
145		−-0.6259557612659932,18.0,B0 FE only,rep4_fold0,69.0,10x5-fold,
146		−-0.7014602714988226,18.0,B0 FE only,rep4_fold1,69.0,10x5-fold,
147		−-0.5912995056749891,17.0,B0 FE only,rep4_fold2,70.0,10x5-fold,
148		−-0.5886871480168484,17.0,B0 FE only,rep4_fold3,70.0,10x5-fold,
149		−-0.5545150648412948,17.0,B0 FE only,rep4_fold4,70.0,10x5-fold,
150		−-0.5791119016815003,18.0,B0 FE only,rep5_fold0,69.0,10x5-fold,
151		−-0.6463348653643614,18.0,B0 FE only,rep5_fold1,69.0,10x5-fold,
152		−-0.631423050333644,17.0,B0 FE only,rep5_fold2,70.0,10x5-fold,
153		−-0.5803390702036926,17.0,B0 FE only,rep5_fold3,70.0,10x5-fold,
154		−-0.5893572684192436,17.0,B0 FE only,rep5_fold4,70.0,10x5-fold,
155		−-0.6471485289801452,18.0,B0 FE only,rep6_fold0,69.0,10x5-fold,
156		−-0.5553637502851452,18.0,B0 FE only,rep6_fold1,69.0,10x5-fold,
157		−-0.5636147183157447,17.0,B0 FE only,rep6_fold2,70.0,10x5-fold,
158		−-0.63411451084136,17.0,B0 FE only,rep6_fold3,70.0,10x5-fold,
159		−-0.6319157472426794,17.0,B0 FE only,rep6_fold4,70.0,10x5-fold,
160		−-0.6448266855298942,18.0,B0 FE only,rep7_fold0,69.0,10x5-fold,
161		−-0.6208906825838825,18.0,B0 FE only,rep7_fold1,69.0,10x5-fold,
162		−-0.5755605616448807,17.0,B0 FE only,rep7_fold2,70.0,10x5-fold,
163		−-0.6497659257041098,17.0,B0 FE only,rep7_fold3,70.0,10x5-fold,
164		−-0.5521465560221865,17.0,B0 FE only,rep7_fold4,70.0,10x5-fold,
165		−-0.583620691159712,18.0,B0 FE only,rep8_fold0,69.0,10x5-fold,
166		−-0.6141841607797137,18.0,B0 FE only,rep8_fold1,69.0,10x5-fold,
167		−-0.6167588901383592,17.0,B0 FE only,rep8_fold2,70.0,10x5-fold,
168		−-0.6247229072680537,17.0,B0 FE only,rep8_fold3,70.0,10x5-fold,
169		−-0.5837698591057986,17.0,B0 FE only,rep8_fold4,70.0,10x5-fold,
170		−-0.6655415522114628,18.0,B0 FE only,rep9_fold0,69.0,10x5-fold,
171		−,,B0 FE only,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
172		−that was "
173		−-0.616058766494441,17.0,B0 FE only,rep9_fold2,70.0,10x5-fold,
174		−-0.598973678147369,17.0,B0 FE only,rep9_fold3,70.0,10x5-fold,
175		−-0.5685000808762086,17.0,B0 FE only,rep9_fold4,70.0,10x5-fold,
176		−-0.6083195876382478,18.0,B1 D only,rep0_fold0,69.0,10x5-fold,
177		−,,B1 D only,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
178		−that was "
179		−-0.546891347396433,17.0,B1 D only,rep0_fold2,70.0,10x5-fold,
180		−-0.6073616258366199,17.0,B1 D only,rep0_fold3,70.0,10x5-fold,
181		−-0.6312694824979594,17.0,B1 D only,rep0_fold4,70.0,10x5-fold,
182		−-0.6223838629775634,18.0,B1 D only,rep1_fold0,69.0,10x5-fold,
183		−-0.5947180870121573,18.0,B1 D only,rep1_fold1,69.0,10x5-fold,
184		−-0.5822865142697975,17.0,B1 D only,rep1_fold2,70.0,10x5-fold,
185		−-0.6199227847693539,17.0,B1 D only,rep1_fold3,70.0,10x5-fold,
186		−-0.5859938247448637,17.0,B1 D only,rep1_fold4,70.0,10x5-fold,
187		−-0.6175189764440715,18.0,B1 D only,rep2_fold0,69.0,10x5-fold,
188		−-0.5569485312822358,18.0,B1 D only,rep2_fold1,69.0,10x5-fold,
189		−-0.6841572629494984,17.0,B1 D only,rep2_fold2,70.0,10x5-fold,
190		−-0.6200529874225703,17.0,B1 D only,rep2_fold3,70.0,10x5-fold,
191		−-0.546098627091856,17.0,B1 D only,rep2_fold4,70.0,10x5-fold,
192		−-0.5660339371317407,18.0,B1 D only,rep3_fold0,69.0,10x5-fold,
193		−-0.640706101223895,18.0,B1 D only,rep3_fold1,69.0,10x5-fold,
194		−-0.628772500289047,17.0,B1 D only,rep3_fold2,70.0,10x5-fold,
195		−-0.6089885321578207,17.0,B1 D only,rep3_fold3,70.0,10x5-fold,
196		−-0.589291658590394,17.0,B1 D only,rep3_fold4,70.0,10x5-fold,
197		−-0.616508747350852,18.0,B1 D only,rep4_fold0,69.0,10x5-fold,
198		−-0.6943559160937426,18.0,B1 D only,rep4_fold1,69.0,10x5-fold,
199		−-0.5866364532840139,17.0,B1 D only,rep4_fold2,70.0,10x5-fold,
200		−-0.6021887502841344,17.0,B1 D only,rep4_fold3,70.0,10x5-fold,
201		−-0.5618567958922737,17.0,B1 D only,rep4_fold4,70.0,10x5-fold,
202		−-0.5731895159730104,18.0,B1 D only,rep5_fold0,69.0,10x5-fold,
203		−-0.639150680801495,18.0,B1 D only,rep5_fold1,69.0,10x5-fold,
204		−-0.6295445658391886,17.0,B1 D only,rep5_fold2,70.0,10x5-fold,
205		−-0.573692260065334,17.0,B1 D only,rep5_fold3,70.0,10x5-fold,
206		−-0.5995650677337133,17.0,B1 D only,rep5_fold4,70.0,10x5-fold,
207		−-0.6518349304983629,18.0,B1 D only,rep6_fold0,69.0,10x5-fold,
208		−-0.5531610824735905,18.0,B1 D only,rep6_fold1,69.0,10x5-fold,
209		−-0.5763361704012375,17.0,B1 D only,rep6_fold2,70.0,10x5-fold,
210		−-0.6282372771818667,17.0,B1 D only,rep6_fold3,70.0,10x5-fold,
211		−-0.6248206977725752,17.0,B1 D only,rep6_fold4,70.0,10x5-fold,
212		−-0.653125147254194,18.0,B1 D only,rep7_fold0,69.0,10x5-fold,
213		−-0.6143976840991052,18.0,B1 D only,rep7_fold1,69.0,10x5-fold,
214		−-0.5674316594988658,17.0,B1 D only,rep7_fold2,70.0,10x5-fold,
215		−-0.644496996076469,17.0,B1 D only,rep7_fold3,70.0,10x5-fold,
216		−-0.5635400325456037,17.0,B1 D only,rep7_fold4,70.0,10x5-fold,
217		−-0.584929487443215,18.0,B1 D only,rep8_fold0,69.0,10x5-fold,
218		−-0.6051446334858768,18.0,B1 D only,rep8_fold1,69.0,10x5-fold,
219		−-0.6116292761407962,17.0,B1 D only,rep8_fold2,70.0,10x5-fold,
220		−-0.6245369876486837,17.0,B1 D only,rep8_fold3,70.0,10x5-fold,
221		−-0.5802696103002862,17.0,B1 D only,rep8_fold4,70.0,10x5-fold,
222		−-0.6769562064244521,18.0,B1 D only,rep9_fold0,69.0,10x5-fold,
223		−,,B1 D only,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
224		−that was "
225		−-0.6094913290463297,17.0,B1 D only,rep9_fold2,70.0,10x5-fold,
226		−-0.5936887920747554,17.0,B1 D only,rep9_fold3,70.0,10x5-fold,
227		−-0.5608830070370011,17.0,B1 D only,rep9_fold4,70.0,10x5-fold,
228		−-0.622860733833226,18.0,B2 K only,rep0_fold0,69.0,10x5-fold,
229		−,,B2 K only,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
230		−that was "
231		−-0.5481025853813067,17.0,B2 K only,rep0_fold2,70.0,10x5-fold,
232		−-0.6157795858450074,17.0,B2 K only,rep0_fold3,70.0,10x5-fold,
233		−-0.647858812398134,17.0,B2 K only,rep0_fold4,70.0,10x5-fold,
234		−-0.6288038612470842,18.0,B2 K only,rep1_fold0,69.0,10x5-fold,
235		−-0.5975185326794321,18.0,B2 K only,rep1_fold1,69.0,10x5-fold,
236		−-0.5785935586562205,17.0,B2 K only,rep1_fold2,70.0,10x5-fold,
237		−-0.6265308235739577,17.0,B2 K only,rep1_fold3,70.0,10x5-fold,
238		−-0.6049837734565248,17.0,B2 K only,rep1_fold4,70.0,10x5-fold,
239		−-0.622723644715846,18.0,B2 K only,rep2_fold0,69.0,10x5-fold,
240		−-0.5479976710611313,18.0,B2 K only,rep2_fold1,69.0,10x5-fold,
241		−-0.691017174609867,17.0,B2 K only,rep2_fold2,70.0,10x5-fold,
242		−-0.6297025936236185,17.0,B2 K only,rep2_fold3,70.0,10x5-fold,
243		−-0.5489775346820496,17.0,B2 K only,rep2_fold4,70.0,10x5-fold,
244		−-0.5569141957997442,18.0,B2 K only,rep3_fold0,69.0,10x5-fold,
245		−-0.6438185192695789,18.0,B2 K only,rep3_fold1,69.0,10x5-fold,
246		−-0.6381378863600627,17.0,B2 K only,rep3_fold2,70.0,10x5-fold,
247		−-0.6142228160425109,17.0,B2 K only,rep3_fold3,70.0,10x5-fold,
248		−-0.5830679679528096,17.0,B2 K only,rep3_fold4,70.0,10x5-fold,
249		−-0.6272397542059387,18.0,B2 K only,rep4_fold0,69.0,10x5-fold,
250		−-0.7005034673848259,18.0,B2 K only,rep4_fold1,69.0,10x5-fold,
251		−-0.594376135777104,17.0,B2 K only,rep4_fold2,70.0,10x5-fold,
252		−-0.5882661065530312,17.0,B2 K only,rep4_fold3,70.0,10x5-fold,
253		−-0.5552266659875161,17.0,B2 K only,rep4_fold4,70.0,10x5-fold,
254		−-0.5833162558472252,18.0,B2 K only,rep5_fold0,69.0,10x5-fold,
255		−-0.6458843675834318,18.0,B2 K only,rep5_fold1,69.0,10x5-fold,
256		−-0.6341119232147508,17.0,B2 K only,rep5_fold2,70.0,10x5-fold,
257		−-0.5801197123671539,17.0,B2 K only,rep5_fold3,70.0,10x5-fold,
258		−-0.5917407814937196,17.0,B2 K only,rep5_fold4,70.0,10x5-fold,
259		−-0.6478841088759187,18.0,B2 K only,rep6_fold0,69.0,10x5-fold,
260		−-0.5562699230921433,18.0,B2 K only,rep6_fold1,69.0,10x5-fold,
261		−-0.5642110792414579,17.0,B2 K only,rep6_fold2,70.0,10x5-fold,
262		−-0.6339541022109576,17.0,B2 K only,rep6_fold3,70.0,10x5-fold,
263		−-0.631241916779392,17.0,B2 K only,rep6_fold4,70.0,10x5-fold,
264		−-0.6457897459223011,18.0,B2 K only,rep7_fold0,69.0,10x5-fold,
265		−-0.6241780267664523,18.0,B2 K only,rep7_fold1,69.0,10x5-fold,
266		−-0.577715100514324,17.0,B2 K only,rep7_fold2,70.0,10x5-fold,
267		−-0.6570781933037084,17.0,B2 K only,rep7_fold3,70.0,10x5-fold,
268		−-0.5521118033697582,17.0,B2 K only,rep7_fold4,70.0,10x5-fold,
269		−-0.5828381767659178,18.0,B2 K only,rep8_fold0,69.0,10x5-fold,
270		−-0.6144580157235265,18.0,B2 K only,rep8_fold1,69.0,10x5-fold,
271		−-0.616659994437666,17.0,B2 K only,rep8_fold2,70.0,10x5-fold,
272		−-0.624088792321915,17.0,B2 K only,rep8_fold3,70.0,10x5-fold,
273		−-0.5844439826069375,17.0,B2 K only,rep8_fold4,70.0,10x5-fold,
274		−-0.6663920964067712,18.0,B2 K only,rep9_fold0,69.0,10x5-fold,
275		−,,B2 K only,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
276		−that was "
277		−-0.61663352215543,17.0,B2 K only,rep9_fold2,70.0,10x5-fold,
278		−-0.5989841600894605,17.0,B2 K only,rep9_fold3,70.0,10x5-fold,
279		−-0.5687535233371137,17.0,B2 K only,rep9_fold4,70.0,10x5-fold,
280		−-0.6224232632942309,18.0,B3 complexity,rep0_fold0,69.0,10x5-fold,
281		−,,B3 complexity,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
282		−that was "
283		−-0.548021220605251,17.0,B3 complexity,rep0_fold2,70.0,10x5-fold,
284		−-0.6185799277109469,17.0,B3 complexity,rep0_fold3,70.0,10x5-fold,
285		−-0.6551767260281819,17.0,B3 complexity,rep0_fold4,70.0,10x5-fold,
286		−-0.6280157326305398,18.0,B3 complexity,rep1_fold0,69.0,10x5-fold,
287		−-0.5974127318195985,18.0,B3 complexity,rep1_fold1,69.0,10x5-fold,
288		−-0.5779100101966184,17.0,B3 complexity,rep1_fold2,70.0,10x5-fold,
289		−-0.6265719380201763,17.0,B3 complexity,rep1_fold3,70.0,10x5-fold,
290		−-0.5998911050375249,17.0,B3 complexity,rep1_fold4,70.0,10x5-fold,
291		−-0.6229682345999569,18.0,B3 complexity,rep2_fold0,69.0,10x5-fold,
292		−-0.5527991529682255,18.0,B3 complexity,rep2_fold1,69.0,10x5-fold,
293		−-0.6918027497298345,17.0,B3 complexity,rep2_fold2,70.0,10x5-fold,
294		−-0.6243495498300767,17.0,B3 complexity,rep2_fold3,70.0,10x5-fold,
295		−-0.5486012589225445,17.0,B3 complexity,rep2_fold4,70.0,10x5-fold,
296		−-0.554348329200956,18.0,B3 complexity,rep3_fold0,69.0,10x5-fold,
297		−-0.643857038455794,18.0,B3 complexity,rep3_fold1,69.0,10x5-fold,
298		−-0.640216594325675,17.0,B3 complexity,rep3_fold2,70.0,10x5-fold,
299		−-0.6139446870181516,17.0,B3 complexity,rep3_fold3,70.0,10x5-fold,
300		−-0.5838240226168955,17.0,B3 complexity,rep3_fold4,70.0,10x5-fold,
301		−-0.6276158169789166,18.0,B3 complexity,rep4_fold0,69.0,10x5-fold,
302		−-0.7025064012583728,18.0,B3 complexity,rep4_fold1,69.0,10x5-fold,
303		−-0.5968080743238756,17.0,B3 complexity,rep4_fold2,70.0,10x5-fold,
304		−-0.5878815702260025,17.0,B3 complexity,rep4_fold3,70.0,10x5-fold,
305		−-0.555541049733102,17.0,B3 complexity,rep4_fold4,70.0,10x5-fold,
306		−-0.5825960687526403,18.0,B3 complexity,rep5_fold0,69.0,10x5-fold,
307		−-0.6458675603618614,18.0,B3 complexity,rep5_fold1,69.0,10x5-fold,
308		−-0.6318087921744449,17.0,B3 complexity,rep5_fold2,70.0,10x5-fold,
309		−-0.5805765904798559,17.0,B3 complexity,rep5_fold3,70.0,10x5-fold,
310		−-0.5898164549103526,17.0,B3 complexity,rep5_fold4,70.0,10x5-fold,
311		−-0.6488127534634986,18.0,B3 complexity,rep6_fold0,69.0,10x5-fold,
312		−-0.5598498827806732,18.0,B3 complexity,rep6_fold1,69.0,10x5-fold,
313		−-0.5631650509389547,17.0,B3 complexity,rep6_fold2,70.0,10x5-fold,
314		−-0.6340273844377895,17.0,B3 complexity,rep6_fold3,70.0,10x5-fold,
315		−-0.6344343111184253,17.0,B3 complexity,rep6_fold4,70.0,10x5-fold,
316		−-0.6495896866751387,18.0,B3 complexity,rep7_fold0,69.0,10x5-fold,
317		−-0.6201545780087081,18.0,B3 complexity,rep7_fold1,69.0,10x5-fold,
318		−-0.5766895733818591,17.0,B3 complexity,rep7_fold2,70.0,10x5-fold,
319		−-0.6526483507711919,17.0,B3 complexity,rep7_fold3,70.0,10x5-fold,
320		−-0.5520706492340092,17.0,B3 complexity,rep7_fold4,70.0,10x5-fold,
321		−-0.5837108392288399,18.0,B3 complexity,rep8_fold0,69.0,10x5-fold,
322		−-0.6148287565836799,18.0,B3 complexity,rep8_fold1,69.0,10x5-fold,
323		−-0.6173938320749792,17.0,B3 complexity,rep8_fold2,70.0,10x5-fold,
324		−-0.6243828632704121,17.0,B3 complexity,rep8_fold3,70.0,10x5-fold,
325		−-0.58370651697469,17.0,B3 complexity,rep8_fold4,70.0,10x5-fold,
326		−-0.6658135394934944,18.0,B3 complexity,rep9_fold0,69.0,10x5-fold,
327		−,,B3 complexity,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
328		−that was "
329		−-0.6160048014502761,17.0,B3 complexity,rep9_fold2,70.0,10x5-fold,
330		−-0.5994598035926632,17.0,B3 complexity,rep9_fold3,70.0,10x5-fold,
331		−-0.568713984582936,17.0,B3 complexity,rep9_fold4,70.0,10x5-fold,
332		−-0.6177748794828175,18.0,B4 persistence,rep0_fold0,69.0,10x5-fold,
333		−,,B4 persistence,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
334		−that was "
335		−-0.541350208035929,17.0,B4 persistence,rep0_fold2,70.0,10x5-fold,
336		−-0.6150391065571179,17.0,B4 persistence,rep0_fold3,70.0,10x5-fold,
337		−-0.6399410433722716,17.0,B4 persistence,rep0_fold4,70.0,10x5-fold,
338		−-0.6295697257883499,18.0,B4 persistence,rep1_fold0,69.0,10x5-fold,
339		−-0.5880278784748856,18.0,B4 persistence,rep1_fold1,69.0,10x5-fold,
340		−-0.5731604710799205,17.0,B4 persistence,rep1_fold2,70.0,10x5-fold,
341		−-0.6375013456439419,17.0,B4 persistence,rep1_fold3,70.0,10x5-fold,
342		−-0.5958581411636649,17.0,B4 persistence,rep1_fold4,70.0,10x5-fold,
343		−-0.6190718851371307,18.0,B4 persistence,rep2_fold0,69.0,10x5-fold,
344		−-0.5482486031623248,18.0,B4 persistence,rep2_fold1,69.0,10x5-fold,
345		−-0.6889472659786482,17.0,B4 persistence,rep2_fold2,70.0,10x5-fold,
346		−-0.611941560462283,17.0,B4 persistence,rep2_fold3,70.0,10x5-fold,
347		−-0.5472265352846163,17.0,B4 persistence,rep2_fold4,70.0,10x5-fold,
348		−-0.5573419404454799,18.0,B4 persistence,rep3_fold0,69.0,10x5-fold,
349		−-0.6380646732677969,18.0,B4 persistence,rep3_fold1,69.0,10x5-fold,
350		−-0.6307203306424725,17.0,B4 persistence,rep3_fold2,70.0,10x5-fold,
351		−-0.6097671103332879,17.0,B4 persistence,rep3_fold3,70.0,10x5-fold,
352		−-0.58655920179526,17.0,B4 persistence,rep3_fold4,70.0,10x5-fold,
353		−-0.6211534067439222,18.0,B4 persistence,rep4_fold0,69.0,10x5-fold,
354		−-0.6961452495181134,18.0,B4 persistence,rep4_fold1,69.0,10x5-fold,
355		−-0.5881216246876273,17.0,B4 persistence,rep4_fold2,70.0,10x5-fold,
356		−-0.5876368465571769,17.0,B4 persistence,rep4_fold3,70.0,10x5-fold,
357		−-0.5550327501500281,17.0,B4 persistence,rep4_fold4,70.0,10x5-fold,
358		−-0.5826526469530944,18.0,B4 persistence,rep5_fold0,69.0,10x5-fold,
359		−-0.6371101593952542,18.0,B4 persistence,rep5_fold1,69.0,10x5-fold,
360		−-0.6300000134772511,17.0,B4 persistence,rep5_fold2,70.0,10x5-fold,
361		−-0.5762926736158204,17.0,B4 persistence,rep5_fold3,70.0,10x5-fold,
362		−-0.5985421162868496,17.0,B4 persistence,rep5_fold4,70.0,10x5-fold,
363		−-0.652928695414056,18.0,B4 persistence,rep6_fold0,69.0,10x5-fold,
364		−-0.5551680870281747,18.0,B4 persistence,rep6_fold1,69.0,10x5-fold,
365		−-0.5643743883025886,17.0,B4 persistence,rep6_fold2,70.0,10x5-fold,
366		−-0.6273228665735541,17.0,B4 persistence,rep6_fold3,70.0,10x5-fold,
367		−-0.6269801961378364,17.0,B4 persistence,rep6_fold4,70.0,10x5-fold,
368		−-0.6397452196954947,18.0,B4 persistence,rep7_fold0,69.0,10x5-fold,
369		−-0.6162470649154671,18.0,B4 persistence,rep7_fold1,69.0,10x5-fold,
370		−-0.5647336910797054,17.0,B4 persistence,rep7_fold2,70.0,10x5-fold,
371		−-0.6445988346816625,17.0,B4 persistence,rep7_fold3,70.0,10x5-fold,
372		−-0.5518545474497174,17.0,B4 persistence,rep7_fold4,70.0,10x5-fold,
373		−-0.5816639710614425,18.0,B4 persistence,rep8_fold0,69.0,10x5-fold,
374		−-0.6104026408192657,18.0,B4 persistence,rep8_fold1,69.0,10x5-fold,
375		−-0.6230006737451382,17.0,B4 persistence,rep8_fold2,70.0,10x5-fold,
376		−-0.6211060770037978,17.0,B4 persistence,rep8_fold3,70.0,10x5-fold,
377		−-0.5788777122446616,17.0,B4 persistence,rep8_fold4,70.0,10x5-fold,
378		−-0.6798072722309922,18.0,B4 persistence,rep9_fold0,69.0,10x5-fold,
379		−,,B4 persistence,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
380		−that was "
381		−-0.6161884466742826,17.0,B4 persistence,rep9_fold2,70.0,10x5-fold,
382		−-0.5979456758715648,17.0,B4 persistence,rep9_fold3,70.0,10x5-fold,
383		−-0.5649801809335007,17.0,B4 persistence,rep9_fold4,70.0,10x5-fold,
384		−-0.6071201113987283,18.0,B5 D+K,rep0_fold0,69.0,10x5-fold,
385		−,,B5 D+K,rep0_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
386		−that was "
387		−-0.5448226199275119,17.0,B5 D+K,rep0_fold2,70.0,10x5-fold,
388		−-0.6078459568722003,17.0,B5 D+K,rep0_fold3,70.0,10x5-fold,
389		−-0.6342417651615544,17.0,B5 D+K,rep0_fold4,70.0,10x5-fold,
390		−-0.6216611883548933,18.0,B5 D+K,rep1_fold0,69.0,10x5-fold,
391		−-0.5933436332383474,18.0,B5 D+K,rep1_fold1,69.0,10x5-fold,
392		−-0.5799203552322584,17.0,B5 D+K,rep1_fold2,70.0,10x5-fold,
393		−-0.6237774969169441,17.0,B5 D+K,rep1_fold3,70.0,10x5-fold,
394		−-0.5962314074144938,17.0,B5 D+K,rep1_fold4,70.0,10x5-fold,
395		−-0.6164285456981456,18.0,B5 D+K,rep2_fold0,69.0,10x5-fold,
396		−-0.5550719506380004,18.0,B5 D+K,rep2_fold1,69.0,10x5-fold,
397		−-0.6823529488772692,17.0,B5 D+K,rep2_fold2,70.0,10x5-fold,
398		−-0.6349211354758763,17.0,B5 D+K,rep2_fold3,70.0,10x5-fold,
399		−-0.5459307682665984,17.0,B5 D+K,rep2_fold4,70.0,10x5-fold,
400		−-0.5642082627660487,18.0,B5 D+K,rep3_fold0,69.0,10x5-fold,
401		−-0.6413358094439828,18.0,B5 D+K,rep3_fold1,69.0,10x5-fold,
402		−-0.6273715788317731,17.0,B5 D+K,rep3_fold2,70.0,10x5-fold,
403		−-0.606168563492821,17.0,B5 D+K,rep3_fold3,70.0,10x5-fold,
404		−-0.5873488241907752,17.0,B5 D+K,rep3_fold4,70.0,10x5-fold,
405		−-0.6167635409415378,18.0,B5 D+K,rep4_fold0,69.0,10x5-fold,
406		−-0.6902898041346739,18.0,B5 D+K,rep4_fold1,69.0,10x5-fold,
407		−-0.590063802353847,17.0,B5 D+K,rep4_fold2,70.0,10x5-fold,
408		−-0.60149969130101,17.0,B5 D+K,rep4_fold3,70.0,10x5-fold,
409		−-0.5610610105956596,17.0,B5 D+K,rep4_fold4,70.0,10x5-fold,
410		−-0.5763076114426743,18.0,B5 D+K,rep5_fold0,69.0,10x5-fold,
411		−-0.6367410766013497,18.0,B5 D+K,rep5_fold1,69.0,10x5-fold,
412		−-0.6283237081659587,17.0,B5 D+K,rep5_fold2,70.0,10x5-fold,
413		−-0.5723113512237697,17.0,B5 D+K,rep5_fold3,70.0,10x5-fold,
414		−-0.5974104639599264,17.0,B5 D+K,rep5_fold4,70.0,10x5-fold,
415		−-0.650233537823597,18.0,B5 D+K,rep6_fold0,69.0,10x5-fold,
416		−-0.5530376972999185,18.0,B5 D+K,rep6_fold1,69.0,10x5-fold,
417		−-0.5745115580682859,17.0,B5 D+K,rep6_fold2,70.0,10x5-fold,
418		−-0.6266934184819944,17.0,B5 D+K,rep6_fold3,70.0,10x5-fold,
419		−-0.6216739607716862,17.0,B5 D+K,rep6_fold4,70.0,10x5-fold,
420		−-0.6517790474103996,18.0,B5 D+K,rep7_fold0,69.0,10x5-fold,
421		−-0.6169432515040979,18.0,B5 D+K,rep7_fold1,69.0,10x5-fold,
422		−-0.5671613279484712,17.0,B5 D+K,rep7_fold2,70.0,10x5-fold,
423		−-0.6516063601653364,17.0,B5 D+K,rep7_fold3,70.0,10x5-fold,
424		−-0.5669222689586157,17.0,B5 D+K,rep7_fold4,70.0,10x5-fold,
425		−-0.5826281169172589,18.0,B5 D+K,rep8_fold0,69.0,10x5-fold,
426		−-0.6043655684533858,18.0,B5 D+K,rep8_fold1,69.0,10x5-fold,
427		−-0.610361676935831,17.0,B5 D+K,rep8_fold2,70.0,10x5-fold,
428		−-0.6213289274365054,17.0,B5 D+K,rep8_fold3,70.0,10x5-fold,
429		−-0.580382971518139,17.0,B5 D+K,rep8_fold4,70.0,10x5-fold,
430		−-0.6811250689340757,18.0,B5 D+K,rep9_fold0,69.0,10x5-fold,
431		−,,B5 D+K,rep9_fold1,,10x5-fold,"predict requires that you use a DataFrame when predicting from a model
432		−that was "
433		−-0.6095615301133424,17.0,B5 D+K,rep9_fold2,70.0,10x5-fold,
434		−-0.5925029199463963,17.0,B5 D+K,rep9_fold3,70.0,10x5-fold,
435		−-0.5588775589829003,17.0,B5 D+K,rep9_fold4,70.0,10x5-fold,
	1	+oos_loglik,n,model,fold,n_train,scheme
	2	+-0.4863985102023952,4,A2 P,loo_Central_Asia,89,leave-one-macro-region-out
	3	+-0.6166133226174284,23,A2 P,loo_East_Asia,70,leave-one-macro-region-out
	4	+-0.5461394212003365,15,A2 P,loo_Europe,78,leave-one-macro-region-out
	5	+-0.6558986957848013,10,A2 P,loo_North_Africa,83,leave-one-macro-region-out
	6	+-0.6534568063077896,4,A2 P,loo_North_America,89,leave-one-macro-region-out
	7	+-0.5494096956688588,3,A2 P,loo_South_America,90,leave-one-macro-region-out
	8	+-0.5578368952910477,5,A2 P,loo_South_Asia,88,leave-one-macro-region-out
	9	+-0.5306657089980731,7,A2 P,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	10	+-0.5909253081411068,18,A2 P,loo_Southwest_Asia,75,leave-one-macro-region-out
	11	+-0.5658937895584326,4,A2 P,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	12	+-0.4892121743389306,4,B0 FE only,loo_Central_Asia,89,leave-one-macro-region-out
	13	+-0.6261057941338579,23,B0 FE only,loo_East_Asia,70,leave-one-macro-region-out
	14	+-0.5476594155230036,15,B0 FE only,loo_Europe,78,leave-one-macro-region-out
	15	+-0.6591780066673476,10,B0 FE only,loo_North_Africa,83,leave-one-macro-region-out
	16	+-0.6936777863668938,4,B0 FE only,loo_North_America,89,leave-one-macro-region-out
	17	+-0.5531812491187513,3,B0 FE only,loo_South_America,90,leave-one-macro-region-out
	18	+-0.5670621848384819,5,B0 FE only,loo_South_Asia,88,leave-one-macro-region-out
	19	+-0.5776188664076132,7,B0 FE only,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	20	+-0.5950053348615014,18,B0 FE only,loo_Southwest_Asia,75,leave-one-macro-region-out
	21	+-0.605891406695747,4,B0 FE only,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	22	+-0.480412079791239,4,B1 D only,loo_Central_Asia,89,leave-one-macro-region-out
	23	+-0.614546271347027,23,B1 D only,loo_East_Asia,70,leave-one-macro-region-out
	24	+-0.5556266751918056,15,B1 D only,loo_Europe,78,leave-one-macro-region-out
	25	+-0.6414956008595407,10,B1 D only,loo_North_Africa,83,leave-one-macro-region-out
	26	+-0.6530718160929676,4,B1 D only,loo_North_America,89,leave-one-macro-region-out
	27	+-0.5355807122128691,3,B1 D only,loo_South_America,90,leave-one-macro-region-out
	28	+-0.5604334875647103,5,B1 D only,loo_South_Asia,88,leave-one-macro-region-out
	29	+-0.5247171011370838,7,B1 D only,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	30	+-0.5876861017896027,18,B1 D only,loo_Southwest_Asia,75,leave-one-macro-region-out
	31	+-0.5618563407255517,4,B1 D only,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	32	+-0.4883930405983222,4,B2 K only,loo_Central_Asia,89,leave-one-macro-region-out
	33	+-0.6260582499031354,23,B2 K only,loo_East_Asia,70,leave-one-macro-region-out
	34	+-0.5477305776079607,15,B2 K only,loo_Europe,78,leave-one-macro-region-out
	35	+-0.6593836398859183,10,B2 K only,loo_North_Africa,83,leave-one-macro-region-out
	36	+-0.6979308625922402,4,B2 K only,loo_North_America,89,leave-one-macro-region-out
	37	+-0.5507344882259749,3,B2 K only,loo_South_America,90,leave-one-macro-region-out
	38	+-0.5675383997887671,5,B2 K only,loo_South_Asia,88,leave-one-macro-region-out
	39	+-0.580212907088958,7,B2 K only,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	40	+-0.5943948394992884,18,B2 K only,loo_Southwest_Asia,75,leave-one-macro-region-out
	41	+-0.6079524740182257,4,B2 K only,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	42	+-0.48874736633636673,4,B3 complexity,loo_Central_Asia,89,leave-one-macro-region-out
	43	+-0.6267837142901096,23,B3 complexity,loo_East_Asia,70,leave-one-macro-region-out
	44	+-0.5479344497965566,15,B3 complexity,loo_Europe,78,leave-one-macro-region-out
	45	+-0.6598733532216607,10,B3 complexity,loo_North_Africa,83,leave-one-macro-region-out
	46	+-0.7011544303415017,4,B3 complexity,loo_North_America,89,leave-one-macro-region-out
	47	+-0.5503228717879296,3,B3 complexity,loo_South_America,90,leave-one-macro-region-out
	48	+-0.5657524463574177,5,B3 complexity,loo_South_Asia,88,leave-one-macro-region-out
	49	+-0.5776803560758247,7,B3 complexity,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	50	+-0.5949651774388359,18,B3 complexity,loo_Southwest_Asia,75,leave-one-macro-region-out
	51	+-0.6055831023524705,4,B3 complexity,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	52	+-0.4829544089351404,4,B4 persistence,loo_Central_Asia,89,leave-one-macro-region-out
	53	+-0.6233123063470912,23,B4 persistence,loo_East_Asia,70,leave-one-macro-region-out
	54	+-0.553011295044539,15,B4 persistence,loo_Europe,78,leave-one-macro-region-out
	55	+-0.6530969076833154,10,B4 persistence,loo_North_Africa,83,leave-one-macro-region-out
	56	+-0.6869099507412111,4,B4 persistence,loo_North_America,89,leave-one-macro-region-out
	57	+-0.5451688887985835,3,B4 persistence,loo_South_America,90,leave-one-macro-region-out
	58	+-0.5609693206878914,5,B4 persistence,loo_South_Asia,88,leave-one-macro-region-out
	59	+-0.5639183822948616,7,B4 persistence,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	60	+-0.5953212451624995,18,B4 persistence,loo_Southwest_Asia,75,leave-one-macro-region-out
	61	+-0.598977727789005,4,B4 persistence,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	62	+-0.4812372007023369,4,B5 D+K,loo_Central_Asia,89,leave-one-macro-region-out
	63	+-0.6141995658337153,23,B5 D+K,loo_East_Asia,70,leave-one-macro-region-out
	64	+-0.5560791467344729,15,B5 D+K,loo_Europe,78,leave-one-macro-region-out
	65	+-0.645582962037851,10,B5 D+K,loo_North_Africa,83,leave-one-macro-region-out
	66	+-0.6503472521110671,4,B5 D+K,loo_North_America,89,leave-one-macro-region-out
	67	+-0.5367923719111669,3,B5 D+K,loo_South_America,90,leave-one-macro-region-out
	68	+-0.5592192953717521,5,B5 D+K,loo_South_Asia,88,leave-one-macro-region-out
	69	+-0.522862537365376,7,B5 D+K,loo_Southeast_Asia_Oceania,86,leave-one-macro-region-out
	70	+-0.5880234830903966,18,B5 D+K,loo_Southwest_Asia,75,leave-one-macro-region-out
	71	+-0.5599915463045968,4,B5 D+K,loo_Sub_Saharan_Africa,89,leave-one-macro-region-out
	72	+-0.6378724753868233,19,A2 P,rep0_fold0,74,10x5-fold
	73	+-0.6476438045436628,17,A2 P,rep0_fold1,76,10x5-fold
	74	+-0.6722841867465483,19,A2 P,rep0_fold2,74,10x5-fold
	75	+-0.558168443331959,19,A2 P,rep0_fold3,74,10x5-fold
	76	+-0.5750489107772118,19,A2 P,rep0_fold4,74,10x5-fold
	77	+-0.553072136633415,19,A2 P,rep1_fold0,74,10x5-fold
	78	+-0.5638583811530476,19,A2 P,rep1_fold1,74,10x5-fold
	79	+-0.626310212866657,19,A2 P,rep1_fold2,74,10x5-fold
	80	+-0.5934845970355244,19,A2 P,rep1_fold3,74,10x5-fold
	81	+-0.6436420682883632,17,A2 P,rep1_fold4,76,10x5-fold
	82	+-0.5883219501792204,20,A2 P,rep2_fold0,73,10x5-fold
	83	+-0.6117323103459469,19,A2 P,rep2_fold1,74,10x5-fold
	84	+-0.5719274911142231,18,A2 P,rep2_fold2,75,10x5-fold
	85	+-0.6033485740794418,18,A2 P,rep2_fold3,75,10x5-fold
	86	+-0.6234943648902918,18,A2 P,rep2_fold4,75,10x5-fold
	87	+-0.656109862407513,18,A2 P,rep3_fold0,75,10x5-fold
	88	+-0.557158615911579,18,A2 P,rep3_fold1,75,10x5-fold
	89	+-0.5947381994626586,19,A2 P,rep3_fold2,74,10x5-fold
	90	+-0.6200023994054562,18,A2 P,rep3_fold3,75,10x5-fold
	91	+-0.5803530149726879,20,A2 P,rep3_fold4,73,10x5-fold
	92	+-0.5872443079593692,19,A2 P,rep4_fold0,74,10x5-fold
	93	+-0.5985389342326792,19,A2 P,rep4_fold1,74,10x5-fold
	94	+-0.6113931721361338,18,A2 P,rep4_fold2,75,10x5-fold
	95	+-0.5544720060532792,19,A2 P,rep4_fold3,74,10x5-fold
	96	+-0.6419570238504169,18,A2 P,rep4_fold4,75,10x5-fold
	97	+-0.6207672336595869,18,A2 P,rep5_fold0,75,10x5-fold
	98	+-0.5966481439941131,19,A2 P,rep5_fold1,74,10x5-fold
	99	+-0.6397512260757561,18,A2 P,rep5_fold2,75,10x5-fold
	100	+-0.5266556605025304,20,A2 P,rep5_fold3,73,10x5-fold
	101	+-0.6101371876347584,18,A2 P,rep5_fold4,75,10x5-fold
	102	+-0.5412999530659527,20,A2 P,rep6_fold0,73,10x5-fold
	103	+-0.585538046053729,19,A2 P,rep6_fold1,74,10x5-fold
	104	+-0.5300611673974372,18,A2 P,rep6_fold2,75,10x5-fold
	105	+-0.6132442972303742,18,A2 P,rep6_fold3,75,10x5-fold
	106	+-0.6991703846772523,18,A2 P,rep6_fold4,75,10x5-fold
	107	+-0.6362081546966806,17,A2 P,rep7_fold0,76,10x5-fold
	108	+-0.6959225153371932,19,A2 P,rep7_fold1,74,10x5-fold
	109	+-0.557278415572337,19,A2 P,rep7_fold2,74,10x5-fold
	110	+-0.5383011727515022,20,A2 P,rep7_fold3,73,10x5-fold
	111	+-0.6300579462596528,18,A2 P,rep7_fold4,75,10x5-fold
	112	+-0.6126164864966035,18,A2 P,rep8_fold0,75,10x5-fold
	113	+-0.6776484750924084,18,A2 P,rep8_fold1,75,10x5-fold
	114	+-0.5723861193417171,19,A2 P,rep8_fold2,74,10x5-fold
	115	+-0.5690265979307906,19,A2 P,rep8_fold3,74,10x5-fold
	116	+-0.5773948857031158,19,A2 P,rep8_fold4,74,10x5-fold
	117	+-0.535280925329609,19,A2 P,rep9_fold0,74,10x5-fold
	118	+-0.6055332623146835,19,A2 P,rep9_fold1,74,10x5-fold
	119	+-0.6039418797952539,18,A2 P,rep9_fold2,75,10x5-fold
	120	+-0.6320159147811587,18,A2 P,rep9_fold3,75,10x5-fold
	121	+-0.6203519735463313,19,A2 P,rep9_fold4,74,10x5-fold
	122	+-0.6785113570742274,19,B0 FE only,rep0_fold0,74,10x5-fold
	123	+-0.6627742444905758,17,B0 FE only,rep0_fold1,76,10x5-fold
	124	+-0.6808175851310345,19,B0 FE only,rep0_fold2,74,10x5-fold
	125	+-0.5601268745776017,19,B0 FE only,rep0_fold3,74,10x5-fold
	126	+-0.5859549259034864,19,B0 FE only,rep0_fold4,74,10x5-fold
	127	+-0.5652917767428121,19,B0 FE only,rep1_fold0,74,10x5-fold
	128	+-0.5689952222647577,19,B0 FE only,rep1_fold1,74,10x5-fold
	129	+-0.6402971722088028,19,B0 FE only,rep1_fold2,74,10x5-fold
	130	+-0.5924832697673466,19,B0 FE only,rep1_fold3,74,10x5-fold
	131	+-0.6563047572356738,17,B0 FE only,rep1_fold4,76,10x5-fold
	132	+-0.5946828408558984,20,B0 FE only,rep2_fold0,73,10x5-fold
	133	+-0.6189965169844655,19,B0 FE only,rep2_fold1,74,10x5-fold
	134	+-0.5725319374178143,18,B0 FE only,rep2_fold2,75,10x5-fold
	135	+-0.5910436731162775,18,B0 FE only,rep2_fold3,75,10x5-fold
	136	+-0.6327432028824488,18,B0 FE only,rep2_fold4,75,10x5-fold
	137	+-0.6589490833304797,18,B0 FE only,rep3_fold0,75,10x5-fold
	138	+-0.5640793689927075,18,B0 FE only,rep3_fold1,75,10x5-fold
	139	+-0.6065870671193895,19,B0 FE only,rep3_fold2,74,10x5-fold
	140	+-0.6267661863678218,18,B0 FE only,rep3_fold3,75,10x5-fold
	141	+-0.5905053756198214,20,B0 FE only,rep3_fold4,73,10x5-fold
	142	+-0.5978968779595755,19,B0 FE only,rep4_fold0,74,10x5-fold
	143	+-0.6053845679138585,19,B0 FE only,rep4_fold1,74,10x5-fold
	144	+-0.6115485612433651,18,B0 FE only,rep4_fold2,75,10x5-fold
	145	+-0.564873132879744,19,B0 FE only,rep4_fold3,74,10x5-fold
	146	+-0.6522727747291481,18,B0 FE only,rep4_fold4,75,10x5-fold
	147	+-0.636414272535366,18,B0 FE only,rep5_fold0,75,10x5-fold
	148	+-0.5947360345601029,19,B0 FE only,rep5_fold1,74,10x5-fold
	149	+-0.6499070428390907,18,B0 FE only,rep5_fold2,75,10x5-fold
	150	+-0.5250887869392054,20,B0 FE only,rep5_fold3,73,10x5-fold
	151	+-0.6318233675530855,18,B0 FE only,rep5_fold4,75,10x5-fold
	152	+-0.5521073305190212,20,B0 FE only,rep6_fold0,73,10x5-fold
	153	+-0.5919122421442115,19,B0 FE only,rep6_fold1,74,10x5-fold
	154	+-0.5337703887191214,18,B0 FE only,rep6_fold2,75,10x5-fold
	155	+-0.6236137172391328,18,B0 FE only,rep6_fold3,75,10x5-fold
	156	+-0.7049140457464247,18,B0 FE only,rep6_fold4,75,10x5-fold
	157	+-0.644855559978993,17,B0 FE only,rep7_fold0,76,10x5-fold
	158	+-0.7046229534579819,19,B0 FE only,rep7_fold1,74,10x5-fold
	159	+-0.5400887581411766,19,B0 FE only,rep7_fold2,74,10x5-fold
	160	+-0.5446441896236986,20,B0 FE only,rep7_fold3,73,10x5-fold
	161	+-0.6710479822860658,18,B0 FE only,rep7_fold4,75,10x5-fold
	162	+-0.6102553756976724,18,B0 FE only,rep8_fold0,75,10x5-fold
	163	+-0.6815279293780482,18,B0 FE only,rep8_fold1,75,10x5-fold
	164	+-0.5643261768453831,19,B0 FE only,rep8_fold2,74,10x5-fold
	165	+-0.5727934145606224,19,B0 FE only,rep8_fold3,74,10x5-fold
	166	+-0.5877778408134525,19,B0 FE only,rep8_fold4,74,10x5-fold
	167	+-0.5388195852281443,19,B0 FE only,rep9_fold0,74,10x5-fold
	168	+-0.6335145480962255,19,B0 FE only,rep9_fold1,74,10x5-fold
	169	+-0.6182810064205335,18,B0 FE only,rep9_fold2,75,10x5-fold
	170	+-0.6449700931487307,18,B0 FE only,rep9_fold3,75,10x5-fold
	171	+-0.6210108439503512,19,B0 FE only,rep9_fold4,74,10x5-fold
	172	+-0.6356854447252611,19,B1 D only,rep0_fold0,74,10x5-fold
	173	+-0.6553211776770863,17,B1 D only,rep0_fold1,76,10x5-fold
	174	+-0.6679381700412771,19,B1 D only,rep0_fold2,74,10x5-fold
	175	+-0.5564535853335865,19,B1 D only,rep0_fold3,74,10x5-fold
	176	+-0.5688101680099225,19,B1 D only,rep0_fold4,74,10x5-fold
	177	+-0.5578407746318227,19,B1 D only,rep1_fold0,74,10x5-fold
	178	+-0.5627905272673703,19,B1 D only,rep1_fold1,74,10x5-fold
	179	+-0.6262507576985021,19,B1 D only,rep1_fold2,74,10x5-fold
	180	+-0.5904495495870067,19,B1 D only,rep1_fold3,74,10x5-fold
	181	+-0.6445472687145267,17,B1 D only,rep1_fold4,76,10x5-fold
	182	+-0.584933249427394,20,B1 D only,rep2_fold0,73,10x5-fold
	183	+-0.6197009772503428,19,B1 D only,rep2_fold1,74,10x5-fold
	184	+-0.5679934940065683,18,B1 D only,rep2_fold2,75,10x5-fold
	185	+-0.6049902452913467,18,B1 D only,rep2_fold3,75,10x5-fold
	186	+-0.6217676789178803,18,B1 D only,rep2_fold4,75,10x5-fold
	187	+-0.6540509119804404,18,B1 D only,rep3_fold0,75,10x5-fold
	188	+-0.5580724206483056,18,B1 D only,rep3_fold1,75,10x5-fold
	189	+-0.5963334441829446,19,B1 D only,rep3_fold2,74,10x5-fold
	190	+-0.6145040492354771,18,B1 D only,rep3_fold3,75,10x5-fold
	191	+-0.5831753017647058,20,B1 D only,rep3_fold4,73,10x5-fold
	192	+-0.5833347514321969,19,B1 D only,rep4_fold0,74,10x5-fold
	193	+-0.5940569919637817,19,B1 D only,rep4_fold1,74,10x5-fold
	194	+-0.616658878178323,18,B1 D only,rep4_fold2,75,10x5-fold
	195	+-0.5599071121958769,19,B1 D only,rep4_fold3,74,10x5-fold
	196	+-0.6417955388754584,18,B1 D only,rep4_fold4,75,10x5-fold
	197	+-0.6236695730365845,18,B1 D only,rep5_fold0,75,10x5-fold
	198	+-0.5949552563760175,19,B1 D only,rep5_fold1,74,10x5-fold
	199	+-0.6393065424446964,18,B1 D only,rep5_fold2,75,10x5-fold
	200	+-0.5249632919529578,20,B1 D only,rep5_fold3,73,10x5-fold
	201	+-0.6119402007516135,18,B1 D only,rep5_fold4,75,10x5-fold
	202	+-0.5409178038460863,20,B1 D only,rep6_fold0,73,10x5-fold
	203	+-0.5960472611966452,19,B1 D only,rep6_fold1,74,10x5-fold
	204	+-0.5208047818462161,18,B1 D only,rep6_fold2,75,10x5-fold
	205	+-0.606959800218638,18,B1 D only,rep6_fold3,75,10x5-fold
	206	+-0.7021282225573476,18,B1 D only,rep6_fold4,75,10x5-fold
	207	+-0.6338286454608977,17,B1 D only,rep7_fold0,76,10x5-fold
	208	+-0.7011843072576646,19,B1 D only,rep7_fold1,74,10x5-fold
	209	+-0.5445052661624574,19,B1 D only,rep7_fold2,74,10x5-fold
	210	+-0.5325404448041982,20,B1 D only,rep7_fold3,73,10x5-fold
	211	+-0.6318024333903307,18,B1 D only,rep7_fold4,75,10x5-fold
	212	+-0.6116646610222912,18,B1 D only,rep8_fold0,75,10x5-fold
	213	+-0.6685797216651185,18,B1 D only,rep8_fold1,75,10x5-fold
	214	+-0.5643918843318513,19,B1 D only,rep8_fold2,74,10x5-fold
	215	+-0.5723092393157977,19,B1 D only,rep8_fold3,74,10x5-fold
	216	+-0.5791633044441097,19,B1 D only,rep8_fold4,74,10x5-fold
	217	+-0.5389448601448642,19,B1 D only,rep9_fold0,74,10x5-fold
	218	+-0.6079807602848534,19,B1 D only,rep9_fold1,74,10x5-fold
	219	+-0.6095520213859511,18,B1 D only,rep9_fold2,75,10x5-fold
	220	+-0.6313563418603431,18,B1 D only,rep9_fold3,75,10x5-fold
	221	+-0.6158644011981763,19,B1 D only,rep9_fold4,74,10x5-fold
	222	+-0.6794162410679633,19,B2 K only,rep0_fold0,74,10x5-fold
	223	+-0.6752333352285095,17,B2 K only,rep0_fold1,76,10x5-fold
	224	+-0.6810527273025005,19,B2 K only,rep0_fold2,74,10x5-fold
	225	+-0.5611212006291333,19,B2 K only,rep0_fold3,74,10x5-fold
	226	+-0.5866475458100563,19,B2 K only,rep0_fold4,74,10x5-fold
	227	+-0.5684592850762812,19,B2 K only,rep1_fold0,74,10x5-fold
	228	+-0.5685999683892149,19,B2 K only,rep1_fold1,74,10x5-fold
	229	+-0.6410171866693567,19,B2 K only,rep1_fold2,74,10x5-fold
	230	+-0.5925243419445171,19,B2 K only,rep1_fold3,74,10x5-fold
	231	+-0.6568158962592102,17,B2 K only,rep1_fold4,76,10x5-fold
	232	+-0.5983340396692196,20,B2 K only,rep2_fold0,73,10x5-fold
	233	+-0.6228132983555745,19,B2 K only,rep2_fold1,74,10x5-fold
	234	+-0.5715759954217449,18,B2 K only,rep2_fold2,75,10x5-fold
	235	+-0.5978375881583571,18,B2 K only,rep2_fold3,75,10x5-fold
	236	+-0.6331933446211148,18,B2 K only,rep2_fold4,75,10x5-fold
	237	+-0.6598134632702075,18,B2 K only,rep3_fold0,75,10x5-fold
	238	+-0.5655365751237913,18,B2 K only,rep3_fold1,75,10x5-fold
	239	+-0.6086740111389098,19,B2 K only,rep3_fold2,74,10x5-fold
	240	+-0.6285006043690673,18,B2 K only,rep3_fold3,75,10x5-fold
	241	+-0.5919101353752837,20,B2 K only,rep3_fold4,73,10x5-fold
	242	+-0.610268506647847,19,B2 K only,rep4_fold0,74,10x5-fold
	243	+-0.6081042250157828,19,B2 K only,rep4_fold1,74,10x5-fold
	244	+-0.6115698596799015,18,B2 K only,rep4_fold2,75,10x5-fold
	245	+-0.5679022449658021,19,B2 K only,rep4_fold3,74,10x5-fold
	246	+-0.6522888622674122,18,B2 K only,rep4_fold4,75,10x5-fold
	247	+-0.6397534146861698,18,B2 K only,rep5_fold0,75,10x5-fold
	248	+-0.5963569948403069,19,B2 K only,rep5_fold1,74,10x5-fold
	249	+-0.6499269897313157,18,B2 K only,rep5_fold2,75,10x5-fold
	250	+-0.5261217125726229,20,B2 K only,rep5_fold3,73,10x5-fold
	251	+-0.6333220913196165,18,B2 K only,rep5_fold4,75,10x5-fold
	252	+-0.5528256833594891,20,B2 K only,rep6_fold0,73,10x5-fold
	253	+-0.5947018431493715,19,B2 K only,rep6_fold1,74,10x5-fold
	254	+-0.5452391361722131,18,B2 K only,rep6_fold2,75,10x5-fold
	255	+-0.6298086124259137,18,B2 K only,rep6_fold3,75,10x5-fold
	256	+-0.7059924426866322,18,B2 K only,rep6_fold4,75,10x5-fold
	257	+-0.6456210440967394,17,B2 K only,rep7_fold0,76,10x5-fold
	258	+-0.7115715864301222,19,B2 K only,rep7_fold1,74,10x5-fold
	259	+-0.5415546204100566,19,B2 K only,rep7_fold2,74,10x5-fold
	260	+-0.5453876798903289,20,B2 K only,rep7_fold3,73,10x5-fold
	261	+-0.6709700310627044,18,B2 K only,rep7_fold4,75,10x5-fold
	262	+-0.6114446116776894,18,B2 K only,rep8_fold0,75,10x5-fold
	263	+-0.6923525658459138,18,B2 K only,rep8_fold1,75,10x5-fold
	264	+-0.5643538323113253,19,B2 K only,rep8_fold2,74,10x5-fold
	265	+-0.5744530208085011,19,B2 K only,rep8_fold3,74,10x5-fold
	266	+-0.5896587645068391,19,B2 K only,rep8_fold4,74,10x5-fold
	267	+-0.5406660120770627,19,B2 K only,rep9_fold0,74,10x5-fold
	268	+-0.6333437912062119,19,B2 K only,rep9_fold1,74,10x5-fold
	269	+-0.6245554325764014,18,B2 K only,rep9_fold2,75,10x5-fold
	270	+-0.6452464695901153,18,B2 K only,rep9_fold3,75,10x5-fold
	271	+-0.6244227367326036,19,B2 K only,rep9_fold4,74,10x5-fold
	272	+-0.6782720270006548,19,B3 complexity,rep0_fold0,74,10x5-fold
	273	+-0.6664715603728109,17,B3 complexity,rep0_fold1,76,10x5-fold
	274	+-0.6804947467624988,19,B3 complexity,rep0_fold2,74,10x5-fold
	275	+-0.5601333397027236,19,B3 complexity,rep0_fold3,74,10x5-fold
	276	+-0.5858671257564613,19,B3 complexity,rep0_fold4,74,10x5-fold
	277	+-0.5650370769304913,19,B3 complexity,rep1_fold0,74,10x5-fold
	278	+-0.5688121258934972,19,B3 complexity,rep1_fold1,74,10x5-fold
	279	+-0.6446838490401826,19,B3 complexity,rep1_fold2,74,10x5-fold
	280	+-0.5931417836793563,19,B3 complexity,rep1_fold3,74,10x5-fold
	281	+-0.6581770201164792,17,B3 complexity,rep1_fold4,76,10x5-fold
	282	+-0.6032385483625,20,B3 complexity,rep2_fold0,73,10x5-fold
	283	+-0.6206198499452557,19,B3 complexity,rep2_fold1,74,10x5-fold
	284	+-0.5716771075714067,18,B3 complexity,rep2_fold2,75,10x5-fold
	285	+-0.6032394197687556,18,B3 complexity,rep2_fold3,75,10x5-fold
	286	+-0.6332712113893304,18,B3 complexity,rep2_fold4,75,10x5-fold
	287	+-0.660589299714756,18,B3 complexity,rep3_fold0,75,10x5-fold
	288	+-0.5667623180967009,18,B3 complexity,rep3_fold1,75,10x5-fold
	289	+-0.6096062317894603,19,B3 complexity,rep3_fold2,74,10x5-fold
	290	+-0.6268650104471554,18,B3 complexity,rep3_fold3,75,10x5-fold
	291	+-0.5904061169240384,20,B3 complexity,rep3_fold4,73,10x5-fold
	292	+-0.6042762820259596,19,B3 complexity,rep4_fold0,74,10x5-fold
	293	+-0.6078027970898242,19,B3 complexity,rep4_fold1,74,10x5-fold
	294	+-0.6118997354825901,18,B3 complexity,rep4_fold2,75,10x5-fold
	295	+-0.5655932843836615,19,B3 complexity,rep4_fold3,74,10x5-fold
	296	+-0.6528022207657026,18,B3 complexity,rep4_fold4,75,10x5-fold
	297	+-0.63680268258605,18,B3 complexity,rep5_fold0,75,10x5-fold
	298	+-0.5951262498626473,19,B3 complexity,rep5_fold1,74,10x5-fold
	299	+-0.6507694427513745,18,B3 complexity,rep5_fold2,75,10x5-fold
	300	+-0.524882759645932,20,B3 complexity,rep5_fold3,73,10x5-fold
	301	+-0.6325532752789995,18,B3 complexity,rep5_fold4,75,10x5-fold
	302	+-0.5524999674270533,20,B3 complexity,rep6_fold0,73,10x5-fold
	303	+-0.5946803049166622,19,B3 complexity,rep6_fold1,74,10x5-fold
	304	+-0.5478815726325732,18,B3 complexity,rep6_fold2,75,10x5-fold
	305	+-0.6310749317752484,18,B3 complexity,rep6_fold3,75,10x5-fold
	306	+-0.7049196236105723,18,B3 complexity,rep6_fold4,75,10x5-fold
	307	+-0.6448428775388669,17,B3 complexity,rep7_fold0,76,10x5-fold
	308	+-0.7113579003755159,19,B3 complexity,rep7_fold1,74,10x5-fold
	309	+-0.5398588099695848,19,B3 complexity,rep7_fold2,74,10x5-fold
	310	+-0.5470172534876638,20,B3 complexity,rep7_fold3,73,10x5-fold
	311	+-0.6726137128632598,18,B3 complexity,rep7_fold4,75,10x5-fold
	312	+-0.6110655707019442,18,B3 complexity,rep8_fold0,75,10x5-fold
	313	+-0.6942024879857703,18,B3 complexity,rep8_fold1,75,10x5-fold
	314	+-0.5641770375316484,19,B3 complexity,rep8_fold2,74,10x5-fold
	315	+-0.5786780426007944,19,B3 complexity,rep8_fold3,74,10x5-fold
	316	+-0.590758760190815,19,B3 complexity,rep8_fold4,74,10x5-fold
	317	+-0.5436653478281098,19,B3 complexity,rep9_fold0,74,10x5-fold
	318	+-0.6341060948474395,19,B3 complexity,rep9_fold1,74,10x5-fold
	319	+-0.623922218882318,18,B3 complexity,rep9_fold2,75,10x5-fold
	320	+-0.6448090874663114,18,B3 complexity,rep9_fold3,75,10x5-fold
	321	+-0.6225350987601201,19,B3 complexity,rep9_fold4,74,10x5-fold
	322	+-0.671985164747131,19,B4 persistence,rep0_fold0,74,10x5-fold
	323	+-0.6582322691892459,17,B4 persistence,rep0_fold1,76,10x5-fold
	324	+-0.6735242415042979,19,B4 persistence,rep0_fold2,74,10x5-fold
	325	+-0.5576017586396346,19,B4 persistence,rep0_fold3,74,10x5-fold
	326	+-0.580759589149185,19,B4 persistence,rep0_fold4,74,10x5-fold
	327	+-0.5633034046366819,19,B4 persistence,rep1_fold0,74,10x5-fold
	328	+-0.5632829325041058,19,B4 persistence,rep1_fold1,74,10x5-fold
	329	+-0.6351330149989289,19,B4 persistence,rep1_fold2,74,10x5-fold
	330	+-0.5927035550010715,19,B4 persistence,rep1_fold3,74,10x5-fold
	331	+-0.6502583137429893,17,B4 persistence,rep1_fold4,76,10x5-fold
	332	+-0.5975777994496894,20,B4 persistence,rep2_fold0,73,10x5-fold
	333	+-0.6146894856889363,19,B4 persistence,rep2_fold1,74,10x5-fold
	334	+-0.5664857536847796,18,B4 persistence,rep2_fold2,75,10x5-fold
	335	+-0.5881312659296789,18,B4 persistence,rep2_fold3,75,10x5-fold
	336	+-0.6297036700907955,18,B4 persistence,rep2_fold4,75,10x5-fold
	337	+-0.6619913803481575,18,B4 persistence,rep3_fold0,75,10x5-fold
	338	+-0.5614865927328828,18,B4 persistence,rep3_fold1,75,10x5-fold
	339	+-0.6018874550671651,19,B4 persistence,rep3_fold2,74,10x5-fold
	340	+-0.6189338937968742,18,B4 persistence,rep3_fold3,75,10x5-fold
	341	+-0.5886984632291674,20,B4 persistence,rep3_fold4,73,10x5-fold
	342	+-0.5927450878182084,19,B4 persistence,rep4_fold0,74,10x5-fold
	343	+-0.598778991085936,19,B4 persistence,rep4_fold1,74,10x5-fold
	344	+-0.6116579228459538,18,B4 persistence,rep4_fold2,75,10x5-fold
	345	+-0.5624758831746948,19,B4 persistence,rep4_fold3,74,10x5-fold
	346	+-0.6453306053080126,18,B4 persistence,rep4_fold4,75,10x5-fold
	347	+-0.6311865271208418,18,B4 persistence,rep5_fold0,75,10x5-fold
	348	+-0.6020541294621151,19,B4 persistence,rep5_fold1,74,10x5-fold
	349	+-0.6541112465538,18,B4 persistence,rep5_fold2,75,10x5-fold
	350	+-0.5209835704860011,20,B4 persistence,rep5_fold3,73,10x5-fold
	351	+-0.6248240233639719,18,B4 persistence,rep5_fold4,75,10x5-fold
	352	+-0.5485244586179201,20,B4 persistence,rep6_fold0,73,10x5-fold
	353	+-0.5896579558810591,19,B4 persistence,rep6_fold1,74,10x5-fold
	354	+-0.5289368789051853,18,B4 persistence,rep6_fold2,75,10x5-fold
	355	+-0.6192830925910812,18,B4 persistence,rep6_fold3,75,10x5-fold
	356	+-0.7137694562981892,18,B4 persistence,rep6_fold4,75,10x5-fold
	357	+-0.6422026456270938,17,B4 persistence,rep7_fold0,76,10x5-fold
	358	+-0.7027488241129795,19,B4 persistence,rep7_fold1,74,10x5-fold
	359	+-0.5411451988967334,19,B4 persistence,rep7_fold2,74,10x5-fold
	360	+-0.539522713586498,20,B4 persistence,rep7_fold3,73,10x5-fold
	361	+-0.6590032566537575,18,B4 persistence,rep7_fold4,75,10x5-fold
	362	+-0.6225616782357821,18,B4 persistence,rep8_fold0,75,10x5-fold
	363	+-0.6780312410586735,18,B4 persistence,rep8_fold1,75,10x5-fold
	364	+-0.5620546706652177,19,B4 persistence,rep8_fold2,74,10x5-fold
	365	+-0.5711418679988245,19,B4 persistence,rep8_fold3,74,10x5-fold
	366	+-0.5832363994567088,19,B4 persistence,rep8_fold4,74,10x5-fold
	367	+-0.5345815408150943,19,B4 persistence,rep9_fold0,74,10x5-fold
	368	+-0.6243650997622889,19,B4 persistence,rep9_fold1,74,10x5-fold
	369	+-0.6137189154704589,18,B4 persistence,rep9_fold2,75,10x5-fold
	370	+-0.6407980460796847,18,B4 persistence,rep9_fold3,75,10x5-fold
	371	+-0.6208565323773663,19,B4 persistence,rep9_fold4,74,10x5-fold
	372	+-0.6337948842819158,19,B5 D+K,rep0_fold0,74,10x5-fold
	373	+-0.6596753404433631,17,B5 D+K,rep0_fold1,76,10x5-fold
	374	+-0.6692213522232436,19,B5 D+K,rep0_fold2,74,10x5-fold
	375	+-0.5567168597634137,19,B5 D+K,rep0_fold3,74,10x5-fold
	376	+-0.5721943483073698,19,B5 D+K,rep0_fold4,74,10x5-fold
	377	+-0.5591026304247548,19,B5 D+K,rep1_fold0,74,10x5-fold
	378	+-0.5631897242368877,19,B5 D+K,rep1_fold1,74,10x5-fold
	379	+-0.6261084864620554,19,B5 D+K,rep1_fold2,74,10x5-fold
	380	+-0.5920110809571945,19,B5 D+K,rep1_fold3,74,10x5-fold
	381	+-0.6435381227283641,17,B5 D+K,rep1_fold4,76,10x5-fold
	382	+-0.5872466822190383,20,B5 D+K,rep2_fold0,73,10x5-fold
	383	+-0.6211811256606243,19,B5 D+K,rep2_fold1,74,10x5-fold
	384	+-0.5685072251963441,18,B5 D+K,rep2_fold2,75,10x5-fold
	385	+-0.6054449025341113,18,B5 D+K,rep2_fold3,75,10x5-fold
	386	+-0.6221258470593989,18,B5 D+K,rep2_fold4,75,10x5-fold
	387	+-0.6548442428965751,18,B5 D+K,rep3_fold0,75,10x5-fold
	388	+-0.5577859789586922,18,B5 D+K,rep3_fold1,75,10x5-fold
	389	+-0.5951273764464787,19,B5 D+K,rep3_fold2,74,10x5-fold
	390	+-0.6167510261573316,18,B5 D+K,rep3_fold3,75,10x5-fold
	391	+-0.5828881223316793,20,B5 D+K,rep3_fold4,73,10x5-fold
	392	+-0.5953416377187439,19,B5 D+K,rep4_fold0,74,10x5-fold
	393	+-0.5976548390819251,19,B5 D+K,rep4_fold1,74,10x5-fold
	394	+-0.6157384171455546,18,B5 D+K,rep4_fold2,75,10x5-fold
	395	+-0.5560487791224314,19,B5 D+K,rep4_fold3,74,10x5-fold
	396	+-0.6412271540719632,18,B5 D+K,rep4_fold4,75,10x5-fold
	397	+-0.6242459138566026,18,B5 D+K,rep5_fold0,75,10x5-fold
	398	+-0.5958904248518524,19,B5 D+K,rep5_fold1,74,10x5-fold
	399	+-0.6385934101737542,18,B5 D+K,rep5_fold2,75,10x5-fold
	400	+-0.5249694823728337,20,B5 D+K,rep5_fold3,73,10x5-fold
	401	+-0.6105819186162433,18,B5 D+K,rep5_fold4,75,10x5-fold
	402	+-0.5405701733448982,20,B5 D+K,rep6_fold0,73,10x5-fold
	403	+-0.5969800644402191,19,B5 D+K,rep6_fold1,74,10x5-fold
	404	+-0.5337580106622606,18,B5 D+K,rep6_fold2,75,10x5-fold
	405	+-0.6133796880055206,18,B5 D+K,rep6_fold3,75,10x5-fold
	406	+-0.7015359359689189,18,B5 D+K,rep6_fold4,75,10x5-fold
	407	+-0.6348404643046642,17,B5 D+K,rep7_fold0,76,10x5-fold
	408	+-0.7013263206291614,19,B5 D+K,rep7_fold1,74,10x5-fold
	409	+-0.5520701716423618,19,B5 D+K,rep7_fold2,74,10x5-fold
	410	+-0.5338125532883111,20,B5 D+K,rep7_fold3,73,10x5-fold
	411	+-0.6274481073771072,18,B5 D+K,rep7_fold4,75,10x5-fold
	412	+-0.6123948039475721,18,B5 D+K,rep8_fold0,75,10x5-fold
	413	+-0.6798217477880797,18,B5 D+K,rep8_fold1,75,10x5-fold
	414	+-0.5699974946711587,19,B5 D+K,rep8_fold2,74,10x5-fold
	415	+-0.5721301963253125,19,B5 D+K,rep8_fold3,74,10x5-fold
	416	+-0.5787699185206181,19,B5 D+K,rep8_fold4,74,10x5-fold
	417	+-0.5386548821922236,19,B5 D+K,rep9_fold0,74,10x5-fold
	418	+-0.6042877935927196,19,B5 D+K,rep9_fold1,74,10x5-fold
	419	+-0.610654978308027,18,B5 D+K,rep9_fold2,75,10x5-fold
	420	+-0.6306542871932357,18,B5 D+K,rep9_fold3,75,10x5-fold
	421	+-0.6181047939706994,19,B5 D+K,rep9_fold4,74,10x5-fold

modified results/empirical/h2_seshat_cv_folds_binary.csv +351 −358

Large diff (709 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h2_seshat_cv_wins.csv +12 −12

Large diff (25 lines) is not loaded by default to keep the page fast.Load diff

added results/empirical/h2_seshat_items.csv +15 −0

Large diff (15 lines) is not loaded by default to keep the page fast.Load diff

added results/empirical/h2_seshat_items_dimensionality.json +6 −0

Large diff (7 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h2_seshat_primary.csv +13 −11

Large diff (28 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h2_seshat_robustness.csv +20 −11

Large diff (32 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h2_seshat_y2_y3.csv +3 −0

Large diff (8 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h4_modern_curve.csv +40 −40

Large diff (81 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h4_modern_folds.csv +15 −15

Large diff (31 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h4_modern_summary.json +5 −5

Large diff (15 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h4_seshat_folds.csv +51 −53

Large diff (104 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h4_seshat_summary.json +10 −10

Large diff (23 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h5_modern_blocks.csv +10 −10

Large diff (21 lines) is not loaded by default to keep the page fast.Load diff

modified results/empirical/h5_seshat_blocks.csv +26 −16

Large diff (43 lines) is not loaded by default to keep the page fast.Load diff

added results/empirical/measurement_crisis_sample.json +10 −0

Large diff (11 lines) is not loaded by default to keep the page fast.Load diff

modified results/logs/execution_log.md +2 −0

Large diff (5 lines) is not loaded by default to keep the page fast.Load diff

modified results/logs/final_research_audit.md +12 −8

Large diff (39 lines) is not loaded by default to keep the page fast.Load diff

modified results/logs/h2_modern_run.log +77 −74

Large diff (173 lines) is not loaded by default to keep the page fast.Load diff

modified results/logs/h2_seshat_run.log +124 −99

Large diff (239 lines) is not loaded by default to keep the page fast.Load diff

modified results/logs/sim_analyze.log +36 −37

Large diff (121 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/ablation_admin_capacity.csv +14 −14

Large diff (29 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/cells_institution_vs_none.csv +432 −432

Large diff (867 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/cells_main.csv +432 −432

Large diff (1,189 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/institution_loss_shares.csv +3 −3

Large diff (10 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/paired_vs_none.csv +4 −4

Large diff (11 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/regime_by_N.csv +12 −12

Large diff (28 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/runs_ablation.parquet +0 −0

Binary file not shown.

modified results/simulated/runs_main.parquet +0 −0

Binary file not shown.

modified results/simulated/shock_effects.csv +11 −11

Large diff (26 lines) is not loaded by default to keep the page fast.Load diff

added results/simulated/shock_effects_by_bandwidth.csv +29 −0

Large diff (29 lines) is not loaded by default to keep the page fast.Load diff

modified results/simulated/simulation_summary.json +97 −97

Large diff (334 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/econometrics/common.py +22 −5

Large diff (48 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/econometrics/h2_modern.py +17 −8

Large diff (87 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/econometrics/h2_seshat.py +60 −17

Large diff (158 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/harmonization/seshat.py +8 −4

Large diff (30 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/measurement/build_indices.py +34 −2

Large diff (54 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/simulations/analyze.py +3 −1

Large diff (17 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/simulations/model.py +42 −14

Large diff (94 lines) is not loaded by default to keep the page fast.Load diff

modified src/wp35/visualization/make_all.py +5 −0

Large diff (11 lines) is not loaded by default to keep the page fast.Load diff

modified tests/test_harmonization.py +1 −1

Large diff (9 lines) is not loaded by default to keep the page fast.Load diff