paper: model/implementation/limitations/appendices; docs consistency fixes

Simon-Pierre Boucher committed 6 days ago (Oct 27, 2026) parent 74b1f46 Browse files

6 changed files +315 −272

modified docs/assumptions.md +1 −1

				@@ -8,7 +8,7 @@ Version 1.0, 2026-10-27. Each row states the assumption as implemented in `src
8	\|---\|---\|---\|---\|---\|	8	\|---\|---\|---\|---\|---\|
9	\| A1 \| Discrete model time: 600 ticks, outcomes on t ≥ 100; no calendar mapping \| Defines the horizon of every metric \| Avoids false precision; 30 dividend periods suffice for ACF/VR estimates (≥ 500 returns) \| `n_ticks` 1,000 on the control (free); report whether rankings of cells change \|	9	\| A1 \| Discrete model time: 600 ticks, outcomes on t ≥ 100; no calendar mapping \| Defines the horizon of every metric \| Avoids false precision; 30 dividend periods suffice for ACF/VR estimates (≥ 500 returns) \| `n_ticks` 1,000 on the control (free); report whether rankings of cells change \|
10	\| A2 \| Dividend every 20 ticks: D = μ + x + J, x_t = 0.98 x_{t−1} + 2 ε_t (cents), μ = 50 c, r = 0.25 % per period; one asset by default \| Fundamental process; V ≈ 200.00 and P/D ≈ 400 \| Persistent, stationary cash-flow state with a closed-form PV (brief §Economy) \| Sweep ρ ∈ {0.95, 0.98}, σ ∈ {1, 2, 4} on the control (free) \|	10	\| A2 \| Dividend every 20 ticks: D = μ + x + J, x_t = 0.98 x_{t−1} + 2 ε_t (cents), μ = 50 c, r = 0.25 % per period; one asset by default \| Fundamental process; V ≈ 200.00 and P/D ≈ 400 \| Persistent, stationary cash-flow state with a closed-form PV (brief §Economy) \| Sweep ρ ∈ {0.95, 0.98}, σ ∈ {1, 2, 4} on the control (free) \|
11	−\| A3 \| "Efficient price" = full-information PV V_t = μ/r + x_t·ρ/(1 + r − ρ), risk-neutral, constant discount rate, no risk premium \| Benchmark for `msd_log`, `mad_pct`, bubbles, absorption \| The only unambiguous benchmark in a simulated economy; the latent state is never shown to agents \| Also report deviations from the public PV (`public_value`) and from the end-of-run mean price \|	11	+\| A3 \| "Efficient price" = full-information PV V_t = μ/r + x_n/(1 + r − ρ) with x_n the latent deviation of the upcoming dividend, risk-neutral, constant discount rate, no risk premium \| Benchmark for `msd_log`, `mad_pct`, bubbles, absorption \| The only unambiguous benchmark in a simulated economy; the latent state is never shown to agents \| Also report deviations from the public PV (`public_value`) and from the end-of-run mean price \|
12	\| A4 \| Public PV = Kalman/HMM posterior on announced dividends; in the baseline (no jumps, one regime) the dividend reveals x_t exactly, so V^pub ≡ V \| Information structure \| Simplest case for calibration; but it makes private signals redundant and H1 unidentified \| Confirmatory base uses noisy dividends (`jump_prob` 1, `jump_sigma` ≈ 4 c → SD(V^pub/V − 1) ≈ 0.5 %); both variants reported \|	12	\| A4 \| Public PV = Kalman/HMM posterior on announced dividends; in the baseline (no jumps, one regime) the dividend reveals x_t exactly, so V^pub ≡ V \| Information structure \| Simplest case for calibration; but it makes private signals redundant and H1 unidentified \| Confirmatory base uses noisy dividends (`jump_prob` 1, `jump_sigma` ≈ 4 c → SD(V^pub/V − 1) ≈ 0.5 %); both variants reported \|
13	\| A5 \| No look-ahead: private signals s = x_t + N(0, σ²/precision) are about the current state; news reports x_t with noise; nobody sees future draws \| Rules out mechanical information advantages \| Standard in noisy rational-expectations settings \| Unit test on the signal function; prompt audit (`prompting.py` builds only from `Observation`) \|	13	\| A5 \| No look-ahead: private signals s = x_t + N(0, σ²/precision) are about the current state; news reports x_t with noise; nobody sees future draws \| Rules out mechanical information advantages \| Standard in noisy rational-expectations settings \| Unit test on the signal function; prompt audit (`prompting.py` builds only from `Observation`) \|
14	\| A6 \| Shock: latent jump in x at t = 300 revealed through dividends/news; placebo at t = 450. Baseline size 40 c (+8.4 % on V); confirmatory size 10 c (+2.1 %) \| Event-study outcomes (`time_to_absorb`, `response_ratio`, local projections) \| A sub-threshold (< 5 %) jump so that a lagged response is not counted as a bubble; placebo for falsification \| Sizes 10/20/40 c on the control; `news_burst` variant \|	14	\| A6 \| Shock: latent jump in x at t = 300 revealed through dividends/news; placebo at t = 450. Baseline size 40 c (+8.4 % on V); confirmatory size 10 c (+2.1 %) \| Event-study outcomes (`time_to_absorb`, `response_ratio`, local projections) \| A sub-threshold (< 5 %) jump so that a lagged response is not counted as a bubble; placebo for falsification \| Sizes 10/20/40 c on the control; `news_burst` variant \|

modified docs/model_cards.md +1 −1

				`@@ -77,7 +77,7 @@ One card per model or provider used or planned. Prices are USD per 1M tokens (in`
77	\| Field \| Value \|	77	\| Field \| Value \|
78	\|---\|---\|	78	\|---\|---\|
79	\| Implementation \| `llm/library.py`: 7 standardized features of the observation (mispricing vs public PV, momentum, signal, inventory and cash ratios, spread, open-order flag); returns the recorded decision of one of the 5 nearest neighbours (seeded draw), limit price rescaled to the current mid \|	79	\| Implementation \| `llm/library.py`: 7 standardized features of the observation (mispricing vs public PV, momentum, signal, inventory and cash ratios, spread, open-order flag); returns the recorded decision of one of the 5 nearest neighbours (seeded draw), limit price rescaled to the current mid \|
80	−\| Source \| Valid decisions recorded in the T1 gpt-5.4-nano runs (`results/library/nano_v1.npz`) \|	80	+\| Source \| Valid decisions recorded in the T1 gpt-5.4-nano runs (`results/library/gpt-5.4-nano_single.npz`) \|
81	\| Price \| 0 \|	81	\| Price \| 0 \|
82	\| Role \| Scale arm at N = 1,000 and the "cached" side of the cached-vs-online comparison; runs are labelled provider `library` \|	82	\| Role \| Scale arm at N = 1,000 and the "cached" side of the cached-vs-online comparison; runs are labelled provider `library` \|
83	\| Caveat \| Not an adaptive agent; out-of-support at large N must be reported (distance diagnostics, `protocol.md` §5.4) \|	83	\| Caveat \| Not an adaptive agent; out-of-support at large N must be reported (distance diagnostics, `protocol.md` §5.4) \|

modified docs/protocol.md +5 −5

				@@ -13,14 +13,14 @@ Version 1.0, 2026-10-27. Operational companion to `identification.md`. Every num
13	## 2. Commands	13	## 2. Commands
14		14
15	```bash	15	```bash
16	−.venv/bin/python -m pytest tests # 23 tests (unit, property, integration) must pass	16	+.venv/bin/python -m pytest tests # 24 tests (unit, property, integration) must pass
17	.venv/bin/ma run configs/baseline.yaml --seed 1 --out results/runs/baseline/seed0001 # one classical run (~0.5 s)	17	.venv/bin/ma run configs/baseline.yaml --seed 1 --out results/runs/baseline/seed0001 # one classical run (~0.5 s)
18	.venv/bin/ma budget configs/<design>.yaml --tokens-in 760 --tokens-out 230 # calls and USD before any live design	18	.venv/bin/ma budget configs/<design>.yaml --tokens-in 760 --tokens-out 230 # calls and USD before any live design
19	.venv/bin/ma design configs/<design>.yaml --workers 8 --cache results/cache/<design>.sqlite	19	.venv/bin/ma design configs/<design>.yaml --workers 8 --cache results/cache/<design>.sqlite
20	.venv/bin/ma design configs/<design>.yaml --cells share_25 --max-seeds 1 --cache ... # single cell / single seed	20	.venv/bin/ma design configs/<design>.yaml --cells share_25 --max-seeds 1 --cache ... # single cell / single seed
21	```	21	```
22		22
23	−`ma design` runs mock cells in parallel processes (`--workers`) and any cell with a live provider sequentially with a shared cache (the worker count is ignored for API cells). Existing `outcomes.json` files are reused, so an interrupted design resumes without re-calling the API.	23	+`ma design` runs every cell in a process pool (`--workers`; the Makefile uses 3 for the pilot and 5 for the confirmatory design) with one shared SQLite decision cache (`results/cache/llm_cache.sqlite`). An existing run directory is reused only if its manifest `config_hash` equals the requested configuration's hash (otherwise it is deleted and re-run), so an interrupted design resumes without re-calling the API and a changed configuration can never pick up stale runs. Before relaunching after a code change, delete the design directory and make sure no pool worker is still alive (`pkill -f multiprocessing.spawn`).
24		24
25	## 3. Phase 5a — mock dry-run (no cost)	25	## 3. Phase 5a — mock dry-run (no cost)
26		26
				`@@ -68,7 +68,7 @@ Pilot pass criteria: validity ≥ 80 %, failures ≤ 10 %, cost per decision wit`
68	\| \| poslimit_25 \| `exchange.position_limit: 300` \| 101–102 \| 1,350 \| 0.68 \|	68	\| \| poslimit_25 \| `exchange.position_limit: 300` \| 101–102 \| 1,350 \| 0.68 \|
69	\| T5 fairness \| mini_25_equalusd \| mini_25 with `llm.usd_cap` = mean nano spend of share_25 (TBD ≈ 0.34) → decisions stop at equal USD \| 101–103 \| ≤ 630 mini \| ≤ 1.01 \|	69	\| T5 fairness \| mini_25_equalusd \| mini_25 with `llm.usd_cap` = mean nano spend of share_25 (TBD ≈ 0.34) → decisions stop at equal USD \| 101–103 \| ≤ 630 mini \| ≤ 1.01 \|
70	\| T6 extensions \| k2_25 \| two assets, `factor_loading: [0.7, 0.7]`, `statarb: 0.05` \| 101–102 \| 1,350 \| 0.68 \|	70	\| T6 extensions \| k2_25 \| two assets, `factor_loading: [0.7, 0.7]`, `statarb: 0.05` \| 101–102 \| 1,350 \| 0.68 \|
71	−\| \| library_1000 \| `population.n_traders: 1000`, `llm.frozen_library: results/library/nano_v1.npz`, provider `library` \| 101–110 \| 0 \| 0 \|	71	+\| \| library_1000 \| `population.n_traders: 1000`, `llm.frozen_library: results/library/gpt-5.4-nano_single.npz`, provider `library` \| 101–110 \| 0 \| 0 \|
72	\| \| auction_25, interval_10 \| call auction every 5 ticks; `decision_interval: 10` — mock only \| 1–10 \| 0 \| 0 \|	72	\| \| auction_25, interval_10 \| call auction every 5 ticks; `decision_interval: 10` — mock only \| 1–10 \| 0 \| 0 \|
73	\| Confirmatory total \| \| \| \| ≈ 66,000 \| ≈ 33.3 \|	73	\| Confirmatory total \| \| \| \| ≈ 66,000 \| ≈ 33.3 \|
74	\| Pilot (§4) \| \| \| \| 6,480 \| 3.24 \|	74	\| Pilot (§4) \| \| \| \| 6,480 \| 3.24 \|
				@@ -83,11 +83,11 @@ Tiers run T1 → T6; spend is reconciled after each tier (`sum(llm_usd)` over `o
83		83
84	### 5.4 Frozen-library replay (T6)	84	### 5.4 Frozen-library replay (T6)
85		85
86	−Build the library from the T1 nano runs: `DecisionLibrary.from_runs([...T1 run dirs...]).save("results/library/nano_v1.npz")`. Replay at N = 1,000 with `activity` scaled by 100/N (as in `scripts/11_scale_benchmark.py`) so order flow per tick is constant. Report the nearest-neighbour distance distribution; if the median standardized distance at N = 1,000 exceeds the 90th percentile observed at N = 100, the arm is reported as out-of-support, not as evidence.	86	+Build the library from the T1 nano runs: `DecisionLibrary.from_runs([...T1 run dirs...]).save("results/library/gpt-5.4-nano_single.npz")`. Replay at N = 1,000 with `activity` scaled by 100/N (as in `scripts/11_scale_benchmark.py`) so order flow per tick is constant. Report the nearest-neighbour distance distribution; if the median standardized distance at N = 1,000 exceeds the 90th percentile observed at N = 100, the arm is reported as out-of-support, not as evidence.
87		87
88	## 6. Replication checklist	88	## 6. Replication checklist
89		89
90	−1. `git log -1` matches `freeze.json`; `pytest` passes (23 tests).	90	+1. `git log -1` matches `freeze.json`; `pytest` passes (24 tests).
91	2. `ma design` of each tier reproduces `outcomes.parquet` from the cache (`--cache` path, `llm_cached_hits` = 1.0, zero USD).	91	2. `ma design` of each tier reproduces `outcomes.parquet` from the cache (`--cache` path, `llm_cached_hits` = 1.0, zero USD).
92	3. Every run directory has `manifest.json` with `config_hash`, `git_revision`, `prompt_version`, `price_assumptions`, `llm_usage`, `shocks`, `final_conservation_ok`.	92	3. Every run directory has `manifest.json` with `config_hash`, `git_revision`, `prompt_version`, `price_assumptions`, `llm_usage`, `shocks`, `final_conservation_ok`.
93	4. Analysis scripts read only `outcomes.parquet`, `ticks.parquet`, `trades.parquet`, `llm_decisions.parquet`; tables and figures are regenerated by one command (`paper/Makefile`).	93	4. Analysis scripts read only `outcomes.parquet`, `ticks.parquet`, `trades.parquet`, `llm_decisions.parquet`; tables and figures are regenerated by one command (`paper/Makefile`).

modified docs/research_log.md +1 −0

				`@@ -18,6 +18,7 @@ between the pre-analysis plan and the confirmatory runs, so that nothing is revi`
18	\| 11 \| 21:12 \| — \| First git commit; pilot and confirmatory designs launched on the frozen revision \| manifests now carry `git_revision` \| freeze \|	18	\| 11 \| 21:12 \| — \| First git commit; pilot and confirmatory designs launched on the frozen revision \| manifests now carry `git_revision` \| freeze \|
19	\| 12 \| — \| 5 \| Pilot (3 seeds × {5, 25, 50 %}, gpt-5.4-nano) runs concurrently with the confirmatory design on the same revision; pilot seeds (1–3) are never pooled with confirmatory seeds (201–207) \| time budget \| the pilot's role is reduced to cost/validity/variance reporting; the seed count (7) was fixed by the USD cap, not by the pilot variance \|	19	\| 12 \| — \| 5 \| Pilot (3 seeds × {5, 25, 50 %}, gpt-5.4-nano) runs concurrently with the confirmatory design on the same revision; pilot seeds (1–3) are never pooled with confirmatory seeds (201–207) \| time budget \| the pilot's role is reduced to cost/validity/variance reporting; the seed count (7) was fixed by the USD cap, not by the pilot variance \|
20	\| 13 \| 22:05 \| 6 \| Override bug: `deep_update` merged leaf dictionaries, so `provider_mix: {openai: 1.0}` on the baseline `{mock: 1.0}` became a 50/50 mock/OpenAI mixture in every live cell (caught by a co-author reading run manifests). Leaf dicts now replace; regression test added; all live and classical designs discarded and relaunched on commit c1fdf6c \| half of the "LLM" slots were the mock rule \| new freeze (c1fdf6c); nothing from the discarded runs is used \|	20	\| 13 \| 22:05 \| 6 \| Override bug: `deep_update` merged leaf dictionaries, so `provider_mix: {openai: 1.0}` on the baseline `{mock: 1.0}` became a 50/50 mock/OpenAI mixture in every live cell (caught by a co-author reading run manifests). Leaf dicts now replace; regression test added; all live and classical designs discarded and relaunched on commit c1fdf6c \| half of the "LLM" slots were the mock rule \| new freeze (c1fdf6c); nothing from the discarded runs is used \|
		21	+\| 14 \| 22:40 \| 6 \| Orphaned pool workers from killed designs had kept writing runs after the directories were purged, and the resumable runner reused them (manifests with providers [mock, openai]); all workers killed, every run whose manifest hash differs from its cell's configuration deleted (3 pilot, 7 confirmatory), runner now resumes only on an identical `config_hash` (commit 74b1f46) \| co-author audit of manifests \| clean runs (6 pilot, 20 confirmatory) kept; discarded runs cost ≈ 12 USD of API calls that are reported in the cost accounting but never analysed \|
21		22
22	Known deviations from `docs/identification.md` / `docs/protocol.md` (written before items 8–12):	23	Known deviations from `docs/identification.md` / `docs/protocol.md` (written before items 8–12):
23	- Noisy dividends (`jump_prob = 1`) were proposed to make V^pub ≠ V; item 8 achieves this through the timing of information instead (no jumps in the baseline).	24	- Noisy dividends (`jump_prob = 1`) were proposed to make V^pub ≠ V; item 8 achieves this through the timing of information instead (no jumps in the baseline).

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

				`@@ -424,10 +424,10 @@ $L$, $\ell$ & position limit (per-agent order/position cap; \texttt{allow\_short`
424	$\theta$, $h$ & \texttt{exchange.halt\_threshold}, \texttt{halt\_window}, \texttt{halt\_duration} (circuit breaker) \\	424	$\theta$, $h$ & \texttt{exchange.halt\_threshold}, \texttt{halt\_window}, \texttt{halt\_duration} (circuit breaker) \\
425	$z$, $\sigma_z^2$ & noise-trader slots (\texttt{mix[noise]}, activity) \\	425	$z$, $\sigma_z^2$ & noise-trader slots (\texttt{mix[noise]}, activity) \\
426	$N$ & \texttt{population.n\_traders} ($100$, $1{,}000$, $10{,}000$) \\	426	$N$ & \texttt{population.n\_traders} ($100$, $1{,}000$, $10{,}000$) \\
427	−$\tau$ & $R^2$ of the latent value $V_t$ on the mid price (post-warmup) or $1-\operatorname{Var}(V_t-P_t)/\operatorname{Var}(V_t)$ \\	427	+$\tau$ & not estimated directly; its simulator counterparts are the mean squared log pricing error \texttt{msd\_log} $=\overline{(\log P_t-\log V_t)^2}$ and the mean absolute deviation \texttt{mad\_pct}, both decreasing in informativeness \\
428	$\operatorname{Var}(P-v)$ & mean squared pricing error $(P_t-V_t)^2$ \\	428	$\operatorname{Var}(P-v)$ & mean squared pricing error $(P_t-V_t)^2$ \\
429	$\bar\rho$, $\bar S$ & average pairwise correlation of signed order flow across agents; share of agent pairs with same-signed orders within a decision interval \\	429	$\bar\rho$, $\bar S$ & average pairwise correlation of signed order flow across agents; share of agent pairs with same-signed orders within a decision interval \\
430	−$\Delta P^\ast$, $T_\epsilon$ & crash depth after a shock event (\texttt{shocks}); time-to-absorb = ticks until $\|P_t-V^{\mathrm{pub}}_t\|$ re-enters a tolerance band \\	430	+$\Delta P^\ast$, $T_\epsilon$ & crash depth after a shock event (\texttt{shocks}); time-to-absorb = ticks until $\|P_t/V_t-1\|$ stays below 1\,\% for five consecutive ticks after the latent shock (\texttt{time\_to\_absorb}) \\
431	\hline	431	\hline
432	\end{tabular}	432	\end{tabular}
433	\end{table}	433	\end{table}

modified results/calibration/scale_benchmark.json +305 −263

				`@@ -2,372 +2,414 @@`
2	{	2	{
3	"name": "scale_100",	3	"name": "scale_100",
4	"seed": 1,	4	"seed": 1,
5	− "config_hash": "72ec6dde7c77c2e9",	5	+ "config_hash": "13328f5e6f32990c",
6	"n_traders": 100,	6	"n_traders": 100,
7	"llm_share": 0.0,	7	"llm_share": 0.0,
8	"llm_share_traders": 0.0,	8	"llm_share_traders": 0.0,
9	"llm_share_orders": 0.0,	9	"llm_share_orders": 0.0,
10	"llm_share_volume": 0.0,	10	"llm_share_volume": 0.0,
11	"n_llm_decisions": 0,	11	"n_llm_decisions": 0,
12	− "n_trades": 10295,	12	+ "n_trades": 9483,
13	− "wall_s": 0.5701620420441031,	13	+ "wall_s": 0.561950541974511,
14	− "msd_log": 0.00019286955904584808,	14	+ "msd_log": 4.266113421084894e-05,
15	− "mad_pct": 0.7310450811973923,	15	+ "mad_pct": 0.4452912373334502,
16	− "bias_pct": -0.15575191205753852,	16	+ "bias_pct": -0.16250645419823648,
17	− "vr5": 1.0718532448575135,	17	+ "vr5": 1.2808238242517882,
18	− "vr10": 1.2475793867974083,	18	+ "vr10": 1.6956799240598435,
19	− "vol_bp": 32.35812080767805,	19	+ "vol_bp": 13.060749029694724,
20	− "kurtosis": 10.234149771580848,	20	+ "kurtosis": 35.01507543486031,
21	− "acf_absr_1_10": 2.725430443480742,	21	+ "acf_absr_1_10": 3.1656403290576316,
22	− "acf_r_1": 0.040723476211249214,	22	+ "acf_r_1": 0.00014962429838089104,
23	− "spread_bp": 28.878447094188143,	23	+ "spread_bp": 18.99330779045911,
24	− "eff_spread_bp": 41.4787058741681,	24	+ "eff_spread_bp": 24.641460001660576,
25	− "depth5": 132.712,	25	+ "depth5": 140.702,
26	− "lambda_bp_100": 20.621638192405012,	26	+ "lambda_bp_100": 12.598283607509806,
27	− "resiliency": 0.6471442312753932,	27	+ "resiliency": 0.6094844467069042,
28	− "n_bubbles": 1,	28	+ "n_bubbles": 0,
29	− "bubble_amp": 0.080067123441383,	29	+ "bubble_amp": 0.0,
30	− "bubble_dur": 13.0,	30	+ "bubble_dur": 0.0,
31	− "max_dd": 0.05210182210339487,	31	+ "max_dd": 0.028555868633532988,
32	"n_crashes": 0,	32	"n_crashes": 0,
33	− "volume": 57653.0,	33	+ "volume": 49533.0,
34	"n_halts": 0,	34	"n_halts": 0,
35	"share_ticks_halted": 0.0,	35	"share_ticks_halted": 0.0,
36	− "time_to_absorb": 24.0,	36	+ "time_to_absorb": 15.0,
37	− "response_ratio": 1.0602789303452547,	37	+ "response_ratio": 2.2490023898583673,
38	− "pnl_fundamentalist": 17.656477071443046,	38	+ "pnl_fundamentalist": 11.082209219493947,
39	− "pnl_leveraged": 8.519706421361073,	39	+ "pnl_leveraged": 1.4418651763278723,
40	− "pnl_market_maker": 9.050950394737617,	40	+ "pnl_market_maker": 5.710438916819656,
41	− "pnl_meanrev": 4.209786017783919,	41	+ "pnl_meanrev": 2.9877031781727115,
42	− "pnl_noise": 7.296227416578166,	42	+ "pnl_noise": 4.649299724803906,
43	− "pnl_trend": 8.888541562040968,	43	+ "pnl_trend": 2.136350462390413,
44	− "gini_final": 0.27427592144984314,	44	+ "gini_final": 0.27703259343488695,
45	"gini_initial": 0.2804114796674862,	45	"gini_initial": 0.2804114796674862,
46	"n_bankrupt": 0,	46	"n_bankrupt": 0,
47	− "cancel_ratio": 0.15356339084771192,	47	+ "cancel_ratio": 0.18065779169929522,
48	− "reject_ratio": 0.010352588147036759,	48	+ "reject_ratio": 0.00822239624119029,
49	− "fill_ratio": 1.54463615903976,	49	+ "fill_ratio": 1.4851996867658575,
50	− "turnover": 5.643414799728445,	50	+ "turnover": 4.935081466395112,
51	− "mm_pnl": 9.050950394737619,	51	+ "mm_pnl": 5.710438916819655,
52	− "mm_inventory_abs": 447.8,	52	+ "mm_inventory_abs": 450.4,
53	− "n_orders": 13330	53	+ "flow_corr_fundamentalist": 0.6081052996058153,
		54	+ "flow_corr_leveraged": 0.8056656086005239,
		55	+ "flow_corr_market_maker": 0.40285902309116894,
		56	+ "flow_corr_meanrev": 0.5240214666938758,
		57	+ "flow_corr_noise": 0.0021860282452096148,
		58	+ "flow_corr_trend": 0.5675547519775062,
		59	+ "n_orders": 12770
54	},	60	},
55	{	61	{
56	"name": "scale_100",	62	"name": "scale_100",
57	"seed": 2,	63	"seed": 2,
58	− "config_hash": "f50aac64012c79cc",	64	+ "config_hash": "2220c87cfabc484c",
59	"n_traders": 100,	65	"n_traders": 100,
60	"llm_share": 0.0,	66	"llm_share": 0.0,
61	"llm_share_traders": 0.0,	67	"llm_share_traders": 0.0,
62	"llm_share_orders": 0.0,	68	"llm_share_orders": 0.0,
63	"llm_share_volume": 0.0,	69	"llm_share_volume": 0.0,
64	"n_llm_decisions": 0,	70	"n_llm_decisions": 0,
65	− "n_trades": 10527,	71	+ "n_trades": 10135,
66	− "wall_s": 0.5802934590028599,	72	+ "wall_s": 0.5950057920417748,
67	− "msd_log": 0.0002888292585473304,	73	+ "msd_log": 0.00011269807156550713,
68	− "mad_pct": 0.9452873073912066,	74	+ "mad_pct": 0.6999329907345704,
69	− "bias_pct": -0.29914538754731906,	75	+ "bias_pct": -0.1834133029306038,
70	− "vr5": 1.5658770854218884,	76	+ "vr5": 1.1206839869796097,
71	− "vr10": 2.2323861312360114,	77	+ "vr10": 1.2032701908196335,
72	− "vol_bp": 19.93022805615968,	78	+ "vol_bp": 21.182447653590852,
73	− "kurtosis": 14.340120252000448,	79	+ "kurtosis": 63.393307694350604,
74	− "acf_absr_1_10": 2.5967756236642745,	80	+ "acf_absr_1_10": 2.1232276887652786,
75	− "acf_r_1": 0.20815586356877122,	81	+ "acf_r_1": -0.12168840912844274,
76	− "spread_bp": 24.364181008850423,	82	+ "spread_bp": 24.508671511490352,
77	− "eff_spread_bp": 34.16487033612486,	83	+ "eff_spread_bp": 32.80287511527904,
78	− "depth5": 146.782,	84	+ "depth5": 144.488,
79	− "lambda_bp_100": 12.733686146838773,	85	+ "lambda_bp_100": 10.368067778921336,
80	− "resiliency": 0.7132913644500204,	86	+ "resiliency": 0.8008422153703116,
81	− "n_bubbles": 1,	87	+ "n_bubbles": 0,
82	− "bubble_amp": 0.07461228907838036,	88	+ "bubble_amp": 0.0,
83	− "bubble_dur": 22.0,	89	+ "bubble_dur": 0.0,
84	− "max_dd": 0.04207449700924415,	90	+ "max_dd": 0.0463212015900597,
85	"n_crashes": 0,	91	"n_crashes": 0,
86	− "volume": 61390.0,	92	+ "volume": 57065.0,
87	"n_halts": 0,	93	"n_halts": 0,
88	"share_ticks_halted": 0.0,	94	"share_ticks_halted": 0.0,
89	− "time_to_absorb": 39.0,	95	+ "time_to_absorb": 42.0,
90	− "response_ratio": 0.31568646155126934,	96	+ "response_ratio": 0.07229322625373641,
91	− "pnl_fundamentalist": 12.858237774095874,	97	+ "pnl_fundamentalist": 9.728163515969941,
92	− "pnl_leveraged": 8.307969157398325,	98	+ "pnl_leveraged": -0.5134044434277812,
93	− "pnl_market_maker": 7.757582637283514,	99	+ "pnl_market_maker": 5.311276871706401,
94	− "pnl_meanrev": 4.448602248693099,	100	+ "pnl_meanrev": 3.8280035905591077,
95	− "pnl_noise": 6.090876220549884,	101	+ "pnl_noise": 4.225801084530292,
96	− "pnl_trend": 6.275541672164394,	102	+ "pnl_trend": -2.012191826388961,
97	− "gini_final": 0.3192748996527118,	103	+ "gini_final": 0.31913008923115943,
98	"gini_initial": 0.3167311413076369,	104	"gini_initial": 0.3167311413076369,
99	"n_bankrupt": 0,	105	"n_bankrupt": 0,
100	− "cancel_ratio": 0.17202363793600461,	106	+ "cancel_ratio": 0.17056931248597293,
101	− "reject_ratio": 0.011746901124243299,	107	+ "reject_ratio": 0.011520909702999925,
102	− "fill_ratio": 1.5172960507350821,	108	+ "fill_ratio": 1.5164210368818734,
103	− "turnover": 5.789452166802944,	109	+ "turnover": 5.409157808667212,
104	− "mm_pnl": 7.757582637283514,	110	+ "mm_pnl": 5.311276871706401,
105	− "mm_inventory_abs": 537.6,	111	+ "mm_inventory_abs": 561.8,
106	− "n_orders": 13876	112	+ "flow_corr_fundamentalist": 0.6169346471443932,
		113	+ "flow_corr_leveraged": 0.8242786988340965,
		114	+ "flow_corr_market_maker": 0.333751587227477,
		115	+ "flow_corr_meanrev": 0.71243506190734,
		116	+ "flow_corr_noise": 0.03476489760506395,
		117	+ "flow_corr_trend": 0.8565784759401733,
		118	+ "n_orders": 13367
107	},	119	},
108	{	120	{
109	"name": "scale_100",	121	"name": "scale_100",
110	"seed": 3,	122	"seed": 3,
111	− "config_hash": "3d20d4f33dca1657",	123	+ "config_hash": "1652fd3ac20740c7",
112	"n_traders": 100,	124	"n_traders": 100,
113	"llm_share": 0.0,	125	"llm_share": 0.0,
114	"llm_share_traders": 0.0,	126	"llm_share_traders": 0.0,
115	"llm_share_orders": 0.0,	127	"llm_share_orders": 0.0,
116	"llm_share_volume": 0.0,	128	"llm_share_volume": 0.0,
117	"n_llm_decisions": 0,	129	"n_llm_decisions": 0,
118	− "n_trades": 10741,	130	+ "n_trades": 10966,
119	− "wall_s": 0.581291499955114,	131	+ "wall_s": 0.5947290000040084,
120	− "msd_log": 0.0001581647166691645,	132	+ "msd_log": 7.309845091058393e-05,
121	− "mad_pct": 0.7940933085313311,	133	+ "mad_pct": 0.7194719580752089,
122	− "bias_pct": -0.1457811073728486,	134	+ "bias_pct": -0.006526704437844799,
123	− "vr5": 1.4253673844532706,	135	+ "vr5": 1.0296803187890688,
124	− "vr10": 2.1997897604423264,	136	+ "vr10": 1.4045586222425246,
125	− "vol_bp": 18.747558012139187,	137	+ "vol_bp": 22.217350658987264,
126	− "kurtosis": 6.081260113112632,	138	+ "kurtosis": 17.603931390316244,
127	− "acf_absr_1_10": 2.4869862608002076,	139	+ "acf_absr_1_10": 2.234583046044834,
128	− "acf_r_1": 0.08038707277238923,	140	+ "acf_r_1": 0.014418398132741888,
129	− "spread_bp": 23.159298170155786,	141	+ "spread_bp": 24.752507678158395,
130	− "eff_spread_bp": 32.62440165728167,	142	+ "eff_spread_bp": 37.87658823419071,
131	− "depth5": 145.45,	143	+ "depth5": 139.906,
132	− "lambda_bp_100": 9.286127919988914,	144	+ "lambda_bp_100": 13.897996957061956,
133	− "resiliency": 0.7155986676930891,	145	+ "resiliency": 0.532945437189439,
134	"n_bubbles": 0,	146	"n_bubbles": 0,
135	"bubble_amp": 0.0,	147	"bubble_amp": 0.0,
136	"bubble_dur": 0.0,	148	"bubble_dur": 0.0,
137	− "max_dd": 0.030664395229982988,	149	+ "max_dd": 0.028639909995842205,
138	"n_crashes": 0,	150	"n_crashes": 0,
139	− "volume": 62973.0,	151	+ "volume": 65743.0,
140	"n_halts": 0,	152	"n_halts": 0,
141	"share_ticks_halted": 0.0,	153	"share_ticks_halted": 0.0,
142	− "time_to_absorb": 26.0,	154	+ "time_to_absorb": 10.0,
143	− "response_ratio": 0.5264625844839756,	155	+ "response_ratio": 1.6271649201682974,
144	− "pnl_fundamentalist": 13.05203477140921,	156	+ "pnl_fundamentalist": 11.076389214723797,
145	− "pnl_leveraged": 7.137665260941994,	157	+ "pnl_leveraged": -0.4910219812423844,
146	− "pnl_market_maker": 9.437974794622459,	158	+ "pnl_market_maker": 6.7245953688727145,
147	− "pnl_meanrev": 8.854861916786305,	159	+ "pnl_meanrev": 6.267829548921491,
148	− "pnl_noise": 8.383088043861747,	160	+ "pnl_noise": 5.628234679518568,
149	− "pnl_trend": 6.794934289244118,	161	+ "pnl_trend": -1.7485030516832847,
150	− "gini_final": 0.28354147912478633,	162	+ "gini_final": 0.2858138745681142,
151	"gini_initial": 0.2820805035864161,	163	"gini_initial": 0.2820805035864161,
152	"n_bankrupt": 0,	164	"n_bankrupt": 0,
153	− "cancel_ratio": 0.15596595007935363,	165	+ "cancel_ratio": 0.15531763026409706,
154	− "reject_ratio": 0.009955273409320444,	166	+ "reject_ratio": 0.009778729478943612,
155	− "fill_ratio": 1.5497042273842159,	167	+ "fill_ratio": 1.5654532476802283,
156	− "turnover": 5.99405210689453,	168	+ "turnover": 6.273016670855323,
157	− "mm_pnl": 9.437974794622459,	169	+ "mm_pnl": 6.7245953688727145,
158	− "mm_inventory_abs": 508.8,	170	+ "mm_inventory_abs": 471.4,
159	− "n_orders": 13862	171	+ "flow_corr_fundamentalist": 0.6119780392329717,
		172	+ "flow_corr_leveraged": 0.7727630449923826,
		173	+ "flow_corr_market_maker": 0.7384719894286295,
		174	+ "flow_corr_meanrev": 0.6194062558718657,
		175	+ "flow_corr_noise": 0.06085899518277767,
		176	+ "flow_corr_trend": 0.6689081805252811,
		177	+ "n_orders": 14010
160	},	178	},
161	{	179	{
162	"name": "scale_1000",	180	"name": "scale_1000",
163	"seed": 1,	181	"seed": 1,
164	− "config_hash": "ee14f7c7fdd0c799",	182	+ "config_hash": "274ecfd52b3464b5",
165	"n_traders": 1000,	183	"n_traders": 1000,
166	"llm_share": 0.0,	184	"llm_share": 0.0,
167	"llm_share_traders": 0.0,	185	"llm_share_traders": 0.0,
168	"llm_share_orders": 0.0,	186	"llm_share_orders": 0.0,
169	"llm_share_volume": 0.0,	187	"llm_share_volume": 0.0,
170	"n_llm_decisions": 0,	188	"n_llm_decisions": 0,
171	− "n_trades": 11682,	189	+ "n_trades": 11506,
172	− "wall_s": 2.670115082990378,	190	+ "wall_s": 2.6422452499973588,
173	− "msd_log": 0.00018062420246409605,	191	+ "msd_log": 4.459799490563131e-05,
174	− "mad_pct": 0.8500964158284774,	192	+ "mad_pct": 0.5614471743278486,
175	− "bias_pct": -0.18426548373374133,	193	+ "bias_pct": -0.15511334576246724,
176	− "vr5": 2.025387100878808,	194	+ "vr5": 1.1392675588226546,
177	− "vr10": 3.2959357937833555,	195	+ "vr10": 1.6692572723703871,
178	− "vol_bp": 13.822478939583142,	196	+ "vol_bp": 8.726812744253099,
179	− "kurtosis": 8.487363422007448,	197	+ "kurtosis": 8.700730924357568,
180	− "acf_absr_1_10": 3.676658722746514,	198	+ "acf_absr_1_10": 1.9389812585059079,
181	− "acf_r_1": 0.287550868207432,	199	+ "acf_r_1": -0.010783036772180954,
182	− "spread_bp": 13.824129777426267,	200	+ "spread_bp": 13.29861474159063,
183	− "eff_spread_bp": 20.735460509724028,	201	+ "eff_spread_bp": 18.843467334285577,
184	− "depth5": 221.702,	202	+ "depth5": 205.852,
185	− "lambda_bp_100": 4.820764798328324,	203	+ "lambda_bp_100": 4.340416339061836,
186	− "resiliency": 0.7153936557542477,	204	+ "resiliency": 0.7325175655995373,
187	− "n_bubbles": 1,	205	+ "n_bubbles": 0,
188	− "bubble_amp": 0.07026257865057628,	206	+ "bubble_amp": 0.0,
189	− "bubble_dur": 11.0,	207	+ "bubble_dur": 0.0,
190	− "max_dd": 0.06061815440556961,	208	+ "max_dd": 0.03116660312619135,
191	"n_crashes": 0,	209	"n_crashes": 0,
192	− "volume": 71448.0,	210	+ "volume": 67420.0,
193	"n_halts": 0,	211	"n_halts": 0,
194	"share_ticks_halted": 0.0,	212	"share_ticks_halted": 0.0,
195	− "time_to_absorb": 24.0,	213	+ "time_to_absorb": 0.0,
196	− "response_ratio": 0.6796255487489661,	214	+ "response_ratio": 0.404975000785493,
197	− "pnl_fundamentalist": 12.22118924232089,	215	+ "pnl_fundamentalist": 6.729056031841567,
198	− "pnl_leveraged": 8.85051472713129,	216	+ "pnl_leveraged": 5.637427313317174,
199	− "pnl_market_maker": 8.39873997810738,	217	+ "pnl_market_maker": 5.313848759240008,
200	− "pnl_meanrev": 4.758693065304863,	218	+ "pnl_meanrev": 4.035185583879716,
201	− "pnl_noise": 7.802552958684208,	219	+ "pnl_noise": 4.970844080033762,
202	− "pnl_trend": 8.986276033348616,	220	+ "pnl_trend": 5.916862240206084,
203	− "gini_final": 0.2858965594076205,	221	+ "gini_final": 0.2854617843595489,
204	"gini_initial": 0.2855288797803193,	222	"gini_initial": 0.2855288797803193,
205	"n_bankrupt": 0,	223	"n_bankrupt": 0,
206	− "cancel_ratio": 0.0687326986699075,	224	+ "cancel_ratio": 0.06602917568227822,
207	− "reject_ratio": 0.006886773344136115,	225	+ "reject_ratio": 0.003280519299225239,
208	− "fill_ratio": 1.5774762001215312,	226	+ "fill_ratio": 1.6061980875270467,
209	− "turnover": 0.6852169116725306,	227	+ "turnover": 0.6522142457190135,
210	− "mm_pnl": 8.398739978107383,	228	+ "mm_pnl": 5.313848759240008,
211	− "mm_inventory_abs": 493.8,	229	+ "mm_inventory_abs": 493.4,
212	− "n_orders": 14811	230	+ "flow_corr_fundamentalist": 0.1812877758345397,
		231	+ "flow_corr_leveraged": 0.24358914780429702,
		232	+ "flow_corr_market_maker": 0.16855157430290751,
		233	+ "flow_corr_meanrev": 0.20193002732113746,
		234	+ "flow_corr_noise": 0.007352094582687515,
		235	+ "flow_corr_trend": 0.21713676628286416,
		236	+ "n_orders": 14327
213	},	237	},
214	{	238	{
215	"name": "scale_1000",	239	"name": "scale_1000",
216	"seed": 2,	240	"seed": 2,
217	− "config_hash": "b16091b8bb39498f",	241	+ "config_hash": "39b2c173ae4664a2",
218	"n_traders": 1000,	242	"n_traders": 1000,
219	"llm_share": 0.0,	243	"llm_share": 0.0,
220	"llm_share_traders": 0.0,	244	"llm_share_traders": 0.0,
221	"llm_share_orders": 0.0,	245	"llm_share_orders": 0.0,
222	"llm_share_volume": 0.0,	246	"llm_share_volume": 0.0,
223	"n_llm_decisions": 0,	247	"n_llm_decisions": 0,
224	− "n_trades": 11629,	248	+ "n_trades": 11341,
225	− "wall_s": 2.6595235830172896,	249	+ "wall_s": 2.593718208023347,
226	− "msd_log": 0.00041330390214059054,	250	+ "msd_log": 9.380893708595955e-05,
227	− "mad_pct": 1.1917097910197463,	251	+ "mad_pct": 0.7030094050861265,
228	− "bias_pct": -0.15970070791396937,	252	+ "bias_pct": -0.04239496346757861,
229	− "vr5": 1.644328760768095,	253	+ "vr5": 1.4143997446399388,
230	− "vr10": 2.756869312033638,	254	+ "vr10": 2.04652744673238,
231	− "vol_bp": 12.118437319273784,	255	+ "vol_bp": 10.246104277739805,
232	− "kurtosis": 14.604102013472062,	256	+ "kurtosis": 7.874784440120156,
233	− "acf_absr_1_10": 2.6836877474394876,	257	+ "acf_absr_1_10": 2.2459292375683555,
234	− "acf_r_1": 0.02149103385973521,	258	+ "acf_r_1": 0.12341838139966418,
235	− "spread_bp": 12.264625299683127,	259	+ "spread_bp": 14.86376488201993,
236	− "eff_spread_bp": 19.045456466147648,	260	+ "eff_spread_bp": 20.79134773830736,
237	− "depth5": 243.088,	261	+ "depth5": 248.952,
238	− "lambda_bp_100": 5.2096578366086845,	262	+ "lambda_bp_100": 5.1373327775436906,
239	− "resiliency": 0.5268311406120124,	263	+ "resiliency": 0.809942287932077,
240	− "n_bubbles": 1,	264	+ "n_bubbles": 0,
241	− "bubble_amp": 0.07560512002649855,	265	+ "bubble_amp": 0.0,
242	− "bubble_dur": 33.0,	266	+ "bubble_dur": 0.0,
243	− "max_dd": 0.026409804187581987,	267	+ "max_dd": 0.030708092485549177,
244	"n_crashes": 0,	268	"n_crashes": 0,
245	− "volume": 72963.0,	269	+ "volume": 70170.0,
246	"n_halts": 0,	270	"n_halts": 0,
247	"share_ticks_halted": 0.0,	271	"share_ticks_halted": 0.0,
248	− "time_to_absorb": 55.0,	272	+ "time_to_absorb": 40.0,
249	− "response_ratio": 0.23023340213135676,	273	+ "response_ratio": 0.07301615851627377,
250	− "pnl_fundamentalist": 9.410087991005573,	274	+ "pnl_fundamentalist": 5.242388952904068,
251	− "pnl_leveraged": 8.06978747113479,	275	+ "pnl_leveraged": 4.122740890775753,
252	− "pnl_market_maker": 8.58296964348504,	276	+ "pnl_market_maker": 4.750528108764787,
253	− "pnl_meanrev": 7.3927831227808385,	277	+ "pnl_meanrev": 4.3948817690041855,
254	− "pnl_noise": 8.868638893094134,	278	+ "pnl_noise": 4.8087790856789905,
255	− "pnl_trend": 9.12795706441179,	279	+ "pnl_trend": 4.207353638127365,
256	− "gini_final": 0.2822291008133873,	280	+ "gini_final": 0.28185452077762535,
257	"gini_initial": 0.2821557785325072,	281	"gini_initial": 0.2821557785325072,
258	"n_bankrupt": 0,	282	"n_bankrupt": 0,
259	− "cancel_ratio": 0.07004048582995952,	283	+ "cancel_ratio": 0.06047103755569701,
260	− "reject_ratio": 0.011201079622132253,	284	+ "reject_ratio": 0.005799561496569771,
261	− "fill_ratio": 1.569365721997301,	285	+ "fill_ratio": 1.6042152910389702,
262	− "turnover": 0.6991975582965093,	286	+ "turnover": 0.6682737860658555,
263	− "mm_pnl": 8.58296964348504,	287	+ "mm_pnl": 4.750528108764787,
264	− "mm_inventory_abs": 483.22,	288	+ "mm_inventory_abs": 485.56,
265	− "n_orders": 14820	289	+ "flow_corr_fundamentalist": 0.18370073935220338,
		290	+ "flow_corr_leveraged": 0.19047089945265988,
		291	+ "flow_corr_market_maker": 0.1967491510134189,
		292	+ "flow_corr_meanrev": 0.2217762315213713,
		293	+ "flow_corr_noise": 0.0088854995259112,
		294	+ "flow_corr_trend": 0.2426905842417079,
		295	+ "n_orders": 14139
266	},	296	},
267	{	297	{
268	"name": "scale_1000",	298	"name": "scale_1000",
269	"seed": 3,	299	"seed": 3,
270	− "config_hash": "bb000b72ba320264",	300	+ "config_hash": "fd07f4872e9678a6",
271	"n_traders": 1000,	301	"n_traders": 1000,
272	"llm_share": 0.0,	302	"llm_share": 0.0,
273	"llm_share_traders": 0.0,	303	"llm_share_traders": 0.0,
274	"llm_share_orders": 0.0,	304	"llm_share_orders": 0.0,
275	"llm_share_volume": 0.0,	305	"llm_share_volume": 0.0,
276	"n_llm_decisions": 0,	306	"n_llm_decisions": 0,
277	− "n_trades": 12209,	307	+ "n_trades": 11025,
278	− "wall_s": 2.7417157079908065,	308	+ "wall_s": 2.5898144589737058,
279	− "msd_log": 0.00021092782536451138,	309	+ "msd_log": 4.3915838389987096e-05,
280	− "mad_pct": 0.9236328359261194,	310	+ "mad_pct": 0.5188634604617772,
281	− "bias_pct": -0.17586691384915412,	311	+ "bias_pct": -0.046929456624928534,
282	− "vr5": 1.611910893920555,	312	+ "vr5": 1.3221448319933091,
283	− "vr10": 2.6070797852167216,	313	+ "vr10": 1.7858937955551464,
284	− "vol_bp": 13.820990551125412,	314	+ "vol_bp": 10.451851715820833,
285	− "kurtosis": 4.406283908221623,	315	+ "kurtosis": 23.617737355507195,
286	− "acf_absr_1_10": 1.749659430254234,	316	+ "acf_absr_1_10": 3.3090798799738983,
287	− "acf_r_1": 0.14228872111193444,	317	+ "acf_r_1": -0.02597027489657522,
288	− "spread_bp": 16.402115035292773,	318	+ "spread_bp": 13.389367953115482,
289	− "eff_spread_bp": 24.921946560889914,	319	+ "eff_spread_bp": 19.168776766750792,
290	− "depth5": 178.83,	320	+ "depth5": 209.732,
291	− "lambda_bp_100": 6.947876984339655,	321	+ "lambda_bp_100": 5.675299684983716,
292	− "resiliency": 0.5644120216956088,	322	+ "resiliency": 0.8243452955880805,
293	− "n_bubbles": 1,	323	+ "n_bubbles": 0,
294	− "bubble_amp": 0.06792151254244383,	324	+ "bubble_amp": 0.0,
295	− "bubble_dur": 11.0,	325	+ "bubble_dur": 0.0,
296	− "max_dd": 0.03714749884823354,	326	+ "max_dd": 0.03415154749199578,
297	"n_crashes": 0,	327	"n_crashes": 0,
298	− "volume": 77495.0,	328	+ "volume": 64307.0,
299	"n_halts": 0,	329	"n_halts": 0,
300	"share_ticks_halted": 0.0,	330	"share_ticks_halted": 0.0,
301	− "time_to_absorb": 38.0,	331	+ "time_to_absorb": 2.0,
302	− "response_ratio": 0.38870487487733535,	332	+ "response_ratio": 0.9422059728212616,
303	− "pnl_fundamentalist": 11.763531925432737,	333	+ "pnl_fundamentalist": 8.163368835474088,
304	− "pnl_leveraged": 12.30986340715177,	334	+ "pnl_leveraged": 7.482149977000345,
305	− "pnl_market_maker": 9.441394250152205,	335	+ "pnl_market_maker": 6.467759551962172,
306	− "pnl_meanrev": 7.062613962399257,	336	+ "pnl_meanrev": 3.909678059648282,
307	− "pnl_noise": 8.798699442629827,	337	+ "pnl_noise": 6.052221963341724,
308	− "pnl_trend": 11.794205881549368,	338	+ "pnl_trend": 7.145788314430566,
309	− "gini_final": 0.2887824703843839,	339	+ "gini_final": 0.2883606405817487,
310	"gini_initial": 0.28816811800295694,	340	"gini_initial": 0.28816811800295694,
311	"n_bankrupt": 0,	341	"n_bankrupt": 0,
312	− "cancel_ratio": 0.06202264946010008,	342	+ "cancel_ratio": 0.06500290191526407,
313	− "reject_ratio": 0.011324730050039505,	343	+ "reject_ratio": 0.003990133488102147,
314	− "fill_ratio": 1.607716618382934,	344	+ "fill_ratio": 1.599680789320952,
315	− "turnover": 0.7320112276259568,	345	+ "turnover": 0.6165365938981018,
316	− "mm_pnl": 9.441394250152207,	346	+ "mm_pnl": 6.467759551962171,
317	− "mm_inventory_abs": 505.64,	347	+ "mm_inventory_abs": 500.08,
318	− "n_orders": 15188	348	+ "flow_corr_fundamentalist": 0.21833213263619425,
		349	+ "flow_corr_leveraged": 0.27583220335170927,
		350	+ "flow_corr_market_maker": 0.1479981781996154,
		351	+ "flow_corr_meanrev": 0.19060490917691542,
		352	+ "flow_corr_noise": 0.008272326182851597,
		353	+ "flow_corr_trend": 0.312039965563122,
		354	+ "n_orders": 13784
319	},	355	},
320	{	356	{
321	"name": "scale_10000",	357	"name": "scale_10000",
322	"seed": 1,	358	"seed": 1,
323	− "config_hash": "b20b7ea1893420e2",	359	+ "config_hash": "cf1e0784983bcede",
324	"n_traders": 10000,	360	"n_traders": 10000,
325	"llm_share": 0.0,	361	"llm_share": 0.0,
326	"llm_share_traders": 0.0,	362	"llm_share_traders": 0.0,
327	"llm_share_orders": 0.0,	363	"llm_share_orders": 0.0,
328	"llm_share_volume": 0.0,	364	"llm_share_volume": 0.0,
329	"n_llm_decisions": 0,	365	"n_llm_decisions": 0,
330	− "n_trades": 11885,	366	+ "n_trades": 11123,
331	− "wall_s": 23.677518000011332,	367	+ "wall_s": 23.737372540985234,
332	− "msd_log": 0.0003067595731653351,	368	+ "msd_log": 7.441850757126039e-05,
333	− "mad_pct": 1.2324887935614917,	369	+ "mad_pct": 0.749043336362881,
334	− "bias_pct": -0.005074964908155267,	370	+ "bias_pct": 0.10881212943287366,
335	− "vr5": 1.5681375271456242,	371	+ "vr5": 1.1739064526952574,
336	− "vr10": 2.5294578246021526,	372	+ "vr10": 1.6744417894256225,
337	− "vol_bp": 10.572758594723124,	373	+ "vol_bp": 5.5793268470698685,
338	− "kurtosis": 4.752405591681221,	374	+ "kurtosis": 20.562572058837862,
339	− "acf_absr_1_10": 3.335881681903191,	375	+ "acf_absr_1_10": 3.5598949849067014,
340	− "acf_r_1": 0.09791372200435924,	376	+ "acf_r_1": -0.1448973966918179,
341	− "spread_bp": 10.897921046050657,	377	+ "spread_bp": 9.536466029426055,
342	− "eff_spread_bp": 18.2037877380357,	378	+ "eff_spread_bp": 11.7466253784327,
343	− "depth5": 455.138,	379	+ "depth5": 397.06,
344	− "lambda_bp_100": 2.6356634482239287,	380	+ "lambda_bp_100": 1.8309625750286647,
345	− "resiliency": 0.379668239121855,	381	+ "resiliency": 1.3502975380705582,
346	− "n_bubbles": 1,	382	+ "n_bubbles": 0,
347	− "bubble_amp": 0.06949448067157304,	383	+ "bubble_amp": 0.0,
348	− "bubble_dur": 16.0,	384	+ "bubble_dur": 0.0,
349	− "max_dd": 0.04199416623989671,	385	+ "max_dd": 0.01687673064069517,
350	"n_crashes": 0,	386	"n_crashes": 0,
351	− "volume": 84806.0,	387	+ "volume": 66327.0,
352	"n_halts": 0,	388	"n_halts": 0,
353	"share_ticks_halted": 0.0,	389	"share_ticks_halted": 0.0,
354	− "time_to_absorb": 48.0,	390	+ "time_to_absorb": 9.0,
355	− "response_ratio": 0.3440702894130496,	391	+ "response_ratio": 0.7125705166492834,
356	− "pnl_fundamentalist": 9.880501047388295,	392	+ "pnl_fundamentalist": 5.955015700660537,
357	− "pnl_leveraged": 9.530656101327683,	393	+ "pnl_leveraged": 5.907099944853895,
358	− "pnl_market_maker": 9.1641097345847,	394	+ "pnl_market_maker": 5.6798835853738945,
359	− "pnl_meanrev": 8.715541449604128,	395	+ "pnl_meanrev": 5.470757526892383,
360	− "pnl_noise": 9.212122901393542,	396	+ "pnl_noise": 5.705169840108752,
361	− "pnl_trend": 9.50732656220052,	397	+ "pnl_trend": 5.863998450226706,
362	− "gini_final": 0.28254323570340967,	398	+ "gini_final": 0.282549473263338,
363	"gini_initial": 0.282654480272589,	399	"gini_initial": 0.282654480272589,
364	"n_bankrupt": 0,	400	"n_bankrupt": 0,
365	− "cancel_ratio": 0.01489140330342047,	401	+ "cancel_ratio": 0.014714898835070508,
366	− "reject_ratio": 0.020548835999479775,	402	+ "reject_ratio": 0.005177464404932216,
367	− "fill_ratio": 1.5457146573026401,	403	+ "fill_ratio": 1.5154983309489747,
368	− "turnover": 0.08007531890809352,	404	+ "turnover": 0.06402693262156016,
369	− "mm_pnl": 9.1641097345847,	405	+ "mm_pnl": 5.6798835853738945,
370	− "mm_inventory_abs": 480.866,	406	+ "mm_inventory_abs": 487.6,
371	− "n_orders": 15378	407	+ "flow_corr_fundamentalist": 0.05825003390169506,
		408	+ "flow_corr_leveraged": 0.18770916233743581,
		409	+ "flow_corr_market_maker": 0.036910171635462584,
		410	+ "flow_corr_meanrev": 0.06386890242525427,
		411	+ "flow_corr_noise": 0.0017863715797973461,
		412	+ "flow_corr_trend": 0.17703015821496784,
		413	+ "n_orders": 14679
372	}	414	}
373	]	415	]
	\ No newline at end of file
374		416

				@@ -8,7 +8,7 @@ Version 1.0, 2026-10-27. Each row states the assumption as implemented in `src
8	\|---\|---\|---\|---\|---\|	8	\|---\|---\|---\|---\|---\|
9	\| A1 \| Discrete model time: 600 ticks, outcomes on t ≥ 100; no calendar mapping \| Defines the horizon of every metric \| Avoids false precision; 30 dividend periods suffice for ACF/VR estimates (≥ 500 returns) \| `n_ticks` 1,000 on the control (free); report whether rankings of cells change \|	9	\| A1 \| Discrete model time: 600 ticks, outcomes on t ≥ 100; no calendar mapping \| Defines the horizon of every metric \| Avoids false precision; 30 dividend periods suffice for ACF/VR estimates (≥ 500 returns) \| `n_ticks` 1,000 on the control (free); report whether rankings of cells change \|
10	\| A2 \| Dividend every 20 ticks: D = μ + x + J, x_t = 0.98 x_{t−1} + 2 ε_t (cents), μ = 50 c, r = 0.25 % per period; one asset by default \| Fundamental process; V ≈ 200.00 and P/D ≈ 400 \| Persistent, stationary cash-flow state with a closed-form PV (brief §Economy) \| Sweep ρ ∈ {0.95, 0.98}, σ ∈ {1, 2, 4} on the control (free) \|	10	\| A2 \| Dividend every 20 ticks: D = μ + x + J, x_t = 0.98 x_{t−1} + 2 ε_t (cents), μ = 50 c, r = 0.25 % per period; one asset by default \| Fundamental process; V ≈ 200.00 and P/D ≈ 400 \| Persistent, stationary cash-flow state with a closed-form PV (brief §Economy) \| Sweep ρ ∈ {0.95, 0.98}, σ ∈ {1, 2, 4} on the control (free) \|
11	−\| A3 \| "Efficient price" = full-information PV V_t = μ/r + x_t·ρ/(1 + r − ρ), risk-neutral, constant discount rate, no risk premium \| Benchmark for `msd_log`, `mad_pct`, bubbles, absorption \| The only unambiguous benchmark in a simulated economy; the latent state is never shown to agents \| Also report deviations from the public PV (`public_value`) and from the end-of-run mean price \|	11	+\| A3 \| "Efficient price" = full-information PV V_t = μ/r + x_n/(1 + r − ρ) with x_n the latent deviation of the upcoming dividend, risk-neutral, constant discount rate, no risk premium \| Benchmark for `msd_log`, `mad_pct`, bubbles, absorption \| The only unambiguous benchmark in a simulated economy; the latent state is never shown to agents \| Also report deviations from the public PV (`public_value`) and from the end-of-run mean price \|
12	\| A4 \| Public PV = Kalman/HMM posterior on announced dividends; in the baseline (no jumps, one regime) the dividend reveals x_t exactly, so V^pub ≡ V \| Information structure \| Simplest case for calibration; but it makes private signals redundant and H1 unidentified \| Confirmatory base uses noisy dividends (`jump_prob` 1, `jump_sigma` ≈ 4 c → SD(V^pub/V − 1) ≈ 0.5 %); both variants reported \|	12	\| A4 \| Public PV = Kalman/HMM posterior on announced dividends; in the baseline (no jumps, one regime) the dividend reveals x_t exactly, so V^pub ≡ V \| Information structure \| Simplest case for calibration; but it makes private signals redundant and H1 unidentified \| Confirmatory base uses noisy dividends (`jump_prob` 1, `jump_sigma` ≈ 4 c → SD(V^pub/V − 1) ≈ 0.5 %); both variants reported \|
13	\| A5 \| No look-ahead: private signals s = x_t + N(0, σ²/precision) are about the current state; news reports x_t with noise; nobody sees future draws \| Rules out mechanical information advantages \| Standard in noisy rational-expectations settings \| Unit test on the signal function; prompt audit (`prompting.py` builds only from `Observation`) \|	13	\| A5 \| No look-ahead: private signals s = x_t + N(0, σ²/precision) are about the current state; news reports x_t with noise; nobody sees future draws \| Rules out mechanical information advantages \| Standard in noisy rational-expectations settings \| Unit test on the signal function; prompt audit (`prompting.py` builds only from `Observation`) \|
14	\| A6 \| Shock: latent jump in x at t = 300 revealed through dividends/news; placebo at t = 450. Baseline size 40 c (+8.4 % on V); confirmatory size 10 c (+2.1 %) \| Event-study outcomes (`time_to_absorb`, `response_ratio`, local projections) \| A sub-threshold (< 5 %) jump so that a lagged response is not counted as a bubble; placebo for falsification \| Sizes 10/20/40 c on the control; `news_burst` variant \|	14	\| A6 \| Shock: latent jump in x at t = 300 revealed through dividends/news; placebo at t = 450. Baseline size 40 c (+8.4 % on V); confirmatory size 10 c (+2.1 %) \| Event-study outcomes (`time_to_absorb`, `response_ratio`, local projections) \| A sub-threshold (< 5 %) jump so that a lagged response is not counted as a bubble; placebo for falsification \| Sizes 10/20/40 c on the control; `news_burst` variant \|

spb/wp27_uqo

paper: model/implementation/limitations/appendices; docs consistency fixes

Theme