WP27 v1.0: confirmatory results, mechanisms, robustness, intro/abstract/conclusion, README, compiled paper (101 p.)

modified README.md +135 −2

++<!-- Author: Simon-Pierre Boucher — simon-pierre.boucher@uqo.ca -->
++# 🤖 The Invisible Hand of Machines
++**UQO Working Paper No. 27** — *Price Discovery in Markets Populated by LLM Agents*
++
++[![Paper](https://img.shields.io/badge/paper-PDF%20(101%20p.)-b31b1b?logo=latex&logoColor=white)](paper/main.pdf)
++[![Python](https://img.shields.io/badge/python-3.12-3776AB?logo=python&logoColor=white)](pyproject.toml)
++[![Data](https://img.shields.io/badge/data-synthetic%20%C2%B7%20611%20runs%20%C2%B7%2080k%20LLM%20decisions-4051b5)](results/)
++[![Reproducible](https://img.shields.io/badge/reproducible-make%20replicate-2e7d32)](Makefile)
++[![Institution](https://img.shields.io/badge/UQO-D%C3%A9pt.%20des%20sciences%20administratives-16365c)](https://uqo.ca)
++[![Author](https://img.shields.io/badge/author-Simon--Pierre%20Boucher-16365c)](mailto:simon-pierre.boucher@uqo.ca)
++
++> **TL;DR** — An event-sourced limit-order-book exchange with exact accounting and a latent dividend process (fundamental value ≠ public value ≠ price), seven calibrated classical trader types, and an adapter that lets a language model trade from a canonical snapshot through strictly validated JSON. Randomized matched-seed experiments assign 0–100 % of trader slots to `gpt-5.4-nano` (7 seeds per live cell, 30 per classical cell, pre-registered plan). A stylized correlated-error model predicts that a shared model degrades informativeness and amplifies spirals; the simulator confirms the coordination and diversity predictions and **rejects the price predictions**: spreads fall from 19 to 6 bp, depth ×7, impact −60 %, pricing error unchanged, no bubble or crash in 112 live runs, and LLM agents *dampen* leverage-driven fire sales — while latent information is absorbed more slowly and returns become predictable (VR(5) 1.1 → 2.1, ACF₁ 0.02 → 0.32). Mechanism: patient, public-value-anchored liquidity provision by agents that act infrequently.
++
++---
++
++## 📌 Status
++
++| Item | Value |
++|---|---|
++| Version | 1.0 (October 2026) |
++| SSRN | not yet submitted |
++| Journal | — (candidates: JEDC, Quantitative Finance, Journal of Financial Markets) |
++| Code & data | fully reproducible without an API key (`make replicate`); live cells need `OPENAI_API_KEY` (≈ 60 USD for the full design); Anthropic cells implemented but **not run** (key rejected, HTTP 401) |
++| Pre-registration | `docs/identification.md`, `docs/protocol.md`; every post-plan change in `docs/research_log.md` |
++
++## 🎯 What this paper does
++
++1. **Builds a sandbox exchange** (`src/market_agents/exchange`): continuous double auction, price–time priority, limit/market/cancel, partial fills, fees, per-agent latency, order TTL, price band, circuit breaker, position limit, periodic call auction, leverage/margin/forced liquidation, integer-cent accounting checked after every fill, append-only event log with deterministic replay (24 unit/property/integration tests).
++2. **Specifies the economy** (`assets`, `signals`): dividends every 20 ticks with a latent AR(1) deviation of the *next* dividend; latent value V = μ/r + x/(1+r−ρ); public value from a Kalman filter on announced dividends and (possibly misleading) news; private signals of known precision; macro, funding, liquidity and placebo shocks.
++3. **Populates it** (`agents`): fundamentalists, trend followers, mean-reverters, noise traders, bounded inventory-skew market makers, cross-asset arbitrageurs, leveraged traders — calibrated to stylized facts (ACF₁ ≈ 0, fat tails, volatility clustering, 10–30 bp spreads); 10,000 traders run in 24 s.
++4. **Adds LLM agents** (`llm`): OpenAI (Responses API, json_schema), Anthropic (Messages API, json_schema), mock; strict validation (invalid → hold, logged), bounded memory, cohort communication channel, adaptive feedback, prompt personas, USD/token caps, SQLite decision cache, frozen k-NN decision library for zero-cost scaling.
++5. **Derives the mechanisms** (`paper/sections/theory.tex`, proofs in the appendix, Monte Carlo checks in `scripts/90_theory_checks.py`): six propositions on informativeness vs shared-model share, order-flow correlation and synchrony, effective number of models, feedback amplification, leverage spirals, safeguards.
++6. **Runs randomized designs** (`configs/*.yaml`, `experiments/design.py`): common random numbers across matched seeds; LLM slots replace policy slots stratified by type and inherit their information set.
++
++## 🏆 Headline results (confirmatory design, 7 matched seeds per cell)
++
++| Outcome | 0 % LLM | 25 % | 95 % | Paired effect at 95 % |
++|---|---|---|---|---|
++| Quoted spread (bp) | 18.6 | 11.0 | 6.1 | −12.5 [−17.8, −7.3], d = −2.2 |
++| Depth, 5 levels (lots) | 141 | 201 | 956 | +816 |
++| Price impact (bp / 100 lots) | 9.3 | 5.2 | 3.6 | −5.7, Holm p = 0.01 |
++| Mean \|P/V − 1\| (%) | 0.73 | 0.50 | 0.54 | −0.18 [−0.67, +0.30] (n.s.) |
++| Variance ratio (5 ticks) | 1.14 | 1.55 | 2.12 | +0.97, d = 3.5 |
++| Return ACF₁ | 0.02 | 0.15 | 0.32 | +0.30, d = 3.5 |
++| Bubbles / crashes / halts | 0 | 0 | 0 | — |
++| LLM decision synchrony / pairwise flow corr. | — | 0.95 / 0.40 | 0.90 / 0.34 | corr. falls with share |
++| Leverage 19 %: tracking error (0 % vs 25 % LLM) | 2.16 % | 0.96 % | — | −1.2, p = 0.03; bubbles 0.7 → 0 |
++| Diverse prompts / two models at 25 %: pairwise corr. | 0.40 | 0.32 / 0.30 | — | absorb shock 4 ticks faster, spread +0.5 / +4.6 bp |
++| Communication channel at 25 %: pairwise corr. | 0.40 | 0.48 | — | tracking error +0.05 pp, d = 2.3 |
++
++Classical design (16 cells × 30 seeds): leverage and trend-heavy populations create bubbles and crashes; misleading public news raises pricing error; a position limit harms price discovery; circuit breaker inert at a 2 % shock; two-asset contagion through common funding (asset-1 drawdown 3.6 → 11.6 % with untouched fundamentals).
++
++## 🗂 Repository layout
++
++```
++├── CLAUDE.md                 # research protocol (the brief this repo executes)
++├── Makefile · pyproject.toml · uv.lock · .env.example
++├── configs/                  # baseline.yaml, classical_confirmatory, exploratory (mock), pilot_openai, confirmatory_openai
++├── src/market_agents/        # exchange/ assets/ signals/ agents/ llm/ experiments/ metrics/ analysis/ utils/ cli.py
++├── tests/                    # unit, property (hypothesis), integration
++├── scripts/                  # 10 calibration · 11 scale · 40 figures · 45 recompute · 46 thresholds · 47 mechanisms · 50 tables · 60 library · 90 theory checks · smoke_test
++├── docs/                     # research_plan, identification (pre-analysis plan), protocol, assumptions, model_cards, literature_ledger (143 verified refs), novelty_positioning, research_log
++├── results/                  # runs/<design>/<cell>/seed*/ (config, manifest, ticks/trades/agents/llm parquet), tables/, figures/, calibration/, library/
++└── paper/                    # main.tex, sections/, appendices/, tables/, figures/, references.bib → main.pdf
++```
++
++## 🚀 Quick start
++
++```bash
++uv sync --extra dev                      # Python 3.12 environment from uv.lock
++make test                                # 24 tests
++make smoke                               # 100 agents, fixed seed, zero paid calls, accounting verified
++make replicate                           # classical designs + theory checks + figures + tables (no API, ~5 min)
++cp .env.example .env                     # add OPENAI_API_KEY (and ANTHROPIC_API_KEY) for live cells
++make pilot                               # ≈ 4 USD ;  make confirmatory   # ≈ 60 USD, ~3 h with 5 workers
++make library                             # frozen-library replay at N = 1,000 (no API)
++make paper                               # latexmk → paper/main.pdf
++```
++
++Runtime on an Apple-silicon laptop: one 100-trader run 0.6 s, 1,000 traders 2.6 s, 10,000 traders 24 s; a 25 % LLM run ≈ 7 min (API-latency bound).
++
++## 🗃 Data
++
++All data are synthetic and generated by the simulator; nothing is downloaded. Every run directory holds `config.yaml`, `manifest.json` (config hash, git revision, seeds, providers/models, price assumptions, prompt version, realized LLM shares, usage), `ticks/trades/agents/llm_decisions.parquet` and `outcomes.json`. Live decisions (80,031 in the confirmatory design, 35,652 in the dose arm) are cached in `results/cache/llm_cache.sqlite` (not committed; regenerate with the designs). Total API spend for the reported runs: 56 USD (plus ≈ 14 USD on runs discarded after bug fixes, see `docs/research_log.md`).
++
++## 🔬 Methodology
++
++1. Common random numbers: one seed → named streams (fundamentals, news, slots, assignment, LLM slots, activation, tie-breaks, shuffle); matched pairs differ only in who decides (`utils/rng.py`, `experiments/population.py`).
++2. Outcomes at the run level (`metrics/outcomes.py`): pricing error, time to absorb a latent shock, response ratio, variance ratios, spreads, depth, Kyle λ, resiliency, bubbles/crashes at pre-registered thresholds, volatility clustering, contagion metrics, LLM validity/synchrony/pairwise correlation, cost.
++3. Inference (`analysis/estimators.py`): paired t with 95 % CI, Cohen's d_z, Wilcoxon, seed-FE OLS with cluster-robust SE, Holm within families, local projections, threshold sensitivity grid.
++4. Theory ↔ simulator mapping and H1–H6 in `paper/sections/theory.tex`; pre-analysis plan in `docs/identification.md`.
++
++## 🖼 Figures & tables inventory
++
++| Figure | Content |
++|---|---|
++| `fig_baseline_paths` | mid vs latent/public value, three classical economies |
++| `fig_classical_*` | classical mechanism cells (tracking error, absorption, drawdown, spread, impact, bubbles) |
++| `fig_contagion_*`, `fig_wealth_depth` | two-asset comovement/spillover; P&L distribution and depth |
++| `fig_dose_conf`, `fig_paths_conf`, `fig_realized_share_conf`, `fig_sync_conf`, `fig_lp_conf` | dose–response, price paths, realized dose, coordination, local projections |
++| `fig_arms_conf_*`, `fig_cost_quality_conf` | 25 % arms; cost–quality frontier |
++
++Tables (`paper/tables/*.tex`, generated by `scripts/50_tables.py`): stylized facts, scale, classical cells/effects, contagion, dose (pilot and confirmatory), arms cells/effects, LLM reliability, coordination, mechanism association, library, threshold sensitivity; theory checks in `results/tables/theory_checks.csv`.
++
++## ⚠️ Limitations
++
++- Seven matched seeds per live cell: effects below d ≈ 1.2 are not detectable; nulls on pricing error, absorption and drawdown are inconclusive for small effects.
++- One provider family (OpenAI nano/mini); Anthropic cells implemented but not run (key rejected on 2026-10-27).
++- LLM agents decide every 20 ticks from a fixed prompt with no tools; realized order share is 5–10× below the assigned slot share.
++- Partial-equilibrium sandbox; stylized model assumes market-order traders, the data point to limit-order liquidity providers.
++- No claim about live markets.
++
++## 📝 Citation
++
++```bibtex
++@techreport{boucher2026invisiblehand,
++  author      = {Boucher, Simon-Pierre},
++  title       = {The Invisible Hand of Machines: Price Discovery in Markets Populated by LLM Agents},
++  institution = {Universit\'e du Qu\'ebec en Outaouais, D\'epartement des sciences administratives},
++  type        = {Working Paper},
++  number      = {27},
++  year        = {2026},
++  month       = {October}
++}
++```
++
++## 👤 Author & contact
++
++**Simon-Pierre Boucher** — Département des sciences administratives, Université du Québec en Outaouais (UQO), Gatineau, QC.
++📧 [simon-pierre.boucher@uqo.ca](mailto:simon-pierre.boucher@uqo.ca)
++
++## 🗒 Notes
++
++- Built on 2026-10-27 from the protocol in `CLAUDE.md`; numbers 25–26 of the series are unassigned.
++- Deviations from `Recherche-UQO/HARMONISATION.md`: `pyproject.toml` + `uv.lock` instead of `requirements.txt`; results and cache directories are not committed (regenerable).

modified configs/confirmatory_openai.yaml +2 −0

				`@@ -28,3 +28,5 @@ cells:`
28	# --- E. fairness: equal USD ≈ 25 % nano vs 5 % mini (mini costs ~5x per decision); matched latency ---	28	# --- E. fairness: equal USD ≈ 25 % nano vs 5 % mini (mini costs ~5x per decision); matched latency ---
29	- {name: fair_mini_005, overrides: {population.llm_share: 0.05, population.llm.provider_mix: {openai_mini: 1.0}}, treatment: {arm: fairness, share: 0.05, model: gpt-5.4-mini, prompts: single}}	29	- {name: fair_mini_005, overrides: {population.llm_share: 0.05, population.llm.provider_mix: {openai_mini: 1.0}}, treatment: {arm: fairness, share: 0.05, model: gpt-5.4-mini, prompts: single}}
30	- {name: fair_latency_025, overrides: {population.llm_share: 0.25, population.latency.llm_latency: 0.05}, treatment: {arm: fairness, share: 0.25, llm_latency: 0.05}}	30	- {name: fair_latency_025, overrides: {population.llm_share: 0.25, population.latency.llm_latency: 0.05}, treatment: {arm: fairness, share: 0.25, llm_latency: 0.05}}
		31	+ # --- F. matched classical comparator for the leverage interaction (no API): same seeds, 0 % LLM, leveraged 19 % ---
		32	+ - {name: leverage_000, overrides: {population.llm_share: 0.00, population.mix: {fundamentalist: 0.25, trend: 0.06, meanrev: 0.10, noise: 0.35, market_maker: 0.05, statarb: 0.0, leveraged: 0.19}}, treatment: {arm: leverage, share: 0.00, leveraged: 0.19}}

added paper/figures/fig_arms_conf_llm_pairwise_corr.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_llm_synchrony.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_llm_valid_rate.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_mad_pct.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_max_dd.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_n_bubbles.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_spread_bp.pdf +0 −0

Binary file not shown.

added paper/figures/fig_arms_conf_time_to_absorb.pdf +0 −0

Binary file not shown.

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

Binary file not shown.

modified paper/figures/fig_classical_lambda_bp_100.pdf +0 −0

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

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

Binary file not shown.

added paper/figures/fig_cost_quality_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_dose_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_dose_pilot.pdf +0 −0

Binary file not shown.

added paper/figures/fig_lp_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_paths_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_paths_pilot.pdf +0 −0

Binary file not shown.

added paper/figures/fig_realized_share_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_realized_share_pilot.pdf +0 −0

Binary file not shown.

added paper/figures/fig_sync_conf.pdf +0 −0

Binary file not shown.

added paper/figures/fig_sync_pilot.pdf +0 −0

Binary file not shown.

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

Binary file not shown.

modified paper/main.pdf +0 −0

Binary file not shown.

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

+−Abstract to be written.
+ \ No newline at end of file
++What happens to price discovery when a growing share of the traders in a market are autonomous language-model agents? We build an event-sourced continuous double-auction exchange with exact accounting, a latent dividend process that separates fundamental value from public value and transaction price, seven calibrated classical trader types, and an adapter that lets a language model trade from a canonical snapshot through a strictly validated JSON action. We then run randomized, matched-seed experiments in which 0 to 100 percent of trader slots are exogenously assigned to a small reasoning model (gpt-5.4-nano), holding fundamentals, news, endowments, latencies and information sets fixed, with a pre-registered analysis plan, seven matched seeds per live cell and thirty per classical cell. A stylized model with correlated signal errors predicts that a shared model degrades informativeness, raises order-flow correlation and amplifies leverage-driven spirals, and that model diversity restores informativeness. The simulator confirms the predictions on coordination and diversity and rejects them on prices: raising the LLM share tightens quoted spreads from 19 to 6 basis points, multiplies depth by seven, cuts price impact by 60 percent and leaves the pricing error unchanged, while the market absorbs latent information more slowly and returns become predictable (variance ratio 2.1, lag-one autocorrelation 0.32 at full penetration). No bubble or crash appears in 112 live runs, and with leveraged traders present the LLM agents halve the pricing error and remove the bubble episodes the leverage creates. Prompt diversity and a two-model mix lower coordination and shorten absorption at a liquidity cost; a communication channel and memory raise coordination and pricing error slightly. The unifying mechanism is patient, public-value-anchored liquidity provision by agents who act infrequently. All results are causal statements about the simulator, reproducible from a one-command replication package; the Anthropic cells of the design are implemented but could not be run.

modified paper/sections/conclusion.tex +37 −1

++\section{Conclusion}
++\label{sec:conclusion}
++
++We set out to identify, inside a transparent exchange economy, the causal mechanisms through which
++autonomous language-model agents change price discovery as their share of the market grows. The
++instrument is an event-sourced limit-order-book simulator with exact accounting, a latent dividend
++process that keeps fundamental value, public value and price apart, calibrated classical traders, and a
++randomized, matched-seed design in which the LLM share is the treatment. The theory predicted that a
++shared model would degrade informativeness through correlated errors, raise coordination, amplify
++leverage spirals, and that diversity would help. The experiments confirmed the coordination and diversity
++predictions and reversed the price predictions.
++
++Three findings stand out. First, in this economy a small reasoning model trading from a canonical public
++snapshot is, overwhelmingly, a patient liquidity provider anchored on the public present value: spreads,
++impact and volatility fall and depth rises with its share, with no bubble or crash in any live run, and
++with leveraged traders present the LLM agents absorb the fire sales that destabilize the classical market.
++Second, the cost of that patience is slower incorporation of information that is not yet public and a
++price path that becomes predictable; the variance ratio doubles at full penetration. Third, coordination
++among LLM agents is high but not, in itself, harmful: the agents agree on the direction of the public
++value, prompt and model diversity reduce their agreement and speed up absorption at a liquidity cost, and
++a communication channel or memory raises agreement and pricing error by amounts that are precisely
++estimated but small.
++
++The honest summary of the pre-registered hypotheses is that H2 (synchrony), H4 (diversity) and H5
++(communication) are supported in sign, H1 (informativeness) is supported for dynamics and rejected for
++levels, H3 (instability) is rejected and its leverage corollary reversed, and H6 (safeguards) is not
++testable at the shock sizes used. The stylized model that organizes the paper is a model of
++market-order traders with correlated errors; the data call for a model of limit-order traders with a
++common anchor. Writing that model, giving the LLM agents leverage of their own, running the Claude cells
++that the design contains but a rejected key prevented, and raising the seed count to the level that
++detects half-standard-deviation effects are the next steps, and the replication package makes each of
++them a configuration change.
++
++Nothing here should be read as evidence about live markets. The claim is narrower and, we think, more
++useful: when the share of trading decisions taken by one language model is raised exogenously in a
++market where everything else is held fixed, the first-order effect is not herding into bubbles but a
++deep, calm and slow market---and the slowness, not the calm, is where the next question lies.

modified paper/sections/introduction.tex +73 −1

++\section{Introduction}
++\label{sec:introduction}
++
++Trading desks, retail brokers and market-making firms have begun to connect large language models (LLMs)
++to order entry. The models are few, the prompts are similar, and the information they see is the same
++public tape. If a meaningful share of order flow comes to be generated by a handful of foundation models
++answering near-identical questions about near-identical screens, the classical concern about correlated
++trading---that it impairs the aggregation of dispersed information and amplifies shocks---acquires a
++new and concrete source. Regulators have said as much \citep{fsb2024, danielsson2025}; the empirical
++record on what actually happens when LLM agents populate a market is thin, because the counterfactual
++market without them cannot be observed.
++
++This paper builds the counterfactual. We construct a transparent, replayable exchange economy in which
++the share of traders controlled by a language model is an experimental treatment, and we ask six
++questions: whether a rising share improves or impairs informational efficiency; whether a shared model
++generates correlated orders and liquidity spirals even when prompts and private signals differ; how LLM
++agents interact with fundamentalists, trend followers, noise traders, arbitrageurs, market makers and
++leveraged traders; whether memory, communication and adaptive feedback improve decisions while
++synchronizing them; what the causal role of agent identity, information, latency and budget is; and
++which safeguards mitigate instability at what cost. The answers are causal statements about the
++simulator, identified by randomization, and we are explicit that they are not statements about
++institutions.
++
++Three design choices separate this from existing LLM market simulations
++\citep{lopezlira2025trade, yang2025twinmarket, saxena2026, henning2025}. First, the economy keeps four
++prices apart: the latent fundamental value defined by an exogenous dividend process whose next
++innovation is drawn but not yet announced; the public value implied by announced dividends and public
++news; the transaction price; and the quote midpoint. Price discovery is measured as the convergence of
++the mid to the latent value, not to a known target, and information has value because the latent
++innovation is revealed only through noisy news, private signals and the next dividend. Second, the
++LLM treatment is randomized: a seed fixes the dividend path, the news, the endowments, the latencies and
++the policy assignment of every trader slot, and the treatment replaces a stratified subset of slots by
++LLM agents that inherit the information set of the slot they replace. Matched pairs differ only in who
++decides. Third, the mechanisms are derived before they are tested. A stylized model of traders whose
++signal errors share a model-specific component delivers closed-form comparative statics for price
++informativeness, order-flow correlation, synchrony, the effective number of models, feedback
++amplification and the leverage spiral; the experimental design maps each primitive to a simulator
++parameter and pre-registers the sign of every prediction, the primary outcomes, the thresholds that
++define bubbles and crashes and the estimators.
++
++The experiments use OpenAI's \code{gpt-5.4-nano} as the shared model (with \code{gpt-5.4-mini} in a
++two-model and an equal-budget arm), 100-trader markets of 600 ticks, and seven matched seeds per live
++cell: six shares from 0 to 100\,\% of slots, three diversity and cognition arms, a leverage interaction
++with a matched classical comparator, two safeguards, two fairness arms, and a frozen-decision-library
++replay at 1,000 traders. A classical design of sixteen cells and thirty seeds validates the stylized facts
++and the mechanisms without any language model. The analysis is paired at the run level with Holm
++adjustment within hypothesis families; with seven seeds, effects below roughly 1.2 standard deviations are
++not detectable and are reported as such.
++
++The results separate cleanly into what the theory got right and what it got wrong. It got coordination
++right: LLM agents deciding at the same tick are on the same side nine times out of ten, a communication
++channel makes them more alike, and prompt or model diversity makes them less so. It got prices wrong:
++raising the LLM share from 0 to 95\,\% of slots tightens the quoted spread from 18.6 to 6.1 basis points,
++raises depth from 141 to 956 lots, cuts price impact by 60\,\% and leaves the deviation of price from
++latent value unchanged at about half a percent; no bubble, crash or halt occurs in any of 112 live runs;
++and in the presence of leveraged trend followers the LLM agents halve the pricing error and remove the
++bubble episodes that leverage creates in the classical economy. The cost appears in dynamics rather than
++levels: latent information is absorbed more slowly as the share rises, the five-tick variance ratio
++climbs from 1.1 to 2.1 and the first-order autocorrelation of returns from 0.02 to 0.32, so that the
++price reaches its value in a smooth, predictable drift that trend followers exploit. The mechanism that
++unifies these facts is not correlated error but patient liquidity provision: a small reasoning model given
++the public present value and asked every twenty ticks what to do mostly rests a limit order near that
++value and waits. A market populated by such agents is deep, calm, and slow to learn what it does not
++already know.
++
++Section~\ref{sec:literature} positions the paper. Section~\ref{sec:model} describes the economy,
++Section~\ref{sec:theory} the stylized model and its propositions, Section~\ref{sec:identification} the
++experimental design and pre-analysis plan, Section~\ref{sec:implementation} the software and its
++validation. Section~\ref{sec:results} reports the classical and live experiments,
++Section~\ref{sec:mechanisms} reads them against the theory, Section~\ref{sec:robustness} probes
++thresholds, caching, dose, mechanism and scale, Section~\ref{sec:limitations} states what the design
++cannot say, and Section~\ref{sec:conclusion} concludes. Appendices give the proofs, the prompts, the
++algorithms, every hyperparameter and the replication package.

modified paper/sections/mechanisms.tex +63 −1

++\section{Mechanisms: what the evidence says about the theory}
++\label{sec:mechanisms}
++
++Table~\ref{tab:predictions} listed six predictions. Reading them against Section~\ref{sec:results} gives
++a mixed scorecard that is more informative than a clean confirmation would have been.
++
++\paragraph{H1 (informativeness falls in the shared-model share): not supported on levels, supported on dynamics.}
++The stylized model predicts that a common error component in the signals of a growing share of traders
++leaves a persistent gap between price and value. The simulator shows no such gap: the tracking error
++is, if anything, lower with LLM agents at every share. The reason is visible in the prompts and the
++decision logs. The LLM agents are given the public present value and told how it moves with the next
++dividend; most of them hold no private signal (they inherit the information set of the slot they replace,
++and only fundamentalists are informed); their modal decision is a limit order near the public value.
++Their ``common error'' is therefore not a biased signal about the fundamental but a common
++\emph{anchor} on the public value, and a population anchored on the public value prices the stock at the
++public value---which is close to the latent value except in the window between a latent shock and its
++revelation. In that window the prediction holds: the shock is absorbed more slowly as the share rises
++(Section~\ref{sec:results_dose}), and the persistence of returns (variance ratio 2.1, autocorrelation
++0.32 at 95\,\%) is the signature of a price that moves toward new information in many small steps against
++a wall of stale limit orders. The misleading-news cell of the classical design shows what a genuinely
++biased common signal does (Section~\ref{sec:results_classical}); the LLM cells did not inject one.
++
++\paragraph{H2 (coordination rises with the share): supported for synchrony, rejected for flow correlation.}
++Decision synchrony is near its ceiling at every share, consistent with a single model answering a single
++snapshot. Pairwise flow correlation falls with the share because the agents trade with one another; the
++stylized model, in which every trader submits a market order against a common market maker, has no room
++for this. The classical analogue shows that rule-based agents of one type are more correlated than
++LLM agents, which qualifies the common intuition that language models are the more homogeneous traders:
++in this economy they are more homogeneous than a mixed classical population and less homogeneous than any
++single classical rule.
++
++\paragraph{H3 (bubbles and crashes rise with penetration): rejected; H3$'$ (leverage interaction): reversed.}
++No bubble, crash or halt occurs in 112 live runs. With leverage present, the LLM agents reduce the
++instability the leveraged traders create. Proposition~\ref{prop:spiral} is about an exposed cohort whose
++positions are same-signed and leveraged; the LLM agents in this design are unleveraged and their
++positions lean against the exposed cohort, so the proposition's premise is not met and its conclusion does
++not apply. A design in which the LLM agents themselves carry leverage is the natural next test.
++
++\paragraph{H4 (diversity restores informativeness): supported in sign, small in size.}
++Diverse prompts and the two-model mix lower coordination and shorten the absorption of the latent shock
++by about four ticks, as Proposition~\ref{prop:diversity} predicts; they also lower liquidity provision,
++which the proposition does not consider because its traders do not supply liquidity.
++
++\paragraph{H5 (communication raises synchrony): supported.}
++The channel raises pairwise correlation to the highest level of all cells and the tracking error by a
++small but precisely estimated amount; memory does the same, less strongly.
++
++\paragraph{H6 (safeguards trade depth for delay): not testable at these shock sizes.}
++The breaker never fires; the position limit binds and removes liquidity without removing any
++instability, because there was none to remove.
++
++\paragraph{The unifying mechanism.}
++The one regularity that runs through every arm is liquidity provision by value-anchored agents who act
++infrequently. It explains the tighter spreads and deeper books, the lower volume, the slower absorption of
++latent information, the predictable returns, the trend followers' higher profits, the dampening of
++leverage-driven fire sales, the liquidity cost of diversity (more market orders) and of position limits,
++and the behaviour of the frozen library at $N = 1{,}000$. It is not the mechanism Section~\ref{sec:theory}
++set out to test---correlated errors in a market-order population---but it is the mechanism the data
++identify, and it suggests a different stylized model: a population of patient, public-value-anchored
++limit-order traders whose aggregate order book slows price discovery of private information while
++stabilizing prices against non-fundamental shocks. Whether that is good or bad for a market depends on the
++mix of information and noise in its order flow, which the information-quality and misleading-news cells
++begin to map.

modified paper/sections/results.tex +4 −1

				`@@ -1 +1,4 @@`
1	−% results — to be written	1	+\section{Results}
		2	+\label{sec:results}
		3	+\input{sections/results_classical}
		4	+\input{sections/results_llm}

added paper/sections/results_classical.tex +134 −0

++\subsection{The classical economy: stylized facts and mechanism checks}
++\label{sec:results_classical}
++
++Before any language model enters the market, the classical economy must (i) reproduce the stylized facts
++it was calibrated to, and (ii) respond to the mechanisms of Section~\ref{sec:theory} in the predicted
++direction. Table~\ref{tab:stylized_facts} summarizes the 30 control seeds of the classical design
++(seeds 101--130): first-order autocorrelation of tick returns is indistinguishable from zero
++($-0.007$, seed SD 0.15), the variance ratio at five ticks is 1.18, excess kurtosis averages 20 with a
++minimum of 1.8 across seeds, the first ten autocorrelations of absolute returns sum to $2.0$, the quoted
++spread is 21~bp (effective 30~bp), and the mid tracks the latent value with a mean absolute deviation
++of 0.63\,\%. The market absorbs the latent $+2.2\,\%$ shock of tick~300 in 25~ticks on average, i.e.\
++before the dividend of tick~320 reveals it, through the informed fundamentalists and the news flow; the
++20-tick response ratio of 0.28 says that most of the adjustment happens in the last ticks before the
++announcement. No control seed produces a bubble episode (5\,\% for ten ticks) or a crash (10\,\% in
++twenty ticks), which is the floor effect anticipated in Section~\ref{sec:identification}: in the
++classical economy instability can only be created, not reduced.
++
++Table~\ref{tab:classical_cells} reports cell means and Table~\ref{tab:classical_effects} the paired
++effects against the control for the five primary outcomes, with Holm-adjusted $p$-values within each
++cell. Four results matter for what follows.
++
++\paragraph{Feedback and leverage create instability (Propositions~\ref{prop:feedback} and \ref{prop:spiral}).}
++Raising the trend-follower share from 6 to 16\,\% raises the tracking error by 0.86 percentage points
++($d = 1.26$), the spread by 14~bp and the maximum drawdown by 2.8 points, and creates bubble episodes
++in 13\,\% of seeds where the control has none. Raising the leveraged share from 4 to 19\,\% has larger
++effects on every outcome: tracking error $+2.1$ points, time to absorb $+28$ ticks, spread $+16$~bp,
++0.9 bubble episodes per run and a maximum drawdown of 7.8\,\% against 3.3\,\%; adding a funding shock
++(a 50\,\% cut of the leverage cap twenty ticks after a negative latent shock) pushes the bubble count to
++1.4 and the drawdown to 9.4\,\%, and produces crashes in 30\,\% of seeds. Figure~\ref{fig:baseline_paths}
++shows a representative seed of each economy. Leveraged traders lose money in both cells ($-4.8$ and
++$-6.0$\,\% of initial wealth against $+0.7$\,\% in the control), and the losses are transferred to the
++fundamentalists and market makers who absorb the forced sales.
++
++\paragraph{Information quality and misleading news.}
++Lowering the precision of private signals from 2 to 0.5 has no detectable effect on any primary outcome,
++and raising it to 8 tightens the spread by 3~bp and the drawdown by half a point without changing the
++tracking error. The small effects are a property of this economy rather than a failure of the
++information channel: the latent shock is revealed by the next dividend within at most twenty ticks and
++by public news in the meantime, so private signals shorten the absorption of a shock that the market
++would absorb anyway. Misleading news is different: when half of the public items carry a hidden
++10-cent bias, the tracking error rises by 0.08 points ($d = 0.74$), the spread by 3~bp and the drawdown
++by half a point, because the public value itself is pulled away from the latent value and the
++fundamentalists trade on it. This cell is the classical counterpart of a shared, biased model: a common
++error in the public signal degrades price discovery even when every trader is otherwise rational.
++
++\paragraph{Liquidity withdrawal.}
++When the five market makers stop quoting for sixty ticks at the moment of the latent shock, the shock
++takes 8~ticks longer to absorb ($p = 0.05$ unadjusted), but the tracking error, spread and drawdown
++measured over the whole post-warmup window are unchanged: the remaining limit-order traders supply
++enough liquidity for the price to find its level once the market makers return.
++
++\paragraph{Safeguards and market design.}
++The 5\,\% circuit breaker almost never fires in the classical economy (0.13 halts per run) and has no
++effect; it matters only where prices move by 5\,\% in ten ticks, which the LLM arms below test. The
++position limit of 300 lots is actively harmful here: it caps exactly the traders who absorb order
++imbalances, so the tracking error rises by 1.3 points, the time to absorb by 28 ticks and the spread by
++18~bp, and 0.6 bubble episodes per run appear. The periodic call auction (clearing every five ticks with
++a $\pm 2\,\%$ collar) raises the tracking error by 0.7 points and the drawdown by 2.3 points relative
++to continuous trading at the same shock, which is the expected cost of batching with a mechanical
++collar; it is reported as a mechanism-robustness check in Section~\ref{sec:robustness}, not as a
++recommendation.
++
++\paragraph{Placebo and negative controls.}
++Replacing the latent shock by two placebo events leaves every outcome within its control range except
++the tracking error, which falls by 0.1 points because there is no shock to absorb; no bubble, crash or
++halt appears. The two cells in which 25 or 50\,\% of the slots run the \emph{mock} policy, a seeded
++value-plus-noise rule with independent errors, are the negative control for the LLM treatment: with
++independent errors, replacing a quarter or half of the traders changes neither the tracking error nor the
++time to absorb, bubbles or drawdowns, and tightens the spread (by 4 and 7~bp) because the mock rule
++rests limit orders near the mid. Whatever the live LLM arms do beyond this is attributable to the
++language model's decisions, not to the act of replacing slots.
++
++\paragraph{Two assets.}
++With two assets whose dividend innovations load 0.7 on a common factor, the fundamental return
++correlation across dividend periods is 0.38 while tick-return correlation between the two mids is
++$-0.01$: microstructure noise dominates at the tick frequency and the two books are priced almost
++independently, so excess comovement is negative in the control ($-0.39$). Adding leveraged traders and a
++funding shock to asset~0 produces crashes in asset~0 in half of the seeds and raises the maximum drawdown
++of asset~1 from 3.6 to 11.6\,\% although its fundamentals are untouched, and the tracking error of
++asset~1 from 0.6 to 7.4\,\%: contagion through common funding, not through correlated fundamentals. The
++Diebold--Yilmaz spillover index stays near 1.2\,\% in both cells, because the variance decomposition of a
++VAR on tick returns does not pick up the forced-sale episodes (Table~\ref{tab:contagion}).
++
++\begin{table}[htbp]
++\centering
++\caption{Stylized facts of the classical control economy (30 seeds, post-warm-up window).}
++\label{tab:stylized_facts}
++\small
++\input{tables/tab_stylized_facts}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Classical mechanism checks: cell means (30 matched seeds per cell). Stars: paired $t$-test
++against the control, Holm-adjusted within the cell ($^{*}$ 10\,\%, $^{**}$ 5\,\%, $^{***}$ 1\,\%).}
++\label{tab:classical_cells}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_classical_cells}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Classical mechanism checks: paired effects against the control with 95\,\% confidence intervals,
++Cohen's $d_z$ and Holm-adjusted $p$-values (five primary outcomes per cell).}
++\label{tab:classical_effects}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_classical_effects}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Two-asset economy: comovement and spillovers (30 seeds per cell).}
++\label{tab:contagion}
++\small
++\resizebox{\textwidth}{!}{\input{tables/tab_contagion}}
++\end{table}
++
++\begin{figure}[htbp]
++\centering
++\includegraphics[width=\textwidth]{figures/fig_baseline_paths}
++\caption{Mid price, latent value and public value in one seed (101) of three classical economies. The
++latent $+10$-cent shock of tick~300 is revealed by news and by the dividend of tick~320.}
++\label{fig:baseline_paths}
++\end{figure}
++
++\begin{figure}[htbp]
++\centering
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_mad_pct}\caption{Tracking error}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_time_to_absorb}\caption{Time to absorb the latent shock}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_max_dd}\caption{Maximum drawdown}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_classical_spread_bp}\caption{Quoted spread}\end{subfigure}
++\caption{Classical mechanism checks: cell means with 95\,\% confidence intervals over 30 seeds (control in grey).}
++\label{fig:classical}
++\end{figure}

added paper/sections/results_llm.tex +263 −0

++\subsection{Dose--response: raising the share of LLM-controlled traders}
++\label{sec:results_dose}
++
++Table~\ref{tab:dose_conf} reports the dose--response arm of the confirmatory design: six cells in which
++0, 5, 10, 25, 50 and 100\,\% of the trader slots (the last excluding the five market makers, hence 95
++agents) are controlled by \code{gpt-5.4-nano} with the single generalist prompt, seven matched seeds per
++cell, everything else held fixed. The assigned share overstates the LLM presence in the order flow: an
++LLM agent decides every twenty ticks, so a 25\,\% share of slots is 5\,\% of orders and 25\,\% of traded
++volume, and the 95\,\% cell is 36\,\% of orders and 84\,\% of volume (Figure~\ref{fig:realized_share}).
++We report effects on the assigned share, which is the randomized quantity; the realized shares are
++descriptive. Across the 35,652 live decisions of this arm, 96--98\,\% were schema-valid and feasible
++(Table~\ref{tab:llm_reliability}); the remainder were infeasible orders (typically a buy exceeding free
++cash) and a handful of truncated JSON responses, all converted to \emph{hold} and logged.
++
++\paragraph{Liquidity rises monotonically with the share.}
++The quoted spread falls from 18.6~bp in the control to 15.1, 15.8, 11.0, 7.4 and 6.1~bp as the share
++rises ($d_z$ from $-0.9$ to $-2.2$; Holm-adjusted $p < 0.05$ from 10\,\% upward), the effective spread
++from 27.5 to 7.5~bp, depth within five levels from 141 to 956 lots, and Kyle's price impact from 9.3 to
++3.6~bp per 100 lots. Traded volume falls by 19\,\% at 25\,\% and 70\,\% at 95\,\%, because the LLM agents
++mostly rest limit orders (only 4\,\% of their orders are marketable) and hold them: the market becomes
++deeper, tighter and quieter. Resiliency, the half-life of spread deviations, rises from 0.7 to 2.0
++ticks at 95\,\%, which is the one liquidity statistic that deteriorates---spreads are tighter but, when
++they widen, they take longer to close because fewer agents re-quote.
++
++\paragraph{Pricing error does not rise.}
++The mean absolute deviation of the mid from the latent value is 0.73\,\% in the control and between
++0.50 and 0.56\,\% in every LLM cell; the paired differences ($-0.17$ to $-0.23$ points) are not
++distinguishable from zero with seven seeds (95\,\% CIs span $[-0.7, +0.3]$), and the same holds for the
++squared log error, the maximum drawdown and the number of bubble episodes, which is zero in all 42 runs
++of the arm at every threshold of the sensitivity grid (Section~\ref{sec:robustness}). Hypothesis H1 in
++its directional form---that a shared model degrades price discovery as its share rises---is not
++supported on the level of prices: within the detectable range ($d \gtrsim 1.2$) the tracking error does
++not increase, and the point estimates go the other way.
++
++\paragraph{But information is absorbed more slowly and returns become predictable.}
++The two statistics that respond to the share in the direction H1 predicts are dynamic. The time to
++absorb the latent shock of tick~300 falls from 36 ticks in the control to 20--25 ticks at 5--10\,\%
++(when a few LLM agents, some of them inheriting a private signal, add value-anchored demand) and then
++rises back to 36 ticks at 50\,\% and 40 ticks at 95\,\%; none of these paired differences is significant
++(the control itself varies between 5 and 100 ticks across seeds), but the pattern is monotone from 10\,\%
++upward. The variance ratio at five ticks rises from 1.14 to 1.55 (25\,\%), 1.85 (50\,\%) and 2.12
++(95\,\%; $d_z = 3.5$, Holm $p < 0.001$), and the first-order autocorrelation of tick returns from 0.02 to
++0.15, 0.24 and 0.32 ($d_z = 3.5$ at 95\,\%). Volatility clustering rises with the share as well (the sum of
++the first ten absolute-return autocorrelations from 2.2 to 3.3, $p = 0.03$ unadjusted at 95\,\%), while
++tick volatility falls by half. The joint picture is a market whose price moves less, in smaller and more
++persistent steps: LLM agents anchor on the public present value shown in their snapshot, lean against
++incoming order flow with resting limit orders, and thereby slow the incorporation of information that
++is not yet public. At 95\,\% the price still reaches the latent value---the tracking error is
++unchanged---but it gets there in a smooth drift that a trend follower could exploit, and indeed the
++trend followers' P\&L rises from 0.6 to 3--4\,\% of initial wealth in the LLM cells.
++
++\paragraph{Instability does not appear.}
++No bubble (5\,\% for ten ticks), no crash (10\,\% in twenty ticks) and no halt occurs in any run of the
++dose arm, and the maximum drawdown is 2.2--2.9\,\% throughout. Hypothesis H3 (bubbles and crashes rise
++with penetration) is rejected in this economy for the shares and the shock sizes tested; the leverage arm
++below asks whether LLM agents amplify instability generated elsewhere.
++
++\subsection{Coordination: are the LLM agents trading together?}
++\label{sec:results_coord}
++
++Table~\ref{tab:coordination} measures coordination among the LLM agents directly from their logged
++decisions and from the trades. Their decisions are highly synchronized at every share: in a decision
++round, the absolute net sign of the placed orders averages 0.89--0.95, i.e.\ nine out of ten agents
++deciding at the same tick are on the same side. The pairwise correlation of agents' net order signs over
++20-tick windows is 0.48 at 5\,\% and \emph{declines} to 0.34 at 95\,\% ($-0.12$ per unit share, cluster-robust
++$p < 0.01$), and the correlation of their realized net traded volume declines from 0.41 to zero. Two
++things are going on. First, LLM agents are coordinated by the snapshot they share: with one model, one
++prompt and one public state, the direction of their orders is nearly common, which is the mechanism
++Propositions~\ref{prop:corr} and \ref{prop:diversity} describe. Second, as their number grows they
++increasingly trade \emph{with each other}: an agent whose limit order rests on the bid is filled by
++another LLM agent's sell, so realized flows decorrelate even though decisions do not. The classical
++analogue puts the numbers in perspective: in the control economy the fundamentalists' flows are
++correlated at 0.53 and the trend followers' at 0.76---rule-based classical agents of one type are more
++correlated among themselves than the LLM agents, because they respond to the same statistic with the same
++rule. H2 is therefore supported for decision synchrony and rejected for the cross-sectional prediction
++that order-flow correlation rises with the share.
++
++The within-share association between coordination and outcomes is null (Table~\ref{tab:mechanism_association}):
++holding the share and the seed fixed, runs in which the LLM agents were more synchronized do not have a
++larger pricing error, a slower absorption or a wider spread. Coordination in this market is pervasive but
++not harmful, because the common direction is toward the public value.
++
++\subsection{Diversity, memory and communication at a fixed share}
++\label{sec:results_arms}
++
++Tables~\ref{tab:arms_conf} and \ref{tab:arms_effects} compare, at the 25\,\% share and on the same seven
++seeds, the single-model single-prompt cell with the arms that vary what the LLM agents are given.
++
++\paragraph{Diversity lowers coordination and liquidity provision (H4, in part).}
++Giving the 25 agents eight different persona prompts cuts the pairwise order-flow correlation from
++0.40 to 0.32 and the synchrony from 0.95 to 0.84; mixing two models (nano and mini, with the diverse
++prompts) cuts them to 0.30 and 0.83. Both arms absorb the latent shock four ticks faster
++($p \approx 0.03$) and the two-model mix has the highest 20-tick response ratio of all cells (0.55
++against 0.34). The cost is liquidity: the two-model cell has a spread of 15.6~bp against 11.0~bp
++($d_z = 2.1$, Holm $p < 0.01$) and a drawdown higher by 0.6 points, because the mini model places
++market orders five times more often (17\,\% of its orders) and holds more; the diverse-prompt cell has a
++tracking error higher by 0.04 points (Holm $p = 0.045$) but an unchanged spread. The direction of
++Proposition~\ref{prop:diversity}---less correlated errors, faster information incorporation---is
++observed; the magnitude on price levels is small, and the liquidity side effect is not in the stylized
++model.
++
++\paragraph{Memory and a communication channel raise coordination slightly and pricing error slightly.}
++Bounded memory of the agent's five previous decisions raises the pairwise correlation to 0.44 and the
++tracking error by 0.04 points ($d_z = 1.6$, Holm $p = 0.015$) with a larger drawdown (0.4 points,
++$d_z = 2.1$); the communication channel, in which cohorts of five agents read one another's rationales,
++raises the pairwise correlation to 0.48, the highest of all 25\,\% cells, and the tracking error by 0.05
++points ($d_z = 2.3$, Holm $p = 0.003$), while leaving spreads unchanged. H5 is supported in sign: shared
++rationales make decisions more alike and prices slightly less accurate. Adaptive P\&L feedback changes
++nothing. These effects are an order of magnitude smaller than the leverage effects below and would not
++survive in a design powered for $d \approx 0.5$ unless they replicate.
++
++\paragraph{Equal budgets and matched latency.}
++Five agents of the larger \code{gpt-5.4-mini} model (the equal-USD counterpart of 25 nano agents, at
++0.30 against 0.33 USD per run) absorb the shock nine ticks faster than 25 nano agents but leave the
++spread at 22~bp instead of 11 and the tracking error 0.14 points higher: fewer, more active agents
++provide less liquidity. Against five nano agents (the equal-decision counterpart) the mini agents are
++more synchronized (0.98 against 0.89, $p = 0.03$), earn more (5.7 against 3.8\,\% of initial wealth,
++$p = 0.03$) and absorb the shock five ticks faster. Giving the 25 nano agents the fastest latency in the
++market (0.05 tick instead of a draw from $[0.05, 1]$) changes no primary outcome beyond a 0.4-point
++larger drawdown: in this economy, where an LLM decision is taken every twenty ticks, the latency of its
++execution is second order.
++
++\subsection{Leverage and safeguards}
++\label{sec:results_leverage}
++
++The leverage arm raises the leveraged share from 4 to 19\,\% with 25\,\% LLM agents, and the matched
++comparator runs the same leverage with no LLM agents on the same seeds. Leverage harms every outcome in
++both cells. Relative to the classical comparator, however, the LLM agents \emph{dampen} the damage: the
++tracking error is 0.96 instead of 2.16\,\% ($d_z = -1.1$, $p = 0.03$), the latent shock is absorbed in 16
++instead of 53 ticks ($p = 0.055$), bubble episodes fall from 0.71 per run to none ($p = 0.09$) and the
++maximum drawdown from 7.1 to 5.5\,\% (not significant). The spread is the same (38--39~bp) and the
++leveraged traders lose slightly more ($-8.0$ against $-5.9$\,\%). The mechanism is the one documented in the
++dose arm: LLM agents rest value-anchored bids under the price and absorb the forced sales of the
++leveraged trend followers, which in the classical economy are absorbed only by the cash-constrained
++fundamentalists. This is the opposite of Proposition~\ref{prop:spiral}'s corollary that same-signed LLM
++positions enlarge the exposed cohort: here the LLM agents are not the leveraged cohort, and they lean
++against it.
++
++The 5\,\% circuit breaker never fires at 25\,\% share, so that cell is identical to the 25\,\% cell by
++construction (every decision is a cache hit of the same prompt), which we report rather than hide. The
++position limit of 300 lots doubles the spread (22 against 11~bp, $d_z = 1.7$) and lowers the variance ratio
++below one, i.e.\ it removes the LLM agents' ability to accumulate inventory and with it both their
++liquidity provision and the return persistence; the tracking error and drawdown are unchanged. Among the
++safeguards tested, only the one that binds (the position limit) has an effect, and it trades the
++liquidity gain for nothing in this economy.
++
++\subsection{Scaling with a frozen decision library}
++\label{sec:results_library}
++
++The 34,418 valid decisions recorded in the dose arm form a frozen library replayed by nearest-neighbour
++lookup on a seven-dimensional feature vector (mispricing relative to the public value, momentum, signal,
++inventory and cash ratios, spread, open orders). Table~\ref{tab:library} replays it with 25\,\% library
++agents at $N = 100$ and $N = 1{,}000$ (activity scaled so that order flow per tick is constant). At
++$N = 100$ the library reproduces the online pattern qualitatively---tighter spread (11.9 against 17.3~bp),
++deeper book, higher variance ratio (1.55 against 1.33) and autocorrelation (0.16 against 0.04)---with a
++lower feasibility rate (72\,\% of replayed decisions were feasible in the new state against 98\,\% online),
++which is the expected out-of-support cost of replaying decisions taken elsewhere. At $N = 1{,}000$, 250
++library agents take the spread from 13.0 to 5.5~bp, the depth from 217 to 1,223 lots and the variance
++ratio from 1.14 to 1.89, with a tracking error of 0.61 against 0.73\,\% and no bubble or crash. The
++library is not an adaptive agent and its runs are not pooled with the online results; they show that the
++liquidity-and-persistence signature of the online cells survives a tenfold larger population at zero API
++cost.
++
++\begin{table}[htbp]
++\centering
++\caption{Dose--response in the assigned share of LLM-controlled trader slots (\code{gpt-5.4-nano}, single
++prompt, seven matched seeds per cell). Stars: paired $t$-test against the 0\,\% cell, Holm-adjusted across
++the five shares ($^{*}$ 10\,\%, $^{**}$ 5\,\%, $^{***}$ 1\,\%).}
++\label{tab:dose_conf}
++\small
++\resizebox{\textwidth}{!}{\input{tables/tab_dose_conf}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Coordination among LLM agents by share, and the classical analogue in the 0\,\% cell. Trend:
++cluster-robust OLS on the share with seed fixed effects.}
++\label{tab:coordination}
++\small
++\resizebox{\textwidth}{!}{\input{tables/tab_coordination}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Within-share association between coordination and outcomes (seed and share fixed effects,
++cluster-robust by seed; descriptive, not a mediation estimate).}
++\label{tab:mechanism_association}
++\small
++\input{tables/tab_mechanism_association}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Arms at the 25\,\% share: cell means (seven matched seeds). Stars: paired test against the
++25\,\% single-model single-prompt cell, Holm-adjusted within the cell.}
++\label{tab:arms_conf}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_arms_conf}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Arms at the 25\,\% share: paired effects against the 25\,\% single-model single-prompt cell.}
++\label{tab:arms_effects}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_arms_effects}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{LLM decision reliability and cost by cell (all decisions of the confirmatory design).}
++\label{tab:llm_reliability}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_llm_reliability_conf}}
++\end{table}
++
++\begin{table}[htbp]
++\centering
++\caption{Frozen decision library replayed at $N=100$ and $N=1{,}000$ (five seeds per cell, no API calls).}
++\label{tab:library}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_library}}
++\end{table}
++
++\begin{figure}[htbp]
++\centering
++\includegraphics[width=\textwidth]{figures/fig_dose_conf}
++\caption{Paired effects of the assigned LLM share on nine outcomes (difference from the 0\,\% cell with
++95\,\% confidence intervals over seven matched seeds).}
++\label{fig:dose}
++\end{figure}
++
++\begin{figure}[htbp]
++\centering
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_realized_share_conf}\caption{Assigned vs realized volume share}\label{fig:realized_share}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_sync_conf}\caption{Pairwise order-flow correlation of LLM agents}\end{subfigure}
++\caption{Realized dose and coordination in the dose--response arm (one point per run).}
++\label{fig:realized}
++\end{figure}
++
++\begin{figure}[htbp]
++\centering
++\includegraphics[width=\textwidth]{figures/fig_paths_conf}
++\caption{Mid price, latent and public value in seed 201 at four LLM shares.}
++\label{fig:paths_conf}
++\end{figure}
++
++\begin{figure}[htbp]
++\centering
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_arms_conf_spread_bp}\caption{Quoted spread}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_arms_conf_mad_pct}\caption{Tracking error}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_arms_conf_llm_pairwise_corr}\caption{Pairwise order-flow correlation (LLM)}\end{subfigure}
++\begin{subfigure}{0.49\textwidth}\includegraphics[width=\textwidth]{figures/fig_arms_conf_max_dd}\caption{Maximum drawdown}\end{subfigure}
++\caption{Arms of the confirmatory design (cell means with 95\,\% confidence intervals; the 25\,\% single-model cell in grey).}
++\label{fig:arms}
++\end{figure}
++
++\begin{figure}[htbp]
++\centering
++\includegraphics[width=0.6\textwidth]{figures/fig_cost_quality_conf}
++\caption{Cost--quality frontier: API cost per run against the tracking error, by cell.}
++\label{fig:cost_quality}
++\end{figure}

modified paper/sections/robustness.tex +59 −1

++\section{Robustness}
++\label{sec:robustness}
++
++\paragraph{Bubble and crash thresholds.}
++The pre-registered sensitivity grid (bubble 3/5/8\,\% for ten ticks, crash 7/10/15\,\% in twenty ticks)
++is computed from the saved tick series of every run (Table~\ref{tab:threshold_sensitivity}). In the
++classical design the leverage cells produce 0.3--2.3 bubble episodes per run depending on the bubble
++threshold and 0--1.2 crashes depending on the crash threshold, so the leverage results are threshold-robust
++in sign; in the live design every count is zero at every threshold except 0.14 bubble episodes per run
++in the 25\,\%-plus-leverage cell at the 3\,\% threshold. The floor effect is therefore not an artefact of a
++strict definition.
++
++\paragraph{Cached decisions and the breaker cell.}
++Every provider response is cached by the hash of its prompt, and an identical prompt is replayed rather
++than re-queried. This makes matched comparisons sharper (the breaker cell is literally the 25\,\% cell
++until a halt changes a snapshot) and means that provider non-determinism enters the results only once per
++distinct prompt. Across the confirmatory design 7,370 of 80,031 decisions were cache hits: 4,824 in the breaker cell
++(identical to the 25\,\% cell) and the rest in the first decision rounds, where the market state has not
++yet diverged from a previously queried run; after that, prompts are unique to their run.
++
++\paragraph{Realized versus assigned dose.}
++Figure~\ref{fig:realized_share} shows that the realized volume share is close to the assigned slot share
++up to 50\,\% and below it at 95\,\% (84\,\%), while the order share is five to ten times smaller. Re-reading
++the dose--response on the realized volume share does not change any sign; we do not estimate it because
++the realized share is an outcome.
++
++\paragraph{Market mechanism.}
++Under a periodic call auction (classical design) the tracking error rises by 0.7 points and the drawdown
++by 2.3 points relative to continuous trading, with no change in the time to absorb; the batch mechanism
++with a collar is a different economy rather than a robustness check of the same one, and we did not run
++LLM cells under it.
++
++\paragraph{Population size.}
++The classical stylized facts hold at $N = 1{,}000$ and $N = 10{,}000$ with activity scaled by $100/N$
++(Table~\ref{tab:scale}), and the frozen-library replay reproduces the liquidity-and-persistence signature
++of the online cells at $N = 1{,}000$ (Table~\ref{tab:library}). The online LLM cells are all at $N = 100$.
++
++\paragraph{Information quality.}
++The classical information cells (precision 0.5 and 8, misleading news) bound what private information
++can do in this economy: the latent shock is revealed within twenty ticks in any case, so precision moves
++only the speed of absorption, and only biased public information moves the tracking error. The live
++design inherits information sets from the replaced slots; an information-equalization arm in which every
++LLM agent receives a signal (\code{llm.informed\_share}) is implemented and was not run within the budget.
++
++\paragraph{Seed count and inference.}
++All live contrasts are paired over seven seeds. The effects reported as significant have $d_z$ between 1
++and 4 and survive Holm adjustment within their family; the null results on pricing error, absorption and
++drawdown have confidence intervals that include effects of the size found in the classical leverage cells,
++so they are inconclusive for small effects and conclusive only against large ones. The classical design
++(30 seeds) has the power to detect $d \approx 0.5$.
++
++\begin{table}[htbp]
++\centering
++\caption{Threshold sensitivity: paired effect on bubble and crash counts against the control, for every
++combination of bubble (b) and crash (c) thresholds in the pre-registered grid.}
++\label{tab:threshold_sensitivity}
++\scriptsize
++\resizebox{\textwidth}{!}{\input{tables/tab_threshold_sensitivity}}
++\end{table}

added paper/tables/tab_arms_conf.tex +21 −0

++\begin{tabular}{lrrrrrrrr}
++\toprule
++Cell & Mean $|P/V-1|$ (\%) & Time to absorb (ticks) & Quoted spread (bp) & Bubble episodes & Max drawdown & LLM order-flow corr. & LLM synchrony & LLM valid-decision rate \\
++\midrule
++0\% (control) & 0.73 & 36.43 & 18.61$^{***}$ & 0.00 & 0.03 & -- & -- & -- \\
++25\% single model/prompt & 0.50 & 28.43 & 10.95 & 0.00 & 0.02 & 0.40 & 0.95 & 0.98 \\
++25\% diverse prompts & 0.53$^{**}$ & 24.00$^{**}$ & 11.47 & 0.00 & 0.03$^{*}$ & 0.32$^{**}$ & 0.83$^{***}$ & 0.97$^{**}$ \\
++25\% two models & 0.50 & 24.00$^{**}$ & 15.57$^{***}$ & 0.00 & 0.03$^{***}$ & 0.29$^{**}$ & 0.83$^{***}$ & 0.98 \\
++25\% memory (5) & 0.54$^{***}$ & 31.14 & 10.71 & 0.00 & 0.03$^{***}$ & 0.44 & 0.95 & 0.99$^{*}$ \\
++25\% communication channel & 0.55$^{***}$ & 28.00 & 11.63 & 0.00 & 0.02 & 0.48$^{**}$ & 0.93 & 0.98 \\
++25\% adaptive feedback & 0.50 & 26.57 & 10.62 & 0.00 & 0.02 & 0.41 & 0.95 & 0.98 \\
++0\% + leveraged 19\% (classical) & 2.16$^{***}$ & 53.00 & 38.82$^{**}$ & 0.71$^{*}$ & 0.07$^{***}$ & -- & -- & -- \\
++25\% + leveraged 19\% & 0.96$^{***}$ & 16.29$^{**}$ & 38.51$^{***}$ & 0.00 & 0.06$^{***}$ & 0.46 & 0.95 & 0.98 \\
++25\% + circuit breaker & 0.50 & 28.43 & 10.95 & 0.00 & 0.02 & 0.40 & 0.95 & 0.98 \\
++25\% + position limit & 0.51 & 34.00 & 22.48$^{***}$ & 0.00 & 0.02 & 0.39 & 0.94 & 0.96$^{***}$ \\
++5\% gpt-5.4-mini (equal USD) & 0.64$^{**}$ & 19.57$^{**}$ & 22.20$^{***}$ & 0.00 & 0.03$^{*}$ & 0.52$^{**}$ & 0.98$^{**}$ & 0.99 \\
++25\% matched fast latency & 0.51 & 29.57 & 12.09 & 0.00 & 0.03$^{**}$ & 0.43 & 0.95 & 0.98 \\
++\midrule
++Seeds per cell & 7 & & & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added paper/tables/tab_arms_effects.tex +78 −0

++\begin{tabular}{llrrrrr}
++\toprule
++Cell & Outcome & Control & Effect & 95\% CI & $d$ & $p$ (Holm) \\
++\midrule
++0\% (control) & Mean $|P/V-1|$ (\%) & 0.50 & 0.23 & [-0.24, 0.70] & 0.45 & 0.612 \\
++ & Time to absorb (ticks) & 28.43 & 8.00 & [-21.43, 37.43] & 0.25 & 0.612 \\
++ & Quoted spread (bp) & 10.95 & 7.66$^{**}$ & [2.89, 12.42] & 1.49 & 0.031 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01 & [-0.00, 0.02] & 0.54 & 0.612 \\
++\addlinespace
++25\% diverse prompts & Mean $|P/V-1|$ (\%) & 0.50 & 0.04$^{**}$ & [0.01, 0.06] & 1.36 & 0.045 \\
++ & Time to absorb (ticks) & 28.43 & -4.43$^{*}$ & [-8.35, -0.51] & -1.05 & 0.098 \\
++ & Quoted spread (bp) & 10.95 & 0.52 & [-0.39, 1.42] & 0.53 & 0.211 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.01] & 0.75 & 0.186 \\
++\addlinespace
++25\% two models & Mean $|P/V-1|$ (\%) & 0.50 & 0.01 & [-0.01, 0.03] & 0.29 & 0.469 \\
++ & Time to absorb (ticks) & 28.43 & -4.43$^{**}$ & [-8.00, -0.85] & -1.15 & 0.046 \\
++ & Quoted spread (bp) & 10.95 & 4.62$^{***}$ & [2.61, 6.64] & 2.12 & 0.005 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01$^{***}$ & [0.00, 0.01] & 1.90 & 0.007 \\
++\addlinespace
++25\% memory (5) & Mean $|P/V-1|$ (\%) & 0.50 & 0.04$^{**}$ & [0.02, 0.06] & 1.63 & 0.015 \\
++ & Time to absorb (ticks) & 28.43 & 2.71 & [-2.56, 7.99] & 0.48 & 0.510 \\
++ & Quoted spread (bp) & 10.95 & -0.24 & [-1.30, 0.81] & -0.21 & 0.593 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00$^{***}$ & [0.00, 0.01] & 2.05 & 0.006 \\
++\addlinespace
++25\% communication channel & Mean $|P/V-1|$ (\%) & 0.50 & 0.05$^{***}$ & [0.03, 0.07] & 2.35 & 0.003 \\
++ & Time to absorb (ticks) & 28.43 & -0.43 & [-2.68, 1.83] & -0.18 & 0.666 \\
++ & Quoted spread (bp) & 10.95 & 0.68 & [-0.54, 1.90] & 0.51 & 0.666 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.01] & 0.46 & 0.666 \\
++\addlinespace
++25\% adaptive feedback & Mean $|P/V-1|$ (\%) & 0.50 & -0.00 & [-0.03, 0.02] & -0.02 & 1.000 \\
++ & Time to absorb (ticks) & 28.43 & -1.86 & [-9.42, 5.70] & -0.23 & 1.000 \\
++ & Quoted spread (bp) & 10.95 & -0.33 & [-2.13, 1.48] & -0.17 & 1.000 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.00] & 0.09 & 1.000 \\
++\addlinespace
++0\% + leveraged 19\% (classical) & Mean $|P/V-1|$ (\%) & 0.50 & 1.67$^{**}$ & [0.58, 2.75] & 1.42 & 0.038 \\
++ & Time to absorb (ticks) & 28.43 & 24.57 & [-8.14, 57.29] & 0.69 & 0.188 \\
++ & Quoted spread (bp) & 10.95 & 27.87$^{**}$ & [8.90, 46.85] & 1.36 & 0.038 \\
++ & Bubble episodes & 0.00 & 0.71 & [-0.17, 1.59] & 0.75 & 0.188 \\
++ & Max drawdown & 0.02 & 0.05$^{**}$ & [0.02, 0.08] & 1.50 & 0.037 \\
++\addlinespace
++25\% + leveraged 19\% & Mean $|P/V-1|$ (\%) & 0.50 & 0.46$^{***}$ & [0.36, 0.56] & 4.12 & 0.000 \\
++ & Time to absorb (ticks) & 28.43 & -12.14$^{**}$ & [-23.60, -0.69] & -0.98 & 0.041 \\
++ & Quoted spread (bp) & 10.95 & 27.56$^{***}$ & [19.31, 35.81] & 3.09 & 0.000 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.03$^{***}$ & [0.03, 0.04] & 4.24 & 0.000 \\
++\addlinespace
++25\% + circuit breaker & Mean $|P/V-1|$ (\%) & 0.50 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Time to absorb (ticks) & 28.43 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Quoted spread (bp) & 10.95 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [0.00, 0.00] & -- & -- \\
++\addlinespace
++25\% + position limit & Mean $|P/V-1|$ (\%) & 0.50 & 0.02 & [-0.04, 0.07] & 0.27 & 1.000 \\
++ & Time to absorb (ticks) & 28.43 & 5.57 & [-8.04, 19.18] & 0.38 & 1.000 \\
++ & Quoted spread (bp) & 10.95 & 11.53$^{**}$ & [5.15, 17.90] & 1.67 & 0.018 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & -0.00 & [-0.01, 0.00] & -0.13 & 1.000 \\
++\addlinespace
++5\% gpt-5.4-mini (equal USD) & Mean $|P/V-1|$ (\%) & 0.50 & 0.14$^{*}$ & [0.02, 0.27] & 1.04 & 0.067 \\
++ & Time to absorb (ticks) & 28.43 & -8.86$^{*}$ & [-15.49, -2.22] & -1.23 & 0.051 \\
++ & Quoted spread (bp) & 10.95 & 11.25$^{***}$ & [5.85, 16.64] & 1.93 & 0.009 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01$^{*}$ & [-0.00, 0.02] & 0.79 & 0.082 \\
++\addlinespace
++25\% matched fast latency & Mean $|P/V-1|$ (\%) & 0.50 & 0.01 & [-0.00, 0.03] & 0.71 & 0.333 \\
++ & Time to absorb (ticks) & 28.43 & 1.14 & [-3.62, 5.90] & 0.22 & 0.578 \\
++ & Quoted spread (bp) & 10.95 & 1.14 & [-0.92, 3.19] & 0.51 & 0.451 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00$^{**}$ & [0.00, 0.01] & 1.37 & 0.044 \\
++\addlinespace
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

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

++\begin{tabular}{lrrrrrr}
++\toprule
++Statistic & 5\% & 10\% & 25\% & 50\% & 100\% & Trend per unit share (SE) \\
++\midrule
++Pairwise order-flow correlation (LLM) & 0.475 & 0.420 & 0.399 & 0.389 & 0.338 & -0.117 (0.024)$^{***}$ \\
++Synchrony (LLM) & 0.889 & 0.938 & 0.948 & 0.910 & 0.900 & -0.019 (0.025) \\
++Flow correlation from trades (LLM) & 0.409 & 0.352 & 0.205 & 0.052 & -0.008 & -0.429 (0.021)$^{***}$ \\
++Hold rate & 0.142 & 0.112 & 0.118 & 0.108 & 0.102 & -0.029 (0.023) \\
++Valid-decision rate & 0.981 & 0.964 & 0.981 & 0.973 & 0.957 & -0.019 (0.009)$^{**}$ \\
++\midrule
++Flow correlation, fundamentalist (0\% cell) & \multicolumn{5}{r}{0.531} & \\
++Flow correlation, trend (0\% cell) & \multicolumn{5}{r}{0.757} & \\
++Flow correlation, noise (0\% cell) & \multicolumn{5}{r}{0.035} & \\
++Flow correlation, market maker (0\% cell) & \multicolumn{5}{r}{0.364} & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added paper/tables/tab_dose_conf.tex +23 −0

++\begin{tabular}{lrrrrrr}
++\toprule
++Outcome & 0\% & 5\% & 10\% & 25\% & 50\% & 100\% \\
++\midrule
++Sq. log pricing error ($\times 10^{4}$) & 1.00 & 0.47 & 0.52 & 0.47 & 0.52 & 0.61 \\
++Mean $|P/V-1|$ (\%) & 0.73 & 0.52 & 0.56 & 0.50 & 0.51 & 0.54 \\
++Time to absorb (ticks) & 36.43 & 25.00 & 20.29 & 28.43 & 35.57 & 39.71 \\
++Quoted spread (bp) & 18.61 & 15.09$^{*}$ & 15.80$^{**}$ & 10.95$^{**}$ & 7.38$^{***}$ & 6.07$^{***}$ \\
++Price impact (bp/100 lots) & 9.32 & 8.03 & 8.77 & 5.21 & 5.40 & 3.61$^{**}$ \\
++Bubble episodes & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 \\
++Max drawdown & 0.03 & 0.03 & 0.03 & 0.02 & 0.02 & 0.02 \\
++Vol. clustering $\sum_{1}^{10}\mathrm{ACF}|r|$ & 2.22 & 1.72 & 1.79 & 2.27 & 2.73 & 3.31 \\
++LLM order-flow corr. & -- & 0.48 & 0.42 & 0.40 & 0.39 & 0.34 \\
++LLM valid-decision rate & -- & 0.98 & 0.96 & 0.98 & 0.97 & 0.96 \\
++\midrule
++Realized share of traders & 0.000 & 0.050 & 0.100 & 0.250 & 0.500 & 0.950 \\
++Realized share of orders & 0.000 & 0.009 & 0.018 & 0.048 & 0.109 & 0.355 \\
++Realized share of volume & 0.000 & 0.056 & 0.109 & 0.248 & 0.432 & 0.837 \\
++API cost per run (USD) & -- & 0.082 & 0.170 & 0.331 & 0.821 & 1.779 \\
++\midrule
++Seeds per cell & 7 & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

modified paper/tables/tab_dose_pilot.tex +13 −13

				`@@ -2,21 +2,21 @@`
2	\toprule	2	\toprule
3	Outcome & 5\% & 25\% & 50\% \\	3	Outcome & 5\% & 25\% & 50\% \\
4	\midrule	4	\midrule
5	−Sq. log pricing error ($\times 10^{4}$) & 0.66 & 0.55 & 2.15 \\	5	+Sq. log pricing error ($\times 10^{4}$) & 0.66 & 0.55 & 0.47 \\
6	−Mean $\|P/V-1\|$ (\%) & 0.54 & 0.48 & 0.69 \\	6	+Mean $\|P/V-1\|$ (\%) & 0.54 & 0.48 & 0.43 \\
7	−Time to absorb (ticks) & 21.67 & 27.00 & 22.00 \\	7	+Time to absorb (ticks) & 21.67 & 27.00 & 21.33 \\
8	−Quoted spread (bp) & 17.93 & 11.58 & 11.27 \\	8	+Quoted spread (bp) & 17.93 & 11.58 & 7.76 \\
9	−Price impact (bp/100 lots) & 10.34 & 8.08 & 13.41 \\	9	+Price impact (bp/100 lots) & 10.34 & 8.08 & 6.68 \\
10	−Bubble episodes & 0.00 & 0.00 & 1.00 \\	10	+Bubble episodes & 0.00 & 0.00 & 0.00 \\
11	−Max drawdown & 0.03 & 0.02 & 0.03 \\	11	+Max drawdown & 0.03 & 0.02 & 0.02 \\
12	−Vol. clustering $\sum_{1}^{10}\mathrm{ACF}\|r\|$ & 2.79 & 2.34 & 3.83 \\	12	+Vol. clustering $\sum_{1}^{10}\mathrm{ACF}\|r\|$ & 2.79 & 2.34 & 3.19 \\
13	−LLM order-flow corr. & 0.38 & 0.41 & 0.20 \\	13	+LLM order-flow corr. & 0.38 & 0.41 & 0.35 \\
14	−LLM valid-decision rate & 0.92 & 0.96 & 0.93 \\	14	+LLM valid-decision rate & 0.92 & 0.96 & 0.97 \\
15	\midrule	15	\midrule
16	Realized share of traders & 0.050 & 0.250 & 0.500 \\	16	Realized share of traders & 0.050 & 0.250 & 0.500 \\
17	−Realized share of orders & 0.008 & 0.046 & 0.086 \\	17	+Realized share of orders & 0.008 & 0.046 & 0.106 \\
18	−Realized share of volume & 0.052 & 0.253 & 0.366 \\	18	+Realized share of volume & 0.052 & 0.253 & 0.446 \\
19	−API cost per run (USD) & 0.088 & 0.436 & 0.462 \\	19	+API cost per run (USD) & 0.088 & 0.436 & 0.888 \\
20	\midrule	20	\midrule
21	Seeds per cell & 3 & & \\	21	Seeds per cell & 3 & & \\
22	\bottomrule	22	\bottomrule

added paper/tables/tab_library.tex +12 −0

++\begin{tabular}{lrrrrrrrrrr}
++\toprule
++Cell & Mean $|P/V-1|$ (\%) & Time to absorb (ticks) & Quoted spread (bp) & Depth, 5 levels (lots) & Price impact (bp/100 lots) & Variance ratio (5) & ACF$_1$ of returns & Max drawdown & Realized LLM volume share & LLM valid-decision rate \\
++\midrule
++$N=100$, 0\% & 0.59 & 25.60 & 17.29 & 134.51 & 9.21 & 1.33 & 0.04 & 0.04 & 0.00 & -- \\
++$N=100$, 25\% library & 0.52 & 26.40 & 11.93 & 190.24 & 6.36 & 1.55 & 0.16 & 0.03 & 0.26 & 0.72 \\
++$N=1{,}000$, 0\% & 0.73 & 32.00 & 13.04 & 217.27 & 4.60 & 1.14 & -0.09 & 0.03 & 0.00 & -- \\
++$N=1{,}000$, 25\% library & 0.61 & 40.80 & 5.45 & 1222.83 & 1.18 & 1.89 & 0.21 & 0.02 & 0.63 & 0.61 \\
++\midrule
++Seeds per cell & 5 & & & & & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added paper/tables/tab_llm_reliability_conf.tex +21 −0

++\begin{tabular}{lrrrrrrrrrr}
++\toprule
++Cell & Decisions & Valid & Hold & Market & JSON err. & Schema err. & Infeasible & API err. & Latency (s) & USD/run \\
++\midrule
++adaptive\_025 & 4,824 & 0.976 & 0.10 & 0.04 & 0.000 & 0.000 & 0.023 & 0.000 & 2.75 & 0.43 \\
++breaker\_025 & 4,824 & 0.981 & 0.12 & 0.04 & 0.000 & 0.000 & 0.019 & 0.000 & 0.00 & 0.00 \\
++channel\_025 & 4,824 & 0.984 & 0.09 & 0.02 & 0.000 & 0.000 & 0.016 & 0.000 & 2.42 & 0.41 \\
++div\_models\_025 & 4,824 & 0.983 & 0.20 & 0.17 & 0.000 & 0.000 & 0.017 & 0.000 & 2.62 & 0.94 \\
++div\_prompts\_025 & 4,824 & 0.968 & 0.13 & 0.04 & 0.000 & 0.000 & 0.032 & 0.000 & 2.82 & 0.42 \\
++dose\_005 & 963 & 0.981 & 0.14 & 0.06 & 0.000 & 0.000 & 0.019 & 0.000 & 2.57 & 0.08 \\
++dose\_010 & 1,929 & 0.964 & 0.11 & 0.05 & 0.000 & 0.000 & 0.036 & 0.000 & 2.67 & 0.17 \\
++dose\_025 & 4,824 & 0.981 & 0.12 & 0.04 & 0.000 & 0.000 & 0.019 & 0.000 & 2.17 & 0.33 \\
++dose\_050 & 9,633 & 0.973 & 0.11 & 0.04 & 0.000 & 0.000 & 0.027 & 0.000 & 2.78 & 0.82 \\
++dose\_100 & 18,303 & 0.957 & 0.10 & 0.04 & 0.000 & 0.000 & 0.043 & 0.000 & 3.15 & 1.78 \\
++fair\_latency\_025 & 4,824 & 0.979 & 0.11 & 0.04 & 0.000 & 0.000 & 0.020 & 0.000 & 2.79 & 0.43 \\
++fair\_mini\_005 & 963 & 0.987 & 0.18 & 0.54 & 0.000 & 0.000 & 0.013 & 0.000 & 2.72 & 0.30 \\
++leverage\_025 & 4,824 & 0.983 & 0.07 & 0.07 & 0.000 & 0.000 & 0.017 & 0.000 & 2.88 & 0.44 \\
++memory\_025 & 4,824 & 0.988 & 0.09 & 0.02 & 0.000 & 0.000 & 0.012 & 0.000 & 2.80 & 0.44 \\
++poslimit\_025 & 4,824 & 0.961 & 0.10 & 0.04 & 0.000 & 0.000 & 0.039 & 0.000 & 2.90 & 0.44 \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

modified paper/tables/tab_llm_reliability_pilot.tex +1 −1

				`@@ -4,6 +4,6 @@ Cell & Decisions & Valid & Hold & Market & JSON err. & Schema err. & Infeasible`
4	\midrule	4	\midrule
5	share\_25 & 2,066 & 0.958 & 0.12 & 0.04 & 0.000 & 0.000 & 0.042 & 0.000 & 2.84 & 0.44 \\	5	share\_25 & 2,066 & 0.958 & 0.12 & 0.04 & 0.000 & 0.000 & 0.042 & 0.000 & 2.84 & 0.44 \\
6	share\_5 & 411 & 0.922 & 0.13 & 0.05 & 0.000 & 0.000 & 0.078 & 0.000 & 2.86 & 0.09 \\	6	share\_5 & 411 & 0.922 & 0.13 & 0.05 & 0.000 & 0.000 & 0.078 & 0.000 & 2.86 & 0.09 \\
7	−share\_50 & 4,128 & 0.934 & 0.38 & 0.09 & 0.014 & 0.008 & 0.044 & 0.000 & 1.47 & 0.46 \\	7	+share\_50 & 4,128 & 0.966 & 0.12 & 0.04 & 0.000 & 0.000 & 0.033 & 0.000 & 2.83 & 0.89 \\
8	\bottomrule	8	\bottomrule
9	\end{tabular}	9	\end{tabular}
	\ No newline at end of file

added paper/tables/tab_mechanism_association.tex +20 −0

++\begin{tabular}{llrrr}
++\toprule
++Outcome & Mediator (per 1 SD) & Coefficient & SE & $p$ \\
++\midrule
++mad pct & pairwise corr & +0.011 & 0.019 & 0.545 \\
++mad pct & synchrony & +0.027 & 0.015 & 0.080 \\
++msd log & pairwise corr & +0.000 & 0.000 & 0.269 \\
++msd log & synchrony & +0.000 & 0.000 & 0.220 \\
++time to absorb & pairwise corr & +1.919 & 1.186 & 0.106 \\
++time to absorb & synchrony & +1.068 & 1.575 & 0.498 \\
++n bubbles & pairwise corr & +0.000 & 0.000 & nan \\
++n bubbles & synchrony & +0.000 & 0.000 & nan \\
++max dd & pairwise corr & -0.001 & 0.001 & 0.162 \\
++max dd & synchrony & +0.001 & 0.001 & 0.233 \\
++spread bp & pairwise corr & -0.361 & 1.140 & 0.751 \\
++spread bp & synchrony & +1.053 & 1.032 & 0.307 \\
++lambda bp 100 & pairwise corr & +0.005 & 0.748 & 0.994 \\
++lambda bp 100 & synchrony & +0.820 & 0.724 & 0.257 \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

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

				`@@ -2,8 +2,8 @@`
2	\toprule	2	\toprule
3	Traders & Seeds & Wall-clock (s) & Orders & Trades & Mean $\|P/V-1\|$ (\%) & ACF$_1$ & Spread (bp) \\	3	Traders & Seeds & Wall-clock (s) & Orders & Trades & Mean $\|P/V-1\|$ (\%) & ACF$_1$ & Spread (bp) \\
4	\midrule	4	\midrule
5	−100 & 3 & 0.6 & 13,689 & 10,521 & 0.82 & 0.11 & 25.5 \\	5	+100 & 3 & 0.6 & 13,382 & 10,194 & 0.62 & -0.04 & 22.8 \\
6	−1,000 & 3 & 2.7 & 14,939 & 11,840 & 0.99 & 0.15 & 14.2 \\	6	+1,000 & 3 & 2.6 & 14,083 & 11,290 & 0.59 & 0.03 & 13.9 \\
7	−10,000 & 1 & 23.7 & 15,378 & 11,885 & 1.23 & 0.10 & 10.9 \\	7	+10,000 & 1 & 23.7 & 14,679 & 11,123 & 0.75 & -0.14 & 9.5 \\
8	\bottomrule	8	\bottomrule
9	\end{tabular}	9	\end{tabular}
	\ No newline at end of file

modified paper/tables/tab_threshold_sensitivity.tex +7 −1

				`@@ -10,7 +10,13 @@ classical confirmatory & mock llm 50 indep & n bubbles & +0.00 & +0.00 & +0.00 &`
10	classical confirmatory & mock llm 50 indep & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	10	classical confirmatory & mock llm 50 indep & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
11	classical confirmatory & trend heavy & n bubbles & +1.10 & +1.10 & +1.10 & +0.13 & +0.13 & +0.13 & +0.00 & +0.00 & +0.00 \\	11	classical confirmatory & trend heavy & n bubbles & +1.10 & +1.10 & +1.10 & +0.13 & +0.13 & +0.13 & +0.00 & +0.00 & +0.00 \\
12	classical confirmatory & trend heavy & n crashes & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 \\	12	classical confirmatory & trend heavy & n crashes & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 \\
13	−confirmatory openai & dose 025 & n bubbles & -0.17 & -0.17 & -0.17 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	13	+confirmatory openai & dose 025 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
14	confirmatory openai & dose 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	14	confirmatory openai & dose 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		15	+confirmatory openai & dose 050 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		16	+confirmatory openai & dose 050 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		17	+confirmatory openai & dose 100 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		18	+confirmatory openai & dose 100 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		19	+confirmatory openai & leverage 025 & n bubbles & +0.14 & +0.14 & +0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		20	+confirmatory openai & leverage 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
15	\bottomrule	21	\bottomrule
16	\end{tabular}	22	\end{tabular}
	\ No newline at end of file

added results/figures/fig_arms_conf_llm_pairwise_corr.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_llm_pairwise_corr.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_llm_synchrony.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_llm_synchrony.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_llm_valid_rate.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_llm_valid_rate.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_mad_pct.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_mad_pct.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_max_dd.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_max_dd.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_n_bubbles.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_n_bubbles.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_spread_bp.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_spread_bp.png +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_time_to_absorb.pdf +0 −0

Binary file not shown.

added results/figures/fig_arms_conf_time_to_absorb.png +0 −0

Binary file not shown.

modified results/figures/fig_baseline_paths.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_lambda_bp_100.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_mad_pct.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_max_dd.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_n_bubbles.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_spread_bp.pdf +0 −0

Binary file not shown.

modified results/figures/fig_classical_time_to_absorb.pdf +0 −0

Binary file not shown.

modified results/figures/fig_contagion_excess.pdf +0 −0

Binary file not shown.

modified results/figures/fig_contagion_spillover.pdf +0 −0

Binary file not shown.

added results/figures/fig_cost_quality_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_cost_quality_conf.png +0 −0

Binary file not shown.

added results/figures/fig_dose_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_dose_conf.png +0 −0

Binary file not shown.

added results/figures/fig_dose_pilot.pdf +0 −0

Binary file not shown.

added results/figures/fig_dose_pilot.png +0 −0

Binary file not shown.

added results/figures/fig_lp_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_lp_conf.png +0 −0

Binary file not shown.

added results/figures/fig_paths_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_paths_conf.png +0 −0

Binary file not shown.

added results/figures/fig_paths_pilot.pdf +0 −0

Binary file not shown.

added results/figures/fig_paths_pilot.png +0 −0

Binary file not shown.

added results/figures/fig_realized_share_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_realized_share_conf.png +0 −0

Binary file not shown.

added results/figures/fig_realized_share_pilot.pdf +0 −0

Binary file not shown.

added results/figures/fig_realized_share_pilot.png +0 −0

Binary file not shown.

added results/figures/fig_sync_conf.pdf +0 −0

Binary file not shown.

added results/figures/fig_sync_conf.png +0 −0

Binary file not shown.

added results/figures/fig_sync_pilot.pdf +0 −0

Binary file not shown.

added results/figures/fig_sync_pilot.png +0 −0

Binary file not shown.

modified results/figures/fig_wealth_depth.pdf +0 −0

Binary file not shown.

added results/library/gpt-5.4-nano_single.npz +0 −0

Binary file not shown.

added results/tables/mechanisms.csv +91 −0

++block,share,stat,mean,se,n,beta_share,p,mediator,beta_per_sd,cell,control_mean,diff,ci_lo,ci_hi,cohen_d
++coordination_by_share,0.05,llm_pairwise_corr,0.47534864727094867,0.03328242408336079,7.0,,,,,,,,,,
++coordination_by_share,0.1,llm_pairwise_corr,0.42038565837320985,0.021151905035378154,7.0,,,,,,,,,,
++coordination_by_share,0.25,llm_pairwise_corr,0.3991565429691119,0.023863626575511054,7.0,,,,,,,,,,
++coordination_by_share,0.5,llm_pairwise_corr,0.3888324707624974,0.014449132963520161,7.0,,,,,,,,,,
++coordination_by_share,1.0,llm_pairwise_corr,0.3377276491814324,0.017540446190826658,7.0,,,,,,,,,,
++trend,,llm_pairwise_corr,,0.02406034390156864,35.0,-0.11677006558193814,1.21474757860886e-06,,,,,,,,
++coordination_by_share,0.05,llm_synchrony,0.8885936895083236,0.04054455660742793,7.0,,,,,,,,,,
++coordination_by_share,0.1,llm_synchrony,0.937576714581571,0.0112422144925865,7.0,,,,,,,,,,
++coordination_by_share,0.25,llm_synchrony,0.9478745048304666,0.007771805352358979,7.0,,,,,,,,,,
++coordination_by_share,0.5,llm_synchrony,0.9100850697356939,0.005184768871850125,7.0,,,,,,,,,,
++coordination_by_share,1.0,llm_synchrony,0.9001570732047995,0.01090077469350093,7.0,,,,,,,,,,
++trend,,llm_synchrony,,0.02531619244749916,35.0,-0.01935902071855436,0.4444565971629735,,,,,,,,
++coordination_by_share,0.05,flow_corr_llm,0.40878392723989243,0.037925680055498366,7.0,,,,,,,,,,
++coordination_by_share,0.1,flow_corr_llm,0.35198335633709704,0.020177349223348986,7.0,,,,,,,,,,
++coordination_by_share,0.25,flow_corr_llm,0.20534245322984754,0.01676676582222103,7.0,,,,,,,,,,
++coordination_by_share,0.5,flow_corr_llm,0.052199011151800465,0.006239730795931206,7.0,,,,,,,,,,
++coordination_by_share,1.0,flow_corr_llm,-0.00765017341785877,0.0004944611276987119,7.0,,,,,,,,,,
++trend,,flow_corr_llm,,0.021124831335010516,35.0,-0.42890179247809024,1.2046444840931742e-91,,,,,,,,
++coordination_by_share,0.05,llm_hold_rate,0.142094857619341,0.036078729974285385,7.0,,,,,,,,,,
++coordination_by_share,0.1,llm_hold_rate,0.11240950577744738,0.013990008800765381,7.0,,,,,,,,,,
++coordination_by_share,0.25,llm_hold_rate,0.11832981447918316,0.010436114099914885,7.0,,,,,,,,,,
++coordination_by_share,0.5,llm_hold_rate,0.10777159846219628,0.005040603561784755,7.0,,,,,,,,,,
++coordination_by_share,1.0,llm_hold_rate,0.10229018405306711,0.0072313712310514535,7.0,,,,,,,,,,
++trend,,llm_hold_rate,,0.02344495653995385,35.0,-0.0288496427628465,0.21849997069864557,,,,,,,,
++coordination_by_share,0.05,llm_valid_rate,0.9812677848035438,0.005008358994393158,7.0,,,,,,,,,,
++coordination_by_share,0.1,llm_valid_rate,0.9636338933743637,0.011834582780440773,7.0,,,,,,,,,,
++coordination_by_share,0.25,llm_valid_rate,0.9809044022280899,0.00461802137796935,7.0,,,,,,,,,,
++coordination_by_share,0.5,llm_valid_rate,0.9729072026114723,0.002407356798852634,7.0,,,,,,,,,,
++coordination_by_share,1.0,llm_valid_rate,0.9566672027012222,0.0022739197615887496,7.0,,,,,,,,,,
++trend,,llm_valid_rate,,0.008754249915757682,35.0,-0.018691378825055338,0.03275116510081593,,,,,,,,
++classical_flow_corr,,flow_corr_fundamentalist,0.5307853924704865,,7.0,,,,,,,,,,
++classical_flow_corr,,flow_corr_trend,0.7570009177345741,,7.0,,,,,,,,,,
++classical_flow_corr,,flow_corr_noise,0.0345947025538287,,7.0,,,,,,,,,,
++classical_flow_corr,,flow_corr_market_maker,0.36408673419950727,,7.0,,,,,,,,,,
++within_share_association,,mad_pct,,0.01889825137452115,35.0,,0.5446287228883753,llm_pairwise_corr,0.011449096992300026,,,,,,
++within_share_association,,mad_pct,,0.015271226281562668,35.0,,0.08030658567158394,llm_synchrony,0.026707999655927177,,,,,,
++within_share_association,,msd_log,,3.1675765309020856e-06,35.0,,0.2685817818339524,llm_pairwise_corr,3.504399069836127e-06,,,,,,
++within_share_association,,msd_log,,2.611229147170323e-06,35.0,,0.2198863252549994,llm_synchrony,3.203535417537203e-06,,,,,,
++within_share_association,,time_to_absorb,,1.1860425860728134,35.0,,0.10574788325110501,llm_pairwise_corr,1.9185476893186988,,,,,,
++within_share_association,,time_to_absorb,,1.5750785131509206,35.0,,0.49773259947415205,llm_synchrony,1.0680003494349921,,,,,,
++within_share_association,,n_bubbles,,0.0,35.0,,,llm_pairwise_corr,0.0,,,,,,
++within_share_association,,n_bubbles,,0.0,35.0,,,llm_synchrony,0.0,,,,,,
++within_share_association,,max_dd,,0.0009437125314385361,35.0,,0.1619770102946294,llm_pairwise_corr,-0.0013197378307558826,,,,,,
++within_share_association,,max_dd,,0.0009462744987631889,35.0,,0.23338622576458679,llm_synchrony,0.0011276576426681954,,,,,,
++within_share_association,,spread_bp,,1.1400120524632835,35.0,,0.7514553776246395,llm_pairwise_corr,-0.36106565992926143,,,,,,
++within_share_association,,spread_bp,,1.0315214612364594,35.0,,0.3073316330038698,llm_synchrony,1.053015357301787,,,,,,
++within_share_association,,lambda_bp_100,,0.7480775685233,35.0,,0.994318372307264,llm_pairwise_corr,0.005327003878567882,,,,,,
++within_share_association,,lambda_bp_100,,0.7242343053807916,35.0,,0.2574078573006494,llm_synchrony,0.8202236655869927,,,,,,
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.01739651828416276,,,div_prompts_025,0.3991565429691119,-0.07658304241764745,-0.13418548116805878,-0.0189806036672361,-1.2295920573799648
++arm_vs_dose_025,,llm_synchrony,,,7.0,,3.0499680166681946e-05,,,div_prompts_025,0.9478745048304668,-0.11329237341116818,-0.13807945091062127,-0.0885052959117151,-4.227120759536539
++arm_vs_dose_025,,mad_pct,,,7.0,,0.01125406311224627,,,div_prompts_025,0.49799489146906445,0.03607564838243805,0.01161141531965933,0.06053988144521677,1.3638036368725202
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.03262195956062448,,,div_prompts_025,28.428571428571427,-4.428571428571429,-8.347165996444192,-0.5099768606986661,-1.0452077615828625
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,div_prompts_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.09319294024490339,,,div_prompts_025,0.022262207437842796,0.003365427878981055,-0.0007643334949551223,0.007495189252917232,0.7536759108626273
++arm_vs_dose_025,,spread_bp,,,7.0,,0.21130995236649439,,,div_prompts_025,10.950967182666245,0.5175395250524747,-0.38762272739517045,1.4227017775001198,0.5287938466689498
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.020670374345326235,,,div_models_025,0.3991565429691119,-0.10455242277309852,-0.18663498663645145,-0.02246985890974558,-1.1780195345630746
++arm_vs_dose_025,,llm_synchrony,,,7.0,,6.619040488123943e-05,,,div_models_025,0.9478745048304665,-0.1207638282106156,-0.15101433378385073,-0.09051332263738046,-3.692100708889214
++arm_vs_dose_025,,mad_pct,,,7.0,,0.4691691124716501,,,div_models_025,0.49799489146906445,0.0061508715321395725,-0.013333259527725816,0.025635002592004963,0.29196105114831233
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.023094835525958576,,,div_models_025,28.428571428571427,-4.428571428571429,-8.004793515189657,-0.8523493419532007,-1.145271562457717
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,div_models_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.002366099536537163,,,div_models_025,0.022262207437842796,0.005617331023162119,0.0028879831965841923,0.008346678849740045,1.9034454542865162
++arm_vs_dose_025,,spread_bp,,,7.0,,0.0013734235563795775,,,div_models_025,10.950967182666245,4.624012251431416,2.60567456412838,6.642349938734451,2.118821888829357
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.12797288248130367,,,memory_025,0.39915654296911196,0.04099494042611359,-0.01583235974339369,0.09782224059562086,0.6671792649404588
++arm_vs_dose_025,,llm_synchrony,,,7.0,,0.72552019965094,,,memory_025,0.9478745048304668,0.002791966094088177,-0.015773378753609082,0.02135731094178544,0.13908376023975993
++arm_vs_dose_025,,mad_pct,,,7.0,,0.005035366883793889,,,memory_025,0.49799489146906445,0.04041508852081144,0.01747225608118905,0.06335792096043383,1.6291677535777296
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.2550291887882483,,,memory_025,28.428571428571427,2.7142857142857144,-2.563936958241845,7.992508386813274,0.47559486560567094
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,memory_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.0016235920199536235,,,memory_025,0.022262207437842796,0.003624533169558765,0.0019900124560771433,0.005259053883040387,2.050836103730931
++arm_vs_dose_025,,spread_bp,,,7.0,,0.5933852405134774,,,memory_025,10.950967182666243,-0.24372520178532614,-1.3016889858943657,0.8142385823237133,-0.21305853755565282
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.017695198662024434,,,channel_025,0.3991565429691119,0.0845479997794174,0.02068831112013457,0.14840768843870022,1.2244635037603282
++arm_vs_dose_025,,llm_synchrony,,,7.0,,0.2687635647197799,,,channel_025,0.9478745048304668,-0.01775984289528032,-0.0534237423111069,0.01790405652054626,-0.46055298106879067
++arm_vs_dose_025,,mad_pct,,,7.0,,0.0008033601595175057,,,channel_025,0.49799489146906445,0.04880508958229821,0.029580284981921914,0.06802989418267451,2.3478615535882374
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.6584920975921567,,,channel_025,28.428571428571427,-0.42857142857142855,-2.6849640116957296,1.8278211545528726,-0.17566201313073598
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,channel_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.2737436283646387,,,channel_025,0.022262207437842796,0.0021173148349886866,-0.002183905570913413,0.006418535240890786,0.4552637247565824
++arm_vs_dose_025,,spread_bp,,,7.0,,0.2220703951005214,,,channel_025,10.950967182666245,0.679387598283001,-0.5410447578901464,1.8998199544561483,0.5148410959065869
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.4752649496890312,,,adaptive_025,0.3991565429691119,0.010443010178264664,-0.02311730608037605,0.044003326436905375,0.2877855348137093
++arm_vs_dose_025,,llm_synchrony,,,7.0,,0.8468957404426137,,,adaptive_025,0.9478745048304668,-0.001727929909849554,-0.02270132149034496,0.019245461670645853,-0.07619505049446058
++arm_vs_dose_025,,mad_pct,,,7.0,,0.9563956160242966,,,adaptive_025,0.49799489146906445,-0.0005920353553888995,-0.026006864951263742,0.024822794240485942,-0.021544168896346222
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.5697453163501741,,,adaptive_025,28.428571428571427,-1.8571428571428572,-9.41656315624899,5.702277441963277,-0.22720931069951117
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,adaptive_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.8121849899529254,,,adaptive_025,0.022262207437842796,0.00041654736956286964,-0.0036884467321390063,0.0045215414712647456,0.09384716620020697
++arm_vs_dose_025,,spread_bp,,,7.0,,0.6729682972140855,,,adaptive_025,10.950967182666245,-0.3274751104079265,-2.13436112787564,1.479410907059787,-0.1676165295502182
++arm_vs_dose_025,,llm_pairwise_corr,,,7.0,,0.18268660919362473,,,fair_latency_025,0.3991565429691119,0.035148370143347825,-0.021945657151504587,0.09224239743820023,0.5693558893561349
++arm_vs_dose_025,,llm_synchrony,,,7.0,,0.9493508951321226,,,fair_latency_025,0.9478745048304668,-0.0008516535735897073,-0.032319355471130934,0.03061604832395152,-0.02503036888867482
++arm_vs_dose_025,,mad_pct,,,7.0,,0.11094710423607979,,,fair_latency_025,0.49799489146906445,0.014771280924608046,-0.004575141618801825,0.034117703468017915,0.7061334636769161
++arm_vs_dose_025,,time_to_absorb,,,7.0,,0.578201730081321,,,fair_latency_025,28.428571428571427,1.1428571428571428,-3.6159384728281254,5.9016527585424114,0.22210799850754528
++arm_vs_dose_025,,n_bubbles,,,7.0,,,,,fair_latency_025,0.0,0.0,0.0,0.0,
++arm_vs_dose_025,,max_dd,,,7.0,,0.011064523226606303,,,fair_latency_025,0.022262207437842796,0.004449559956979551,0.0014439564209778686,0.007455163492981233,1.3691614874705338
++arm_vs_dose_025,,spread_bp,,,7.0,,0.22574435858849767,,,fair_latency_025,10.950967182666245,1.135196776170052,-0.9225186144041635,3.192912166744268,0.5102172617041314

added results/tables/tab_arms_conf.tex +21 −0

++\begin{tabular}{lrrrrrrrr}
++\toprule
++Cell & Mean $|P/V-1|$ (\%) & Time to absorb (ticks) & Quoted spread (bp) & Bubble episodes & Max drawdown & LLM order-flow corr. & LLM synchrony & LLM valid-decision rate \\
++\midrule
++0\% (control) & 0.73 & 36.43 & 18.61$^{***}$ & 0.00 & 0.03 & -- & -- & -- \\
++25\% single model/prompt & 0.50 & 28.43 & 10.95 & 0.00 & 0.02 & 0.40 & 0.95 & 0.98 \\
++25\% diverse prompts & 0.53$^{**}$ & 24.00$^{**}$ & 11.47 & 0.00 & 0.03$^{*}$ & 0.32$^{**}$ & 0.83$^{***}$ & 0.97$^{**}$ \\
++25\% two models & 0.50 & 24.00$^{**}$ & 15.57$^{***}$ & 0.00 & 0.03$^{***}$ & 0.29$^{**}$ & 0.83$^{***}$ & 0.98 \\
++25\% memory (5) & 0.54$^{***}$ & 31.14 & 10.71 & 0.00 & 0.03$^{***}$ & 0.44 & 0.95 & 0.99$^{*}$ \\
++25\% communication channel & 0.55$^{***}$ & 28.00 & 11.63 & 0.00 & 0.02 & 0.48$^{**}$ & 0.93 & 0.98 \\
++25\% adaptive feedback & 0.50 & 26.57 & 10.62 & 0.00 & 0.02 & 0.41 & 0.95 & 0.98 \\
++0\% + leveraged 19\% (classical) & 2.16$^{***}$ & 53.00 & 38.82$^{**}$ & 0.71$^{*}$ & 0.07$^{***}$ & -- & -- & -- \\
++25\% + leveraged 19\% & 0.96$^{***}$ & 16.29$^{**}$ & 38.51$^{***}$ & 0.00 & 0.06$^{***}$ & 0.46 & 0.95 & 0.98 \\
++25\% + circuit breaker & 0.50 & 28.43 & 10.95 & 0.00 & 0.02 & 0.40 & 0.95 & 0.98 \\
++25\% + position limit & 0.51 & 34.00 & 22.48$^{***}$ & 0.00 & 0.02 & 0.39 & 0.94 & 0.96$^{***}$ \\
++5\% gpt-5.4-mini (equal USD) & 0.64$^{**}$ & 19.57$^{**}$ & 22.20$^{***}$ & 0.00 & 0.03$^{*}$ & 0.52$^{**}$ & 0.98$^{**}$ & 0.99 \\
++25\% matched fast latency & 0.51 & 29.57 & 12.09 & 0.00 & 0.03$^{**}$ & 0.43 & 0.95 & 0.98 \\
++\midrule
++Seeds per cell & 7 & & & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added results/tables/tab_arms_effects.csv +61 −0

++outcome,level,n,control_mean,treat_mean,diff,se,ci_lo,ci_hi,t,p,p_wilcoxon,cohen_d,pct_change,cell,p_holm
++mad_pct,dose_000,7,0.4979948914690645,0.7262545952175093,0.22825970374844498,0.19087029198330097,-0.23878307573699661,0.6953024832338865,1.1958891107496978,0.2768527325717246,0.296875,0.45200359752198294,45.835752064662394,dose_000,0.6119164258886256
++time_to_absorb,dose_000,7,28.428571428571427,36.42857142857143,8.0,12.027745701779143,-21.430833500241466,37.43083350024146,0.665128794568432,0.5306887307713005,0.78125,0.2513950543223199,28.14070351758794,dose_000,0.6119164258886256
++spread_bp,dose_000,7,10.950967182666243,18.60841142837759,7.657444245711345,1.9471731128008254,2.8928832797681467,12.422005211654543,3.932595512628475,0.007689765742454608,0.015625,1.4863813904890733,69.92482141515266,dose_000,0.03075906296981843
++n_bubbles,dose_000,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,dose_000,
++max_dd,dose_000,7,0.022262207437842796,0.028270910068090682,0.006008702630247886,0.004215962457918657,-0.00430738587201559,0.01632479113251136,1.425226787530331,0.2039721419628752,0.15625,0.5386850916675354,26.9905967187867,dose_000,0.6119164258886256
++mad_pct,div_prompts_025,7,0.49799489146906445,0.5340705398515025,0.03607564838243805,0.00999800342269429,0.01161141531965933,0.06053988144521677,3.608285260290127,0.01125406311224627,0.046875,1.3638036368725202,7.244180412376595,div_prompts_025,0.04501625244898508
++time_to_absorb,div_prompts_025,7,28.428571428571427,24.0,-4.428571428571429,1.6014449257904988,-8.347165996444192,-0.5099768606986661,-2.7653598055427446,0.03262195956062448,0.015625,-1.0452077615828625,-15.577889447236183,div_prompts_025,0.09786587868187345
++spread_bp,div_prompts_025,7,10.950967182666245,11.46850670771872,0.5175395250524747,0.3699202535735377,-0.38762272739517045,1.4227017775001198,1.3990570131072622,0.21130995236649439,0.21875,0.5287938466689498,4.7259709249395145,div_prompts_025,0.21130995236649439
++n_bubbles,div_prompts_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,div_prompts_025,
++max_dd,div_prompts_025,7,0.022262207437842796,0.02562763531682385,0.003365427878981055,0.0016877442364765798,-0.0007643334949551223,0.007495189252917232,1.9940390292825958,0.09319294024490339,0.078125,0.7536759108626273,15.117224508744243,div_prompts_025,0.18638588048980678
++mad_pct,div_models_025,7,0.49799489146906445,0.5041457630012041,0.0061508715321395725,0.007962743345555426,-0.013333259527725816,0.025635002592004963,0.7724563338554434,0.4691691124716501,0.375,0.29196105114831233,1.2351274355435162,div_models_025,0.4691691124716501
++time_to_absorb,div_models_025,7,28.428571428571427,24.0,-4.428571428571429,1.4615246907831245,-8.004793515189657,-0.8523493419532007,-3.030103737897497,0.023094835525958576,0.03125,-1.145271562457717,-15.577889447236183,div_models_025,0.04618967105191715
++spread_bp,div_models_025,7,10.950967182666245,15.574979434097658,4.624012251431416,0.824850999989476,2.60567456412838,6.642349938734451,5.6058757902826235,0.0013734235563795775,0.015625,2.118821888829357,42.22469279928571,div_models_025,0.00549369422551831
++n_bubbles,div_models_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,div_models_025,
++max_dd,div_models_025,7,0.022262207437842796,0.027879538461004914,0.005617331023162119,0.0011154254802030518,0.0028879831965841923,0.008346678849740045,5.0360433062184855,0.002366099536537163,0.015625,1.9034454542865162,25.23258773347607,div_models_025,0.007098298609611489
++mad_pct,memory_025,7,0.49799489146906445,0.5384099799898759,0.04041508852081144,0.009376239862865019,0.01747225608118905,0.06335792096043383,4.310372719972432,0.005035366883793889,0.015625,1.6291677535777296,8.115562872861725,memory_025,0.015106100651381668
++time_to_absorb,memory_025,7,28.428571428571427,31.142857142857142,2.7142857142857144,2.1570955529345,-2.563936958241845,7.992508386813274,1.2583057392117916,0.2550291887882483,0.4375,0.47559486560567094,9.547738693467338,memory_025,0.5100583775764966
++spread_bp,memory_025,7,10.950967182666243,10.707241980880918,-0.24372520178532614,0.4323669377848607,-1.3016889858943657,0.8142385823237133,-0.5636999050713729,0.5933852405134774,0.6875,-0.21305853755565282,-2.225604348181291,memory_025,0.5933852405134774
++n_bubbles,memory_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,memory_025,
++max_dd,memory_025,7,0.022262207437842796,0.02588674060740156,0.003624533169558765,0.0006679932964143287,0.0019900124560771433,0.005259053883040387,5.426002310224707,0.0016235920199536235,0.015625,2.050836103730931,16.281104107392064,memory_025,0.006494368079814494
++mad_pct,channel_025,7,0.49799489146906445,0.5467999810513627,0.04880508958229821,0.007856762225161676,0.029580284981921914,0.06802989418267451,6.211857783604226,0.0008033601595175057,0.015625,2.3478615535882374,9.800319324225438,channel_025,0.0032134406380700227
++time_to_absorb,channel_025,7,28.428571428571427,28.0,-0.42857142857142855,0.9221388919541468,-2.6849640116957296,1.8278211545528726,-0.46475800154489005,0.6584920975921567,0.8125,-0.17566201313073598,-1.507537688442211,channel_025,0.6662111853015642
++spread_bp,channel_025,7,10.950967182666245,11.630354780949245,0.679387598283001,0.4987643325206311,-0.5410447578901464,1.8998199544561483,1.3621415044847829,0.2220703951005214,0.21875,0.5148410959065869,6.203904978898764,channel_025,0.6662111853015642
++n_bubbles,channel_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,channel_025,
++max_dd,channel_025,7,0.022262207437842796,0.024379522272831484,0.0021173148349886866,0.0017578158379058229,-0.002183905570913413,0.006418535240890786,1.204514596654877,0.2737436283646387,0.296875,0.4552637247565824,9.510803638409786,channel_025,0.6662111853015642
++mad_pct,adaptive_025,7,0.49799489146906445,0.49740285611367563,-0.0005920353553888995,0.010386491685012081,-0.026006864951263742,0.024822794240485942,-0.05700051310330499,0.9563956160242966,0.9375,-0.021544168896346222,-0.1188838210051552,adaptive_025,1.0
++time_to_absorb,adaptive_025,7,28.428571428571427,26.571428571428573,-1.8571428571428572,3.0893717301540176,-9.41656315624899,5.702277441963277,-0.6011393316693137,0.5697453163501741,0.46875,-0.22720931069951117,-6.532663316582915,adaptive_025,1.0
++spread_bp,adaptive_025,7,10.950967182666245,10.623492072258317,-0.3274751104079265,0.738435271635233,-2.13436112787564,1.479410907059787,-0.44347165281358647,0.6729682972140855,0.578125,-0.1676165295502182,-2.9903761461934697,adaptive_025,1.0
++n_bubbles,adaptive_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,adaptive_025,
++max_dd,adaptive_025,7,0.022262207437842796,0.022678754807405666,0.00041654736956286964,0.0016776223874926403,-0.0036884467321390063,0.0045215414712647456,0.24829626301389413,0.8121849899529254,0.8125,0.09384716620020697,1.8710964342861816,adaptive_025,1.0
++mad_pct,leverage_000,7,0.49799489146906456,2.1640512515986416,1.666056360129577,0.44400817026248535,0.579607506309104,2.75250521395005,3.752310141375666,0.009483940346143012,0.015625,1.4182399251522322,334.5529017807349,leverage_000,0.037935761384572046
++time_to_absorb,leverage_000,7,28.428571428571427,53.0,24.571428571428573,13.369932761267586,-8.143618351128598,57.28647549398575,1.8378124265973486,0.11572846581659263,0.21875,0.694627805308676,86.43216080402011,leverage_000,0.18826553313163297
++spread_bp,leverage_000,7,10.950967182666243,38.823433254015185,27.872466071348935,7.754384368401521,8.89817106197388,46.846761080723994,3.594413785435623,0.011443023682437976,0.015625,1.3585607121892769,254.5205880578924,leverage_000,0.037935761384572046
++n_bubbles,leverage_000,7,0.0,0.7142857142857143,0.7142857142857143,0.3595159254890833,-0.16541806446887886,1.5939894930403073,1.986798535597566,0.09413276656581648,0.25,0.7509392614826383,,leverage_000,0.18826553313163297
++max_dd,leverage_000,7,0.022262207437842796,0.07122341333848585,0.04896120590064306,0.01233474337259311,0.018779176161412992,0.07914323563987313,3.969373696856252,0.007372240987299569,0.015625,1.500282237508961,219.92969941253673,leverage_000,0.036861204936497846
++mad_pct,leverage_025,7,0.49799489146906456,0.9565431854579806,0.45854829398891617,0.04202935092788612,0.3557061771075404,0.561390410870292,10.910192136340445,3.517317812138456e-05,0.015625,4.123665021241331,92.07891523469597,leverage_025,0.00011977777704717322
++time_to_absorb,leverage_025,7,28.428571428571427,16.285714285714285,-12.142857142857142,4.682432067636134,-23.60035566133729,-0.6853586243769971,-2.5932799381726657,0.041030772903837824,0.015625,-0.9801676851968328,-42.71356783919598,leverage_025,0.041030772903837824
++spread_bp,leverage_025,7,10.950967182666243,38.512498640404544,27.561531457738294,3.370629099575898,19.3138991681719,35.80916374730469,8.176969534027391,0.00018016696091380485,0.015625,3.090603980741169,251.68125333590726,leverage_025,0.0003603339218276097
++n_bubbles,leverage_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,leverage_025,
++max_dd,leverage_025,7,0.022262207437842796,0.055290989531730346,0.03302878209388755,0.002943836393578712,0.025825473934607907,0.040232090253167194,11.219639164028301,2.9944444261793304e-05,0.015625,4.2406250039856435,148.36256550975716,leverage_025,0.00011977777704717322
++mad_pct,breaker_025,7,0.49799489146906445,0.49799489146906445,0.0,0.0,0.0,0.0,,,1.0,,0.0,breaker_025,
++time_to_absorb,breaker_025,7,28.428571428571427,28.428571428571427,0.0,0.0,0.0,0.0,,,1.0,,0.0,breaker_025,
++spread_bp,breaker_025,7,10.950967182666245,10.950967182666245,0.0,0.0,0.0,0.0,,,1.0,,0.0,breaker_025,
++n_bubbles,breaker_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,breaker_025,
++max_dd,breaker_025,7,0.022262207437842796,0.022262207437842796,0.0,0.0,0.0,0.0,,,1.0,,0.0,breaker_025,
++mad_pct,poslimit_025,7,0.49799489146906445,0.5135656679667123,0.015570776497647956,0.022076209262130973,-0.03844776157421688,0.06958931456951278,0.7053193015504574,0.5070560760127865,0.8125,0.26658563811375485,3.1266940212408216,poslimit_025,1.0
++time_to_absorb,poslimit_025,7,28.428571428571427,34.0,5.571428571428571,5.562874996914141,-8.040436184958729,19.18329332781587,1.0015376176022606,0.3552319962116441,0.46875,0.3785456378359553,19.597989949748744,poslimit_025,1.0
++spread_bp,poslimit_025,7,10.950967182666243,22.47650890039355,11.525541717727307,2.6056777040711414,5.149678063371392,17.90140537208322,4.423241485207348,0.004455370518958945,0.015625,1.6718281369489467,105.24679259353941,poslimit_025,0.01782148207583578
++n_bubbles,poslimit_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,poslimit_025,
++max_dd,poslimit_025,7,0.022262207437842796,0.021466273439547053,-0.0007959339982957424,0.002237228319847749,-0.006270234487848369,0.004678366491256884,-0.3557678897743921,0.7341884667068254,0.9375,-0.13446762297218295,-3.575269885154157,poslimit_025,1.0
++mad_pct,fair_mini_005,7,0.49799489146906445,0.6416155048779687,0.14362061340890428,0.052278287311719,0.015700252628296885,0.2715409741895117,2.7472325662189294,0.033413967120872544,0.03125,1.0383563091247245,28.839776445342515,fair_mini_005,0.06682793424174509
++time_to_absorb,fair_mini_005,7,28.428571428571427,19.571428571428573,-8.857142857142858,2.7117782904578758,-15.492625293741906,-2.2216604205438104,-3.266175147248987,0.017115954387445786,0.03125,-1.2344981682857983,-31.155778894472366,fair_mini_005,0.05134786316233736
++spread_bp,fair_mini_005,7,10.950967182666245,22.197428906912677,11.246461724246432,2.203617963776555,5.8544028132856125,16.63852063520725,5.103634980798701,0.002213483908255439,0.015625,1.9289927059490388,102.69834195145714,fair_mini_005,0.008853935633021757
++n_bubbles,fair_mini_005,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,fair_mini_005,
++max_dd,fair_mini_005,7,0.022262207437842796,0.029181382674795482,0.006919175236952687,0.0033176776970127515,-0.0011988896381471955,0.015037240112052569,2.085547744189478,0.08210230322942878,0.078125,0.7882629540681587,31.08036458770485,fair_mini_005,0.08210230322942878
++mad_pct,fair_latency_025,7,0.49799489146906445,0.5127661723936725,0.014771280924608046,0.007906464850524606,-0.004575141618801825,0.034117703468017915,1.8682535373097813,0.11094710423607979,0.15625,0.7061334636769161,2.9661510946494603,fair_latency_025,0.33284131270823936
++time_to_absorb,fair_latency_025,7,28.428571428571427,29.571428571428573,1.1428571428571428,1.9448169387297274,-3.6159384728281254,5.9016527585424114,0.5876425282492701,0.578201730081321,1.0,0.22210799850754528,4.0201005025125625,fair_latency_025,0.578201730081321
++spread_bp,fair_latency_025,7,10.950967182666245,12.086163958836295,1.135196776170052,0.840943816431864,-0.9225186144041635,3.192912166744268,1.3499079890814911,0.22574435858849767,0.296875,0.5102172617041314,10.366178230968494,fair_latency_025,0.45148871717699535
++n_bubbles,fair_latency_025,7,0.0,0.0,0.0,0.0,0.0,0.0,,,1.0,,,fair_latency_025,
++max_dd,fair_latency_025,7,0.022262207437842796,0.026711767394822346,0.004449559956979551,0.0012283252192330817,0.0014439564209778686,0.007455163492981233,3.6224608005343097,0.011064523226606303,0.03125,1.3691614874705338,19.98705640221414,fair_latency_025,0.04425809290642521

added results/tables/tab_arms_effects.tex +78 −0

++\begin{tabular}{llrrrrr}
++\toprule
++Cell & Outcome & Control & Effect & 95\% CI & $d$ & $p$ (Holm) \\
++\midrule
++0\% (control) & Mean $|P/V-1|$ (\%) & 0.50 & 0.23 & [-0.24, 0.70] & 0.45 & 0.612 \\
++ & Time to absorb (ticks) & 28.43 & 8.00 & [-21.43, 37.43] & 0.25 & 0.612 \\
++ & Quoted spread (bp) & 10.95 & 7.66$^{**}$ & [2.89, 12.42] & 1.49 & 0.031 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01 & [-0.00, 0.02] & 0.54 & 0.612 \\
++\addlinespace
++25\% diverse prompts & Mean $|P/V-1|$ (\%) & 0.50 & 0.04$^{**}$ & [0.01, 0.06] & 1.36 & 0.045 \\
++ & Time to absorb (ticks) & 28.43 & -4.43$^{*}$ & [-8.35, -0.51] & -1.05 & 0.098 \\
++ & Quoted spread (bp) & 10.95 & 0.52 & [-0.39, 1.42] & 0.53 & 0.211 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.01] & 0.75 & 0.186 \\
++\addlinespace
++25\% two models & Mean $|P/V-1|$ (\%) & 0.50 & 0.01 & [-0.01, 0.03] & 0.29 & 0.469 \\
++ & Time to absorb (ticks) & 28.43 & -4.43$^{**}$ & [-8.00, -0.85] & -1.15 & 0.046 \\
++ & Quoted spread (bp) & 10.95 & 4.62$^{***}$ & [2.61, 6.64] & 2.12 & 0.005 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01$^{***}$ & [0.00, 0.01] & 1.90 & 0.007 \\
++\addlinespace
++25\% memory (5) & Mean $|P/V-1|$ (\%) & 0.50 & 0.04$^{**}$ & [0.02, 0.06] & 1.63 & 0.015 \\
++ & Time to absorb (ticks) & 28.43 & 2.71 & [-2.56, 7.99] & 0.48 & 0.510 \\
++ & Quoted spread (bp) & 10.95 & -0.24 & [-1.30, 0.81] & -0.21 & 0.593 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00$^{***}$ & [0.00, 0.01] & 2.05 & 0.006 \\
++\addlinespace
++25\% communication channel & Mean $|P/V-1|$ (\%) & 0.50 & 0.05$^{***}$ & [0.03, 0.07] & 2.35 & 0.003 \\
++ & Time to absorb (ticks) & 28.43 & -0.43 & [-2.68, 1.83] & -0.18 & 0.666 \\
++ & Quoted spread (bp) & 10.95 & 0.68 & [-0.54, 1.90] & 0.51 & 0.666 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.01] & 0.46 & 0.666 \\
++\addlinespace
++25\% adaptive feedback & Mean $|P/V-1|$ (\%) & 0.50 & -0.00 & [-0.03, 0.02] & -0.02 & 1.000 \\
++ & Time to absorb (ticks) & 28.43 & -1.86 & [-9.42, 5.70] & -0.23 & 1.000 \\
++ & Quoted spread (bp) & 10.95 & -0.33 & [-2.13, 1.48] & -0.17 & 1.000 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [-0.00, 0.00] & 0.09 & 1.000 \\
++\addlinespace
++0\% + leveraged 19\% (classical) & Mean $|P/V-1|$ (\%) & 0.50 & 1.67$^{**}$ & [0.58, 2.75] & 1.42 & 0.038 \\
++ & Time to absorb (ticks) & 28.43 & 24.57 & [-8.14, 57.29] & 0.69 & 0.188 \\
++ & Quoted spread (bp) & 10.95 & 27.87$^{**}$ & [8.90, 46.85] & 1.36 & 0.038 \\
++ & Bubble episodes & 0.00 & 0.71 & [-0.17, 1.59] & 0.75 & 0.188 \\
++ & Max drawdown & 0.02 & 0.05$^{**}$ & [0.02, 0.08] & 1.50 & 0.037 \\
++\addlinespace
++25\% + leveraged 19\% & Mean $|P/V-1|$ (\%) & 0.50 & 0.46$^{***}$ & [0.36, 0.56] & 4.12 & 0.000 \\
++ & Time to absorb (ticks) & 28.43 & -12.14$^{**}$ & [-23.60, -0.69] & -0.98 & 0.041 \\
++ & Quoted spread (bp) & 10.95 & 27.56$^{***}$ & [19.31, 35.81] & 3.09 & 0.000 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.03$^{***}$ & [0.03, 0.04] & 4.24 & 0.000 \\
++\addlinespace
++25\% + circuit breaker & Mean $|P/V-1|$ (\%) & 0.50 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Time to absorb (ticks) & 28.43 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Quoted spread (bp) & 10.95 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00 & [0.00, 0.00] & -- & -- \\
++\addlinespace
++25\% + position limit & Mean $|P/V-1|$ (\%) & 0.50 & 0.02 & [-0.04, 0.07] & 0.27 & 1.000 \\
++ & Time to absorb (ticks) & 28.43 & 5.57 & [-8.04, 19.18] & 0.38 & 1.000 \\
++ & Quoted spread (bp) & 10.95 & 11.53$^{**}$ & [5.15, 17.90] & 1.67 & 0.018 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & -0.00 & [-0.01, 0.00] & -0.13 & 1.000 \\
++\addlinespace
++5\% gpt-5.4-mini (equal USD) & Mean $|P/V-1|$ (\%) & 0.50 & 0.14$^{*}$ & [0.02, 0.27] & 1.04 & 0.067 \\
++ & Time to absorb (ticks) & 28.43 & -8.86$^{*}$ & [-15.49, -2.22] & -1.23 & 0.051 \\
++ & Quoted spread (bp) & 10.95 & 11.25$^{***}$ & [5.85, 16.64] & 1.93 & 0.009 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.01$^{*}$ & [-0.00, 0.02] & 0.79 & 0.082 \\
++\addlinespace
++25\% matched fast latency & Mean $|P/V-1|$ (\%) & 0.50 & 0.01 & [-0.00, 0.03] & 0.71 & 0.333 \\
++ & Time to absorb (ticks) & 28.43 & 1.14 & [-3.62, 5.90] & 0.22 & 0.578 \\
++ & Quoted spread (bp) & 10.95 & 1.14 & [-0.92, 3.19] & 0.51 & 0.451 \\
++ & Bubble episodes & 0.00 & 0.00 & [0.00, 0.00] & -- & -- \\
++ & Max drawdown & 0.02 & 0.00$^{**}$ & [0.00, 0.01] & 1.37 & 0.044 \\
++\addlinespace
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added results/tables/tab_coordination.tex +16 −0

++\begin{tabular}{lrrrrrr}
++\toprule
++Statistic & 5\% & 10\% & 25\% & 50\% & 100\% & Trend per unit share (SE) \\
++\midrule
++Pairwise order-flow correlation (LLM) & 0.475 & 0.420 & 0.399 & 0.389 & 0.338 & -0.117 (0.024)$^{***}$ \\
++Synchrony (LLM) & 0.889 & 0.938 & 0.948 & 0.910 & 0.900 & -0.019 (0.025) \\
++Flow correlation from trades (LLM) & 0.409 & 0.352 & 0.205 & 0.052 & -0.008 & -0.429 (0.021)$^{***}$ \\
++Hold rate & 0.142 & 0.112 & 0.118 & 0.108 & 0.102 & -0.029 (0.023) \\
++Valid-decision rate & 0.981 & 0.964 & 0.981 & 0.973 & 0.957 & -0.019 (0.009)$^{**}$ \\
++\midrule
++Flow correlation, fundamentalist (0\% cell) & \multicolumn{5}{r}{0.531} & \\
++Flow correlation, trend (0\% cell) & \multicolumn{5}{r}{0.757} & \\
++Flow correlation, noise (0\% cell) & \multicolumn{5}{r}{0.035} & \\
++Flow correlation, market maker (0\% cell) & \multicolumn{5}{r}{0.364} & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added results/tables/tab_dose_conf.tex +23 −0

++\begin{tabular}{lrrrrrr}
++\toprule
++Outcome & 0\% & 5\% & 10\% & 25\% & 50\% & 100\% \\
++\midrule
++Sq. log pricing error ($\times 10^{4}$) & 1.00 & 0.47 & 0.52 & 0.47 & 0.52 & 0.61 \\
++Mean $|P/V-1|$ (\%) & 0.73 & 0.52 & 0.56 & 0.50 & 0.51 & 0.54 \\
++Time to absorb (ticks) & 36.43 & 25.00 & 20.29 & 28.43 & 35.57 & 39.71 \\
++Quoted spread (bp) & 18.61 & 15.09$^{*}$ & 15.80$^{**}$ & 10.95$^{**}$ & 7.38$^{***}$ & 6.07$^{***}$ \\
++Price impact (bp/100 lots) & 9.32 & 8.03 & 8.77 & 5.21 & 5.40 & 3.61$^{**}$ \\
++Bubble episodes & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 \\
++Max drawdown & 0.03 & 0.03 & 0.03 & 0.02 & 0.02 & 0.02 \\
++Vol. clustering $\sum_{1}^{10}\mathrm{ACF}|r|$ & 2.22 & 1.72 & 1.79 & 2.27 & 2.73 & 3.31 \\
++LLM order-flow corr. & -- & 0.48 & 0.42 & 0.40 & 0.39 & 0.34 \\
++LLM valid-decision rate & -- & 0.98 & 0.96 & 0.98 & 0.97 & 0.96 \\
++\midrule
++Realized share of traders & 0.000 & 0.050 & 0.100 & 0.250 & 0.500 & 0.950 \\
++Realized share of orders & 0.000 & 0.009 & 0.018 & 0.048 & 0.109 & 0.355 \\
++Realized share of volume & 0.000 & 0.056 & 0.109 & 0.248 & 0.432 & 0.837 \\
++API cost per run (USD) & -- & 0.082 & 0.170 & 0.331 & 0.821 & 1.779 \\
++\midrule
++Seeds per cell & 7 & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

modified results/tables/tab_dose_pilot.tex +13 −13

				`@@ -2,21 +2,21 @@`
2	\toprule	2	\toprule
3	Outcome & 5\% & 25\% & 50\% \\	3	Outcome & 5\% & 25\% & 50\% \\
4	\midrule	4	\midrule
5	−Sq. log pricing error ($\times 10^{4}$) & 0.66 & 0.55 & 2.15 \\	5	+Sq. log pricing error ($\times 10^{4}$) & 0.66 & 0.55 & 0.47 \\
6	−Mean $\|P/V-1\|$ (\%) & 0.54 & 0.48 & 0.69 \\	6	+Mean $\|P/V-1\|$ (\%) & 0.54 & 0.48 & 0.43 \\
7	−Time to absorb (ticks) & 21.67 & 27.00 & 22.00 \\	7	+Time to absorb (ticks) & 21.67 & 27.00 & 21.33 \\
8	−Quoted spread (bp) & 17.93 & 11.58 & 11.27 \\	8	+Quoted spread (bp) & 17.93 & 11.58 & 7.76 \\
9	−Price impact (bp/100 lots) & 10.34 & 8.08 & 13.41 \\	9	+Price impact (bp/100 lots) & 10.34 & 8.08 & 6.68 \\
10	−Bubble episodes & 0.00 & 0.00 & 1.00 \\	10	+Bubble episodes & 0.00 & 0.00 & 0.00 \\
11	−Max drawdown & 0.03 & 0.02 & 0.03 \\	11	+Max drawdown & 0.03 & 0.02 & 0.02 \\
12	−Vol. clustering $\sum_{1}^{10}\mathrm{ACF}\|r\|$ & 2.79 & 2.34 & 3.83 \\	12	+Vol. clustering $\sum_{1}^{10}\mathrm{ACF}\|r\|$ & 2.79 & 2.34 & 3.19 \\
13	−LLM order-flow corr. & 0.38 & 0.41 & 0.20 \\	13	+LLM order-flow corr. & 0.38 & 0.41 & 0.35 \\
14	−LLM valid-decision rate & 0.92 & 0.96 & 0.93 \\	14	+LLM valid-decision rate & 0.92 & 0.96 & 0.97 \\
15	\midrule	15	\midrule
16	Realized share of traders & 0.050 & 0.250 & 0.500 \\	16	Realized share of traders & 0.050 & 0.250 & 0.500 \\
17	−Realized share of orders & 0.008 & 0.046 & 0.086 \\	17	+Realized share of orders & 0.008 & 0.046 & 0.106 \\
18	−Realized share of volume & 0.052 & 0.253 & 0.366 \\	18	+Realized share of volume & 0.052 & 0.253 & 0.446 \\
19	−API cost per run (USD) & 0.088 & 0.436 & 0.462 \\	19	+API cost per run (USD) & 0.088 & 0.436 & 0.888 \\
20	\midrule	20	\midrule
21	Seeds per cell & 3 & & \\	21	Seeds per cell & 3 & & \\
22	\bottomrule	22	\bottomrule

added results/tables/tab_library.tex +12 −0

++\begin{tabular}{lrrrrrrrrrr}
++\toprule
++Cell & Mean $|P/V-1|$ (\%) & Time to absorb (ticks) & Quoted spread (bp) & Depth, 5 levels (lots) & Price impact (bp/100 lots) & Variance ratio (5) & ACF$_1$ of returns & Max drawdown & Realized LLM volume share & LLM valid-decision rate \\
++\midrule
++$N=100$, 0\% & 0.59 & 25.60 & 17.29 & 134.51 & 9.21 & 1.33 & 0.04 & 0.04 & 0.00 & -- \\
++$N=100$, 25\% library & 0.52 & 26.40 & 11.93 & 190.24 & 6.36 & 1.55 & 0.16 & 0.03 & 0.26 & 0.72 \\
++$N=1{,}000$, 0\% & 0.73 & 32.00 & 13.04 & 217.27 & 4.60 & 1.14 & -0.09 & 0.03 & 0.00 & -- \\
++$N=1{,}000$, 25\% library & 0.61 & 40.80 & 5.45 & 1222.83 & 1.18 & 1.89 & 0.21 & 0.02 & 0.63 & 0.61 \\
++\midrule
++Seeds per cell & 5 & & & & & & & & & \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

added results/tables/tab_llm_reliability_conf.csv +16 −0

++cell,n,valid,hold,place,market,json_err,schema_err,feas_err,api_err,lat,in_tok,out_tok,reason_tok,usd
++adaptive_025,4824,0.9761608623548922,0.10219734660033168,0.7848258706467661,0.036898839137645106,0.0,0.0,0.023424543946932008,0.00041459369817578774,2.7546230811371917,848.9463101160862,359.07628524046436,281.6401326699834,0.42632762857142853
++breaker_025,4824,0.9809286898839138,0.1183665008291874,0.7622305140961857,0.04145936981757877,0.0,0.0,0.019071310116086235,0.0,0.0,0.0,0.0,0.0,0.0
++channel_025,4824,0.9838308457711443,0.087893864013267,0.7875207296849088,0.019485903814262025,0.0,0.0,0.01616915422885572,0.0,2.4200344001334964,1062.7259535655057,302.26596185737975,222.5122305140962,0.40685453571428576
++div_models_025,4824,0.9832089552238806,0.2023217247097844,0.7394278606965174,0.17454394693200664,0.0,0.0,0.01658374792703151,0.00020729684908789387,2.623244746028139,855.943407960199,344.46475953565505,267.79187396351574,0.9436430500000001
++div_prompts_025,4824,0.9680762852404643,0.13329187396351574,0.7572553897180763,0.03917910447761194,0.0,0.0,0.031923714759535655,0.0,2.819302460371564,861.0922470978442,352.3410033167496,272.992951907131,0.4221997214285715
++dose_005,963,0.9813084112149533,0.142263759086189,0.8006230529595015,0.05711318795430945,0.0,0.0,0.018691588785046728,0.0,2.566260087597455,851.546209761163,340.6303219106957,260.8172377985462,0.08200593571428572
++dose_010,1929,0.9637117677553136,0.11249351995852773,0.8221876620010368,0.04872991187143598,0.0,0.0,0.036288232244686365,0.0,2.668634628660417,849.784344219803,357.56557801969933,277.6749611197512,0.17000382857142857
++dose_025,4824,0.9809286898839138,0.1183665008291874,0.7622305140961857,0.04145936981757877,0.0,0.0,0.019071310116086235,0.0,2.1728244717523557,653.8884742951907,279.15277777777777,217.96185737976782,0.3305946928571428
++dose_050,9633,0.9729056368732482,0.1077545935845531,0.7080867850098619,0.04193916744524032,0.00010380982040901069,0.0,0.02688674348593377,0.00010380982040901069,2.7813570442340643,780.4892556835877,352.53981106612684,279.7304058963978,0.8212443714285715
++dose_100,18303,0.9566737693274326,0.10227831503032289,0.6688521007485112,0.03890072665683221,0.00010927170409222532,0.0,0.042998415560290666,0.00021854340818445065,3.1543238009138115,861.5051084521663,406.41561492651476,326.105392558597,1.7788438357142857
++fair_latency_025,4824,0.9794776119402985,0.11256218905472637,0.7684494195688225,0.0439469320066335,0.0,0.0,0.0203150912106136,0.00020729684908789387,2.7871667008452095,847.2282338308457,362.7124792703151,283.2649253731343,0.42922315
++fair_mini_005,963,0.9865005192107996,0.18276220145379024,0.8110072689511942,0.5430944963655244,0.0,0.0,0.013499480789200415,0.0,2.724158215312626,852.5368639667705,350.23572170301145,276.30841121495325,0.3047844642857142
++leverage_025,4824,0.9827943615257048,0.06654228855721393,0.8507462686567164,0.07027363184079602,0.0,0.0,0.017205638474295192,0.0,2.882029548260918,848.7464759535656,374.28710613598673,293.6672885572139,0.43940312142857135
++memory_025,4824,0.9877694859038143,0.09266169154228855,0.7906301824212272,0.02197346600331675,0.0,0.0,0.012230514096185738,0.0,2.801034198651676,962.1770315091211,361.65609452736317,283.70812603648426,0.4441563785714286
++poslimit_025,4824,0.960820895522388,0.1046849087893864,0.7935323383084577,0.043532338308457715,0.00020729684908789387,0.0,0.03897180762852405,0.0,2.903965688155319,844.6082089552239,380.1741293532338,299.6455223880597,0.443904

added results/tables/tab_llm_reliability_conf.tex +21 −0

++\begin{tabular}{lrrrrrrrrrr}
++\toprule
++Cell & Decisions & Valid & Hold & Market & JSON err. & Schema err. & Infeasible & API err. & Latency (s) & USD/run \\
++\midrule
++adaptive\_025 & 4,824 & 0.976 & 0.10 & 0.04 & 0.000 & 0.000 & 0.023 & 0.000 & 2.75 & 0.43 \\
++breaker\_025 & 4,824 & 0.981 & 0.12 & 0.04 & 0.000 & 0.000 & 0.019 & 0.000 & 0.00 & 0.00 \\
++channel\_025 & 4,824 & 0.984 & 0.09 & 0.02 & 0.000 & 0.000 & 0.016 & 0.000 & 2.42 & 0.41 \\
++div\_models\_025 & 4,824 & 0.983 & 0.20 & 0.17 & 0.000 & 0.000 & 0.017 & 0.000 & 2.62 & 0.94 \\
++div\_prompts\_025 & 4,824 & 0.968 & 0.13 & 0.04 & 0.000 & 0.000 & 0.032 & 0.000 & 2.82 & 0.42 \\
++dose\_005 & 963 & 0.981 & 0.14 & 0.06 & 0.000 & 0.000 & 0.019 & 0.000 & 2.57 & 0.08 \\
++dose\_010 & 1,929 & 0.964 & 0.11 & 0.05 & 0.000 & 0.000 & 0.036 & 0.000 & 2.67 & 0.17 \\
++dose\_025 & 4,824 & 0.981 & 0.12 & 0.04 & 0.000 & 0.000 & 0.019 & 0.000 & 2.17 & 0.33 \\
++dose\_050 & 9,633 & 0.973 & 0.11 & 0.04 & 0.000 & 0.000 & 0.027 & 0.000 & 2.78 & 0.82 \\
++dose\_100 & 18,303 & 0.957 & 0.10 & 0.04 & 0.000 & 0.000 & 0.043 & 0.000 & 3.15 & 1.78 \\
++fair\_latency\_025 & 4,824 & 0.979 & 0.11 & 0.04 & 0.000 & 0.000 & 0.020 & 0.000 & 2.79 & 0.43 \\
++fair\_mini\_005 & 963 & 0.987 & 0.18 & 0.54 & 0.000 & 0.000 & 0.013 & 0.000 & 2.72 & 0.30 \\
++leverage\_025 & 4,824 & 0.983 & 0.07 & 0.07 & 0.000 & 0.000 & 0.017 & 0.000 & 2.88 & 0.44 \\
++memory\_025 & 4,824 & 0.988 & 0.09 & 0.02 & 0.000 & 0.000 & 0.012 & 0.000 & 2.80 & 0.44 \\
++poslimit\_025 & 4,824 & 0.961 & 0.10 & 0.04 & 0.000 & 0.000 & 0.039 & 0.000 & 2.90 & 0.44 \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

modified results/tables/tab_llm_reliability_pilot.csv +2 −2

				`@@ -1,4 +1,4 @@`
1	cell,n,valid,hold,place,market,json_err,schema_err,feas_err,api_err,lat,in_tok,out_tok,reason_tok,usd	1	cell,n,valid,hold,place,market,json_err,schema_err,feas_err,api_err,lat,in_tok,out_tok,reason_tok,usd
2	−share_25,2066,0.9583736689254598,0.12391093901258471,0.7613746369796709,0.04017424975798645,0.0,0.0,0.041626331074540175,0.0,2.843962665003406,855.4244917715392,369.75556631171344,289.44433688286546,0.4361183833333333	2	+share_25,2066,0.9583736689254598,0.12391093901258471,0.7613746369796709,0.04017424975798645,0.0,0.0,0.041626331074540175,0.0,2.843962665003406,855.4244917715392,369.75556631171344,289.44433688286546,0.43611838333333336
3	share_5,411,0.9221411192214112,0.13138686131386862,0.8272506082725061,0.0535279805352798,0.0,0.0,0.07785888077858881,0.0,2.8592208875806318,855.5206812652068,375.4963503649635,295.345498783455,0.08774501666666668	3	share_5,411,0.9221411192214112,0.13138686131386862,0.8272506082725061,0.0535279805352798,0.0,0.0,0.07785888077858881,0.0,2.8592208875806318,855.5206812652068,375.4963503649635,295.345498783455,0.08774501666666668
4	−share_50,4128,0.9338662790697675,0.3793604651162791,0.5625,0.09278100775193798,0.013565891472868217,0.008478682170542635,0.04408914728682171,0.0,1.4735181390779597,622.8725775193799,208.65213178294573,159.38735465116278,0.46151500000000006	4	+share_50,4128,0.966327519379845,0.12427325581395349,0.690406976744186,0.03682170542635659,0.00024224806201550387,0.0,0.033187984496124034,0.00024224806201550387,2.826030603510146,854.5809108527131,379.47553294573646,299.6671511627907,0.8878785833333334

modified results/tables/tab_llm_reliability_pilot.tex +1 −1

				`@@ -4,6 +4,6 @@ Cell & Decisions & Valid & Hold & Market & JSON err. & Schema err. & Infeasible`
4	\midrule	4	\midrule
5	share\_25 & 2,066 & 0.958 & 0.12 & 0.04 & 0.000 & 0.000 & 0.042 & 0.000 & 2.84 & 0.44 \\	5	share\_25 & 2,066 & 0.958 & 0.12 & 0.04 & 0.000 & 0.000 & 0.042 & 0.000 & 2.84 & 0.44 \\
6	share\_5 & 411 & 0.922 & 0.13 & 0.05 & 0.000 & 0.000 & 0.078 & 0.000 & 2.86 & 0.09 \\	6	share\_5 & 411 & 0.922 & 0.13 & 0.05 & 0.000 & 0.000 & 0.078 & 0.000 & 2.86 & 0.09 \\
7	−share\_50 & 4,128 & 0.934 & 0.38 & 0.09 & 0.014 & 0.008 & 0.044 & 0.000 & 1.47 & 0.46 \\	7	+share\_50 & 4,128 & 0.966 & 0.12 & 0.04 & 0.000 & 0.000 & 0.033 & 0.000 & 2.83 & 0.89 \\
8	\bottomrule	8	\bottomrule
9	\end{tabular}	9	\end{tabular}
	\ No newline at end of file

added results/tables/tab_mechanism_association.tex +20 −0

++\begin{tabular}{llrrr}
++\toprule
++Outcome & Mediator (per 1 SD) & Coefficient & SE & $p$ \\
++\midrule
++mad pct & pairwise corr & +0.011 & 0.019 & 0.545 \\
++mad pct & synchrony & +0.027 & 0.015 & 0.080 \\
++msd log & pairwise corr & +0.000 & 0.000 & 0.269 \\
++msd log & synchrony & +0.000 & 0.000 & 0.220 \\
++time to absorb & pairwise corr & +1.919 & 1.186 & 0.106 \\
++time to absorb & synchrony & +1.068 & 1.575 & 0.498 \\
++n bubbles & pairwise corr & +0.000 & 0.000 & nan \\
++n bubbles & synchrony & +0.000 & 0.000 & nan \\
++max dd & pairwise corr & -0.001 & 0.001 & 0.162 \\
++max dd & synchrony & +0.001 & 0.001 & 0.233 \\
++spread bp & pairwise corr & -0.361 & 1.140 & 0.751 \\
++spread bp & synchrony & +1.053 & 1.032 & 0.307 \\
++lambda bp 100 & pairwise corr & +0.005 & 0.748 & 0.994 \\
++lambda bp 100 & synchrony & +0.820 & 0.724 & 0.257 \\
++\bottomrule
++\end{tabular}
+ \ No newline at end of file

modified results/tables/tab_threshold_sensitivity.tex +7 −1

				`@@ -10,7 +10,13 @@ classical confirmatory & mock llm 50 indep & n bubbles & +0.00 & +0.00 & +0.00 &`
10	classical confirmatory & mock llm 50 indep & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	10	classical confirmatory & mock llm 50 indep & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
11	classical confirmatory & trend heavy & n bubbles & +1.10 & +1.10 & +1.10 & +0.13 & +0.13 & +0.13 & +0.00 & +0.00 & +0.00 \\	11	classical confirmatory & trend heavy & n bubbles & +1.10 & +1.10 & +1.10 & +0.13 & +0.13 & +0.13 & +0.00 & +0.00 & +0.00 \\
12	classical confirmatory & trend heavy & n crashes & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 \\	12	classical confirmatory & trend heavy & n crashes & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 & +0.07 & +0.00 & +0.00 \\
13	−confirmatory openai & dose 025 & n bubbles & -0.17 & -0.17 & -0.17 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	13	+confirmatory openai & dose 025 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
14	confirmatory openai & dose 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\	14	confirmatory openai & dose 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		15	+confirmatory openai & dose 050 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		16	+confirmatory openai & dose 050 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		17	+confirmatory openai & dose 100 & n bubbles & -0.14 & -0.14 & -0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		18	+confirmatory openai & dose 100 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		19	+confirmatory openai & leverage 025 & n bubbles & +0.14 & +0.14 & +0.14 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
		20	+confirmatory openai & leverage 025 & n crashes & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 & +0.00 \\
15	\bottomrule	21	\bottomrule
16	\end{tabular}	22	\end{tabular}
	\ No newline at end of file

modified results/tables/threshold_sensitivity.csv +108 −27

				`@@ -107,30 +107,111 @@ classical_confirmatory,0.08,0.15,trend_heavy,n_crashes,0.0,0.0,,30`
107	classical_confirmatory,0.08,0.15,mock_llm_50_indep,n_bubbles,0.0,0.0,,30	107	classical_confirmatory,0.08,0.15,mock_llm_50_indep,n_bubbles,0.0,0.0,,30
108	classical_confirmatory,0.08,0.15,mock_llm_50_indep,bubble_amp,0.0,0.0,,30	108	classical_confirmatory,0.08,0.15,mock_llm_50_indep,bubble_amp,0.0,0.0,,30
109	classical_confirmatory,0.08,0.15,mock_llm_50_indep,n_crashes,0.0,0.0,,30	109	classical_confirmatory,0.08,0.15,mock_llm_50_indep,n_crashes,0.0,0.0,,30
110	−confirmatory_openai,0.03,0.07,dose_025,n_bubbles,0.16666666666666666,-0.16666666666666666,0.3632174676491226,6	110	+confirmatory_openai,0.03,0.07,dose_025,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
111	−confirmatory_openai,0.03,0.07,dose_025,bubble_amp,0.005669657734009163,-0.005669657734009163,0.3632174676491227,6	111	+confirmatory_openai,0.03,0.07,dose_025,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
112	−confirmatory_openai,0.03,0.07,dose_025,n_crashes,0.0,0.0,,6	112	+confirmatory_openai,0.03,0.07,dose_025,n_crashes,0.0,0.0,,7
113	−confirmatory_openai,0.03,0.1,dose_025,n_bubbles,0.16666666666666666,-0.16666666666666666,0.3632174676491226,6	113	+confirmatory_openai,0.03,0.07,dose_050,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
114	−confirmatory_openai,0.03,0.1,dose_025,bubble_amp,0.005669657734009163,-0.005669657734009163,0.3632174676491227,6	114	+confirmatory_openai,0.03,0.07,dose_050,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
115	−confirmatory_openai,0.03,0.1,dose_025,n_crashes,0.0,0.0,,6	115	+confirmatory_openai,0.03,0.07,dose_050,n_crashes,0.0,0.0,,7
116	−confirmatory_openai,0.03,0.15,dose_025,n_bubbles,0.16666666666666666,-0.16666666666666666,0.3632174676491226,6	116	+confirmatory_openai,0.03,0.07,dose_100,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
117	−confirmatory_openai,0.03,0.15,dose_025,bubble_amp,0.005669657734009163,-0.005669657734009163,0.3632174676491227,6	117	+confirmatory_openai,0.03,0.07,dose_100,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
118	−confirmatory_openai,0.03,0.15,dose_025,n_crashes,0.0,0.0,,6	118	+confirmatory_openai,0.03,0.07,dose_100,n_crashes,0.0,0.0,,7
119	−confirmatory_openai,0.05,0.07,dose_025,n_bubbles,0.0,0.0,,6	119	+confirmatory_openai,0.03,0.07,leverage_025,n_bubbles,0.14285714285714285,0.14285714285714285,0.6036450565101361,7
120	−confirmatory_openai,0.05,0.07,dose_025,bubble_amp,0.0,0.0,,6	120	+confirmatory_openai,0.03,0.07,leverage_025,bubble_amp,0.004859706629150711,0.005624499649447477,0.5681429646056435,7
121	−confirmatory_openai,0.05,0.07,dose_025,n_crashes,0.0,0.0,,6	121	+confirmatory_openai,0.03,0.07,leverage_025,n_crashes,0.0,0.0,,7
122	−confirmatory_openai,0.05,0.1,dose_025,n_bubbles,0.0,0.0,,6	122	+confirmatory_openai,0.03,0.1,dose_025,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
123	−confirmatory_openai,0.05,0.1,dose_025,bubble_amp,0.0,0.0,,6	123	+confirmatory_openai,0.03,0.1,dose_025,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
124	−confirmatory_openai,0.05,0.1,dose_025,n_crashes,0.0,0.0,,6	124	+confirmatory_openai,0.03,0.1,dose_025,n_crashes,0.0,0.0,,7
125	−confirmatory_openai,0.05,0.15,dose_025,n_bubbles,0.0,0.0,,6	125	+confirmatory_openai,0.03,0.1,dose_050,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
126	−confirmatory_openai,0.05,0.15,dose_025,bubble_amp,0.0,0.0,,6	126	+confirmatory_openai,0.03,0.1,dose_050,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
127	−confirmatory_openai,0.05,0.15,dose_025,n_crashes,0.0,0.0,,6	127	+confirmatory_openai,0.03,0.1,dose_050,n_crashes,0.0,0.0,,7
128	−confirmatory_openai,0.08,0.07,dose_025,n_bubbles,0.0,0.0,,6	128	+confirmatory_openai,0.03,0.1,dose_100,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
129	−confirmatory_openai,0.08,0.07,dose_025,bubble_amp,0.0,0.0,,6	129	+confirmatory_openai,0.03,0.1,dose_100,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
130	−confirmatory_openai,0.08,0.07,dose_025,n_crashes,0.0,0.0,,6	130	+confirmatory_openai,0.03,0.1,dose_100,n_crashes,0.0,0.0,,7
131	−confirmatory_openai,0.08,0.1,dose_025,n_bubbles,0.0,0.0,,6	131	+confirmatory_openai,0.03,0.1,leverage_025,n_bubbles,0.14285714285714285,0.14285714285714285,0.6036450565101361,7
132	−confirmatory_openai,0.08,0.1,dose_025,bubble_amp,0.0,0.0,,6	132	+confirmatory_openai,0.03,0.1,leverage_025,bubble_amp,0.004859706629150711,0.005624499649447477,0.5681429646056435,7
133	−confirmatory_openai,0.08,0.1,dose_025,n_crashes,0.0,0.0,,6	133	+confirmatory_openai,0.03,0.1,leverage_025,n_crashes,0.0,0.0,,7
134	−confirmatory_openai,0.08,0.15,dose_025,n_bubbles,0.0,0.0,,6	134	+confirmatory_openai,0.03,0.15,dose_025,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
135	−confirmatory_openai,0.08,0.15,dose_025,bubble_amp,0.0,0.0,,6	135	+confirmatory_openai,0.03,0.15,dose_025,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
136	−confirmatory_openai,0.08,0.15,dose_025,n_crashes,0.0,0.0,,6	136	+confirmatory_openai,0.03,0.15,dose_025,n_crashes,0.0,0.0,,7
		137	+confirmatory_openai,0.03,0.15,dose_050,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
		138	+confirmatory_openai,0.03,0.15,dose_050,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
		139	+confirmatory_openai,0.03,0.15,dose_050,n_crashes,0.0,0.0,,7
		140	+confirmatory_openai,0.03,0.15,dose_100,n_bubbles,0.14285714285714285,-0.14285714285714285,0.3559176837495823,7
		141	+confirmatory_openai,0.03,0.15,dose_100,bubble_amp,0.004859706629150711,-0.004859706629150711,0.3559176837495823,7
		142	+confirmatory_openai,0.03,0.15,dose_100,n_crashes,0.0,0.0,,7
		143	+confirmatory_openai,0.03,0.15,leverage_025,n_bubbles,0.14285714285714285,0.14285714285714285,0.6036450565101361,7
		144	+confirmatory_openai,0.03,0.15,leverage_025,bubble_amp,0.004859706629150711,0.005624499649447477,0.5681429646056435,7
		145	+confirmatory_openai,0.03,0.15,leverage_025,n_crashes,0.0,0.0,,7
		146	+confirmatory_openai,0.05,0.07,dose_025,n_bubbles,0.0,0.0,,7
		147	+confirmatory_openai,0.05,0.07,dose_025,bubble_amp,0.0,0.0,,7
		148	+confirmatory_openai,0.05,0.07,dose_025,n_crashes,0.0,0.0,,7
		149	+confirmatory_openai,0.05,0.07,dose_050,n_bubbles,0.0,0.0,,7
		150	+confirmatory_openai,0.05,0.07,dose_050,bubble_amp,0.0,0.0,,7
		151	+confirmatory_openai,0.05,0.07,dose_050,n_crashes,0.0,0.0,,7
		152	+confirmatory_openai,0.05,0.07,dose_100,n_bubbles,0.0,0.0,,7
		153	+confirmatory_openai,0.05,0.07,dose_100,bubble_amp,0.0,0.0,,7
		154	+confirmatory_openai,0.05,0.07,dose_100,n_crashes,0.0,0.0,,7
		155	+confirmatory_openai,0.05,0.07,leverage_025,n_bubbles,0.0,0.0,,7
		156	+confirmatory_openai,0.05,0.07,leverage_025,bubble_amp,0.0,0.0,,7
		157	+confirmatory_openai,0.05,0.07,leverage_025,n_crashes,0.0,0.0,,7
		158	+confirmatory_openai,0.05,0.1,dose_025,n_bubbles,0.0,0.0,,7
		159	+confirmatory_openai,0.05,0.1,dose_025,bubble_amp,0.0,0.0,,7
		160	+confirmatory_openai,0.05,0.1,dose_025,n_crashes,0.0,0.0,,7
		161	+confirmatory_openai,0.05,0.1,dose_050,n_bubbles,0.0,0.0,,7
		162	+confirmatory_openai,0.05,0.1,dose_050,bubble_amp,0.0,0.0,,7
		163	+confirmatory_openai,0.05,0.1,dose_050,n_crashes,0.0,0.0,,7
		164	+confirmatory_openai,0.05,0.1,dose_100,n_bubbles,0.0,0.0,,7
		165	+confirmatory_openai,0.05,0.1,dose_100,bubble_amp,0.0,0.0,,7
		166	+confirmatory_openai,0.05,0.1,dose_100,n_crashes,0.0,0.0,,7
		167	+confirmatory_openai,0.05,0.1,leverage_025,n_bubbles,0.0,0.0,,7
		168	+confirmatory_openai,0.05,0.1,leverage_025,bubble_amp,0.0,0.0,,7
		169	+confirmatory_openai,0.05,0.1,leverage_025,n_crashes,0.0,0.0,,7
		170	+confirmatory_openai,0.05,0.15,dose_025,n_bubbles,0.0,0.0,,7
		171	+confirmatory_openai,0.05,0.15,dose_025,bubble_amp,0.0,0.0,,7
		172	+confirmatory_openai,0.05,0.15,dose_025,n_crashes,0.0,0.0,,7
		173	+confirmatory_openai,0.05,0.15,dose_050,n_bubbles,0.0,0.0,,7
		174	+confirmatory_openai,0.05,0.15,dose_050,bubble_amp,0.0,0.0,,7
		175	+confirmatory_openai,0.05,0.15,dose_050,n_crashes,0.0,0.0,,7
		176	+confirmatory_openai,0.05,0.15,dose_100,n_bubbles,0.0,0.0,,7
		177	+confirmatory_openai,0.05,0.15,dose_100,bubble_amp,0.0,0.0,,7
		178	+confirmatory_openai,0.05,0.15,dose_100,n_crashes,0.0,0.0,,7
		179	+confirmatory_openai,0.05,0.15,leverage_025,n_bubbles,0.0,0.0,,7
		180	+confirmatory_openai,0.05,0.15,leverage_025,bubble_amp,0.0,0.0,,7
		181	+confirmatory_openai,0.05,0.15,leverage_025,n_crashes,0.0,0.0,,7
		182	+confirmatory_openai,0.08,0.07,dose_025,n_bubbles,0.0,0.0,,7
		183	+confirmatory_openai,0.08,0.07,dose_025,bubble_amp,0.0,0.0,,7
		184	+confirmatory_openai,0.08,0.07,dose_025,n_crashes,0.0,0.0,,7
		185	+confirmatory_openai,0.08,0.07,dose_050,n_bubbles,0.0,0.0,,7
		186	+confirmatory_openai,0.08,0.07,dose_050,bubble_amp,0.0,0.0,,7
		187	+confirmatory_openai,0.08,0.07,dose_050,n_crashes,0.0,0.0,,7
		188	+confirmatory_openai,0.08,0.07,dose_100,n_bubbles,0.0,0.0,,7
		189	+confirmatory_openai,0.08,0.07,dose_100,bubble_amp,0.0,0.0,,7
		190	+confirmatory_openai,0.08,0.07,dose_100,n_crashes,0.0,0.0,,7
		191	+confirmatory_openai,0.08,0.07,leverage_025,n_bubbles,0.0,0.0,,7
		192	+confirmatory_openai,0.08,0.07,leverage_025,bubble_amp,0.0,0.0,,7
		193	+confirmatory_openai,0.08,0.07,leverage_025,n_crashes,0.0,0.0,,7
		194	+confirmatory_openai,0.08,0.1,dose_025,n_bubbles,0.0,0.0,,7
		195	+confirmatory_openai,0.08,0.1,dose_025,bubble_amp,0.0,0.0,,7
		196	+confirmatory_openai,0.08,0.1,dose_025,n_crashes,0.0,0.0,,7
		197	+confirmatory_openai,0.08,0.1,dose_050,n_bubbles,0.0,0.0,,7
		198	+confirmatory_openai,0.08,0.1,dose_050,bubble_amp,0.0,0.0,,7
		199	+confirmatory_openai,0.08,0.1,dose_050,n_crashes,0.0,0.0,,7
		200	+confirmatory_openai,0.08,0.1,dose_100,n_bubbles,0.0,0.0,,7
		201	+confirmatory_openai,0.08,0.1,dose_100,bubble_amp,0.0,0.0,,7
		202	+confirmatory_openai,0.08,0.1,dose_100,n_crashes,0.0,0.0,,7
		203	+confirmatory_openai,0.08,0.1,leverage_025,n_bubbles,0.0,0.0,,7
		204	+confirmatory_openai,0.08,0.1,leverage_025,bubble_amp,0.0,0.0,,7
		205	+confirmatory_openai,0.08,0.1,leverage_025,n_crashes,0.0,0.0,,7
		206	+confirmatory_openai,0.08,0.15,dose_025,n_bubbles,0.0,0.0,,7
		207	+confirmatory_openai,0.08,0.15,dose_025,bubble_amp,0.0,0.0,,7
		208	+confirmatory_openai,0.08,0.15,dose_025,n_crashes,0.0,0.0,,7
		209	+confirmatory_openai,0.08,0.15,dose_050,n_bubbles,0.0,0.0,,7
		210	+confirmatory_openai,0.08,0.15,dose_050,bubble_amp,0.0,0.0,,7
		211	+confirmatory_openai,0.08,0.15,dose_050,n_crashes,0.0,0.0,,7
		212	+confirmatory_openai,0.08,0.15,dose_100,n_bubbles,0.0,0.0,,7
		213	+confirmatory_openai,0.08,0.15,dose_100,bubble_amp,0.0,0.0,,7
		214	+confirmatory_openai,0.08,0.15,dose_100,n_crashes,0.0,0.0,,7
		215	+confirmatory_openai,0.08,0.15,leverage_025,n_bubbles,0.0,0.0,,7
		216	+confirmatory_openai,0.08,0.15,leverage_025,bubble_amp,0.0,0.0,,7
		217	+confirmatory_openai,0.08,0.15,leverage_025,n_crashes,0.0,0.0,,7

modified scripts/40_figures.py +4 −4

Large diff (27 lines) is not loaded by default to keep the page fast.Load diff

modified scripts/50_tables.py +23 −3

Large diff (42 lines) is not loaded by default to keep the page fast.Load diff

modified src/market_agents/analysis/plots.py +10 −4

Large diff (27 lines) is not loaded by default to keep the page fast.Load diff

spb/wp27_uqo

Theme