IMMUTABLE EVENT EVIDENCE

LLM Cross-sectional Strategy Benchmark

Four one-shot model submissions, one point-in-time dataset, one frozen economic contract and four isolated Event accounts.

Evaluation2025-09-01 → 2026-09-01 UTC

Common friction5 bp + 2 bp

Initial equity100,000 USDT

Verified artifacts36 / 36

01 · LEADERBOARD

After-cost formal results

Positive-return strategies rank ahead of non-positive strategies; within each group the frozen Sharpe-first tie-break applies.

#	Submission / strategy	Total return	Sharpe	Max drawdown	Turnover	Trades	Run ID
1	ChatGPT 6 Astra UltraBuffered Relative Momentum	+42.32%	1.04	-24.67%	196.1×	24,058	`evt-a01175782a0fa6997144d0da` ↗
2	Grok 4.7 XHighResidual Momentum-Carry Buffer	-3.91%	0.16	-43.88%	270.0×	32,200	`evt-c7bdf2892da257893f46be6c` ↗
3	Kimi K3 MaxTri-Blend Carry-Momentum-Reversal	-47.06%	-1.55	-53.15%	535.5×	53,600	`evt-498d8f170a3ee9272dbbb66c` ↗
4	Claude Fable 5.1 MaxTrailing-Funding Carry Reversal	-38.69%	-1.76	-49.03%	538.5×	115,847	`evt-ed4f9aeb18febee2690b12d4` ↗

02 · PATH

Daily ending equity

03 · MONTHLY CONSISTENCY

Compounded monthly returns

Submission	2025-09	2025-10	2025-11	2025-12	2026-01	2026-02	2026-03	2026-04	2026-05	2026-06	2026-07	2026-08	2026-09
ChatGPT 6 Astra Ultra	-0.63%	+15.82%	-10.02%	+7.56%	-1.81%	+8.76%	+5.80%	+12.67%	+0.27%	+14.96%	-4.41%	-8.84%	-0.08%
Grok 4.7 XHigh	-14.31%	+3.88%	+1.69%	+7.75%	+26.14%	+0.31%	-1.62%	+14.17%	-2.29%	-11.88%	-18.44%	-1.20%	-0.08%
Kimi K3 Max	-15.36%	+2.35%	-4.10%	+3.02%	+9.84%	+0.44%	-9.84%	-8.88%	-12.77%	-20.53%	-5.25%	+3.95%	-0.04%
Claude Fable 5.1 Max	+1.70%	-3.44%	-3.97%	-9.53%	+4.86%	-9.22%	+3.92%	-15.60%	-15.19%	+2.22%	-3.02%	+2.45%	-0.08%

The 2026-09 column contains only the frozen terminal-liquidation timestamp and its execution costs.

04 · STRATEGY AUDIT

What each model actually submitted

ChatGPT 6 Astra Ultra

OpenAI · ChatGPT 6 Astra · Effort Ultra

Daily volatility-normalized 28-day relative momentum, skipping the latest day, with 15/15 buffered selections and exact post-cost basket sizing.

Ending equity142,318.69

Avg / max gross0.88× / 1.00×

Fees + slippage17,183.21

Funding cashflow+46,716.70

Risk events2,030

Audit dispositionadmissible_pending_implementation

Risk-event mix and identity

eligibility_loss: 1,641, exposure_repair: 389

evt-a01175782a0fa6997144d0da

Original response SHA-256: 63cd97d05872b6e99354c67dc0928c6688f0e3b853c8a73e8339c571f8865df2

Grok 4.7 XHigh

xAI · Grok 4.7 · Effort XHigh

Daily market-residual momentum plus seven-day funding carry, buffered 15/15 selection and a minute-level exposure enforcer.

Ending equity96,088.10

Avg / max gross0.86× / 1.07×

Fees + slippage20,346.16

Funding cashflow+100,113.68

Risk events600

Audit dispositionadmissible_with_frozen_benchmark_overrides

Risk-event mix and identity

exposure_enforcer: 600

evt-c7bdf2892da257893f46be6c

Original response SHA-256: 23676a65a8ea7cbf2e87c647f9218b56b29ec0196d43bc42f6c0f75c90afcaf4

Kimi K3 Max

Kimi · Kimi K3 · Effort Max

Four-hour robust blend of carry, medium-horizon momentum and 24-hour reversal, with 10/10 selections and a drawdown breaker.

Ending equity52,944.25

Avg / max gross0.51× / 1.15×

Fees + slippage29,373.14

Funding cashflow+64,994.52

Risk events1

Audit dispositionadmissible_with_frozen_benchmark_overrides

Risk-event mix and identity

drawdown_breaker: 1

evt-498d8f170a3ee9272dbbb66c

Original response SHA-256: 53299f9fb77a88e36dafc6cb19822ca2086a8c066db113f0c7501d38e5a9a9c7

Claude Fable 5.1 Max

Anthropic · Claude Fable 5.1 · Effort Max

Hourly trailing-funding carry reversal, inverse-volatility weights, rank holding bands, drift controls and flash/velocity exits.

Ending equity61,307.98

Avg / max gross1.00× / 1.07×

Fees + slippage29,492.01

Funding cashflow+81,641.71

Risk events1,638

Audit dispositionadmissible_with_recorded_contract_noncompliance

Risk-event mix and identity

market_relative_velocity: 1,225, market_relative_flash: 413

evt-ed4f9aeb18febee2690b12d4

Original response SHA-256: 37d6d708492ed6d70a750c3cd7e525d644c98f956caf39f1de0dec2098e185cc

05 · QUALIFICATIONS

How to read this benchmark

This is a backtest comparison, not evidence of live profitability. High turnover, funding dependence, transient exposure excursions and model-specific interpretation defects remain visible rather than being optimized away.

Every response was accepted once; there was no follow-up clarification or parameter search.
Signals use completed causal observations and scheduled fills occur no earlier than the next minute open.
Reported fee, slippage, funding, risk events and exposure paths come from immutable artifacts.
Model names and effort labels are owner-supplied benchmark metadata, not product availability claims.

06 · PROVENANCE

Reproducibility identity

Source commit: 2b1ca8f6826866fcf70e1cc39fefae4c8a668c7f

DatasetSnapshot SHA-256: a49846e623dee5ef16cebe28a48a12e384c07582237a782eeef311208ab75e96

All manifest-listed artifact hashes were recomputed before this report was written.