← REGISTRY
CASE FILE · COMPARISON-PRICE-ONLY-VS-DEPTH-AWARE
FILL-MODEL COMPARISONEVIDENCE RECORD38 windows · 459 recorded events · captured 2026-07-06
5 → 0
APPARENT TRADES → VERIFIABLE TRADES
Same spec. Same data. Only the honesty of the fill model changed, and the P&L disappeared. Proofline discloses the assumption instead of hiding it.
THINKENIGMATIC-MEANREV (COMPILE RECORD #02)
POLYMARKET BTC 5M UP/DOWN
POLYMARKET BTC 5M UP/DOWN
PROOFLINE
FILL-MODEL COMPARISON
RECORD SEALED
SAME RULES, TWO FILL MODELS10 ATTEMPTS REFUSEDREPRODUCIBLE
WHAT HAPPENED?
This record shows why fill-checking matters. The same rules ran twice on the same saved market data. The loose method assumed every shown price fills and produced a P&L. The honest method demanded proof there was enough available to trade, and the data had none, so it refused every trade. The P&L a backtest shows you can be an artifact of how generously it counts fills.
WHAT WAS TESTED?
Same strategy, two ways of counting fills: Polymarket BTC 5m up/down · 38 windows · 459 recorded events · captured 2026-07-06.
WHY TRUST IT?
Both runs used the same rules and the same saved market data; the only change was how strictly fills were checked. Both runs reproduce exactly when rerun.
WHAT IS STILL UNKNOWN?
Neither number predicts anything. The point is how easily a backtest can manufacture a result by assuming trades fill.
EVIDENCE · EXPAND TO INSPECT
PLAIN STORY FIRST · PROOF UNDERNEATH
The loose method assumes every shown price fills. It produced a P&L out of trades the data cannot prove.
The honest method demands recorded proof there was enough available to trade. This dataset carries none, so every trade was refused.
Positions close against the venue's official resolution, never a guess.
- ✓Trade signals
- ✓Whether trades could go through
- ✓Fees
- ✓Final market results
- ✓Reproducibility
v0 fills (depth_unverified)5
v0.1 fills0
v0.1 rejected (depth_unavailable)10
Real venue settlements applied38
- The price-only arm's fills are labeled depth_unverified by the engine itself; the label marks the weak assumption.
- ARB-085 calibration later measured what price-only firing does against real books: 81.7% fill on Polymarket, 69.2% on Kalshi (see the calibration record).
TECHNICAL PROOF · FOR THOSE WHO WANT IT
NEVER SHOWN BEFORE THE STORY
FINGERPRINTS (HASHES)
REPORT (PRICE-ONLY ARM)
e5953b82…
REPORT (DEPTH-AWARE ARM)
fe533c6a…
REPRODUCIBILITY
Bit-identical reruns confirmed for both arms (price-only report e5953b82…, depth report fe533c6a…).
EVIDENCE FILES · PROOFLINE REPOSITORY
compiler-benchmark/replays/depth-aware-0706/comparison.jsonSEALED
compiler-benchmark/replays/depth-aware-0706/report_depth.jsonSEALED
compiler-benchmark/replays/02-thinkenigmatic-meanrev/report.jsonSEALED
compiler-benchmark/replays/02-thinkenigmatic-meanrev/truth_report.htmlSEALED
compiler-benchmark/replays/STATUS.mdSEALED
Written by the Proofline engine; the site only displays them. See the method for how a record is reproduced.
ENGINE RECORD
COMPARISONREPLAY ONLYBYTE-REPRODUCIBLE
Trust levelREPLAY ONLY
Run statusCOMPLETED
P&L statusCOMPUTABLE
P&L DETAIL
Price-only v0: net−$14.28 (5 fills, all depth_unverified)
Price-only v0: fees / maxDD$1.67 / −$22.97
Depth-aware v0.1: net$0.00 (0 fills; nothing to settle)
Spec from compile record #02 (COMPILED · HIGH · 0 assumed defaults).
VERIFIED UNDER THE DATA AND ASSUMPTIONS SHOWN. NO GUARANTEE OF FUTURE RESULTS.