BTC 15-Minute Favorite Continuation

VERDICT: INCONCLUSIVE — PROMISING HOLDOUT, NOT YET SIGNIFICANT

The original underdog-rebound hypothesis failed. Buying either side below 10 cents between minutes 5 and 10 remained negative across every tested threshold and rebound target, including before the conservative execution haircut.

The opposite rule was more promising: buy the 85–95 cent favorite between minutes 7 and 8, exit at 99 cents, stop at 80 cents, or force an exit at minute 14. With a one-cent adverse haircut at both entry and exit, favorite continuation produced PF 1.14 on the first 70% of markets and PF 1.56 on the untouched final 30%.

That supports a frozen-rule forward paper test, not a validated verdict. The combined bootstrap interval crosses zero, the sample spans only six weeks, the placebo was not run, and doubling the execution haircut turns the result negative.

What was tested

• 3,085 usable public Polymarket BTC 15-minute markets from 2026-03-14 through 2026-04-25.
• Chronological split: first 70% for rule-family selection, final 30% untouched.
• One trade per market; entry from 0.85 through 0.95 during minute 7–8.
• Exit at 0.99, stop at 0.80, otherwise first observation at/after minute 14.
• Execution haircut: +0.01 to entry and −0.01 from exit.

Results

Sample	Trades	Win rate	Mean return	Median return	Profit factor	95% bootstrap interval
Training	137	72.3%	+0.66%	+5.35%	1.14	−1.81% to +2.96%
Holdout	45	77.8%	+1.83%	+5.35%	1.56	−1.78% to +4.88%
Combined	182	73.6%	+0.95%	+5.35%	1.21	−1.07% to +2.82%

Why the rebound failed

Contracts below 10 cents usually continued decaying. The best evolved rebound rule still lost about 12% per trade on the holdout before the conservative execution haircut. Entry bands, timing, fresh-low confirmation, stops, and earlier forced exits did not reverse the result.

The sample instead favored persistence. By minutes 7–8, a side already priced at 85–95 cents often remained dominant through minute 14. This is a different archetype, not a cosmetic parameter change.

11-gate scorecard

Gate	Status	Detail
G1 minimum sample	PASS	182 combined observations, including 45 untouched holdout trades.
G2 PF ≥ 1.20 net of costs	PASS	Combined PF 1.212 after the one-cent-per-side haircut.
G3 Sharpe ≥ 0.6	NOT EVALUABLE	Six-week event sample is not an honest annualized equity record.
G4 max drawdown ≤ 12%	NOT EVALUABLE	No preregistered account sizing model.
G5 positive in ≥60% of years	NOT EVALUABLE	Only six weeks of history.
G6 bootstrap lower bound >0	FAIL	Combined 95% interval is −1.07% to +2.82%.
G7 placebo beats p95	NOT RUN	Requires a preregistered market-block placebo.
G8 2× cost PF >1.0	FAIL	Two-cent-per-side stress produces PF 0.752 and −1.25% mean return.
G9 DSR / multiple testing	NOT EVALUABLE	Standard multiple-testing correction was not run.
G10 concentration ≤40%	FAIL	Best week contributes 41.4% of positive weekly P/L.
G11 holdout persistence	PASS	Holdout PF 1.56 exceeds 0.9× training PF 1.14.

Data and execution caveats

The source contains minute-resolution observed outcome prices, not guaranteed fills. Targets can occur between samples, while displayed observations may not be executable at the modeled size. Final validation needs side-specific CLOB bids/asks, depth, actual fees, partial-fill handling, Chainlink opening/current prices, and immutable resolution records.

Next step

Freeze v1 and paper trade it for at least 200 signals or 30 calendar days, whichever is longer. Pass only with at least 100 completed trades, PF ≥1.20 net of actual costs, a positive bootstrap lower bound, and no week contributing more than 40% of profit. Do not tune during collection and do not deploy capital before a clean pass.

---

Generated 2026-07-19. Public historical observations; no claim of executable fills.