Trading-strategy family
FUNDING_CONTRA_V1Funding-rate crowding fadeFAIL
tested Oct 24, 2026 2 variants tried
What was tested
When funding has been unusually high for 3 days (the crowd is long), short BTC; when unusually low, go long. “Unusual” means the top or bottom 10% (or 5%) of the previous 180 days. Take profit at 2R, exit after 72 hours at the latest.
Signal from Binance BTCUSDT 8-hour funding; trades priced on Kraken PF_XBTUSD candles. Same costs and R convention as TSMOM_D_V1. IS Oct 3, 2022 to Apr 1, 2025, OOS to Sep 30, 2026.
Pre-registered pass bar and result
The bar was written and hashed before the run. Results are on data the selection step never saw (out-of-sample).
| Criterion | Bar | Result | Verdict |
|---|---|---|---|
| Fee gate: IS gross captured move ≥ 2× the cost wall | ≥ 26.6 bp | −11.1 bp | FAIL |
| IS mean R > 0 and gate passed | > 0 | −0.048 | FAIL |
| OOS mean R above random-direction 95th percentile | > +0.166 | +0.149 (info only) | FAIL |
| OOS mean R > 0 / OOS trades ≥ 30 | > 0 / ≥ 30 | +0.149 / 100 (info only) | INFO |
| Monte Carlo pass probability | ≥ MF4 + 5 pp | 62.7% (info only) | INFO |
Why it failed
It lost money in-sample under both variants (−0.048R and −0.146R) and failed the fee gate, which under the rules ends the test there.
The out-of-sample result (+0.149R) looks good but is still below the random-direction 95th percentile (+0.166R). Adopting it now would mean choosing on the out-of-sample answer, which is exactly what the process forbids. It can only come back as a new FUNDING_CONTRA_V2, pre-registered and tested on data after Oct 24, 2026.
Variants tried
2 variants: top/bottom 10% (q10) or 5% (q05). Both were negative in IS, so the pick (q10) was forced. IS / OOS (OOS information only): q10 −0.048 / +0.149, q05 −0.146 / +0.117.
What this test cannot tell
Process slip, disclosed: the rules said no out-of-sample run after a failed fee gate, but the script computed it anyway. Those numbers are labelled information only and do not affect the verdict.
Rules lock
Rules file SHA-256:
4657ac56e6f5c8eacec015fed9ff07d9f85ba84d06527695ed8d897c9a9c7b39
Locked:
“not recorded” means the number or time was not written down at the time, and we have not filled it in afterwards.
Submit a rule
Have a rule with numbers? Send it to us on X @ProjectAnchorHQ or Telegram t.me/ProjectAnchorHQ (market, entry, exit, stop). Free; we only need the written rule, never keys, accounts or funds. We publish the result pass or fail, and credit you only if you say yes.