Crowded longs underperform; crowded shorts do not bounce

Verdict: WATCH · Asset: BTCUSDT global long/short account ratio · Horizon: next 8h · Sample: chronological 60/20/20

Hypothesis: extreme retail long/short account ratio predicts fade — too many longs underperform, too many shorts bounce. Series: count_long_short_ratio from the same Binance vision metrics zips as OI (globalLongShortAccountRatio). Frozen discovery P90=2.572, P10=0.901.

What happened

Q1 HIGH clears. Crowded-long underperforms MID on validation (t=−5.14, −56 bp / 8h) and OOS (t=−2.16).

Q1 LOW fails. No bounce vs MID on val or OOS. The joint two-sided fade that showed up in discovery does not replicate. Prereg required both sides for promotion — so this is not PROMOTED.

Q1b |ratio−1| lifts next |r| +59% on val (vol marker, same family as OI/volume).

Q2: HIGH dummy is not redundant with OI Δ or funding (joint OLS t≈−8.3; overlap with extreme OI 6%, with high funding 0%).

Q3: HIGH-only short survives 5–10 bps in mean (val +37 bp net at 10 bps; OOS +7 bp). Hit rate ~50%. Both-sides book dies at 5 bps OOS because LOW failed.

Frozen HIGH rate halves after discovery (10% → ~4.5%) — nonstationary tail.

Trading

A one-sided crowded-long short is the interesting leftover. It is not a finished strategy: hit rate is a coin, the tail is shrinking, and the prereg two-sided claim failed. Needs a new prereg on HIGH-only before anyone treats it as a book.

Final Verdict

WATCH

Do not fade both tails. Do not ignore crowded longs either — this is the first positioning series in this log that added something incremental to OI and funding on chronological validation.