SIGNALS / CAPITULATION-REVERSION

Capitulation Reversion Monitor

Monitor snapshot · —

Flags symbols whose 24-hour price and base-unit open interest indicate a sharp deleveraging event. The page separates the current trigger state, a static 2024–2026 held-out backtest, and an accumulating paper record. None is a live trade recommendation.

Read this before the numbers. This page separates three things: the current trigger state, a static held-out backtest, and an accumulating paper record. The paper-record summary statistics currently combine labeled historical backfill with trades captured after the live recorder started. Use the Source column to distinguish them. Until the API exposes source-separated statistics, this combined block must not be described as a pure forward holdout.

E1 is the validated symmetric ±5×ATR research profile. The live monitor’s suggested +3×ATR target and catastrophic stop correspond to E2 and differ from the validated headline configuration. Do not attach E1’s headline statistics to the live E2 levels. Nothing here is trading advice.
Live Monitor — symbols currently in capitulation
SymbolTierMode24h Px%24h OI%ATR% EntryTP (+3×ATR)Hard stopTime-stopSizing
Exit profile
Leverage
HELD-OUT BACKTEST — 2 years, real OI, 14 majors
The static study uses 14 majors, real base-unit open interest, a 4-hour grid, next-bar entry, and 0.22% modeled round-trip costs over 2024-03-19 to 2026-03-19. “TEST” is the held-out portion of this one research dataset; it is not live-trading evidence. Always show n, train/test split, regime split, and drawdown next to profit factor or win rate. backtest — not live-trade-proven
Equity curve · E1_L1
Left of the line = TRAIN (breakeven). Right = held-out TEST. 5% equity per trade, compounded, net of fees.
Drawdown (FULL window)
Compounded portfolio drawdown from the running peak.
Statistics · TEST
Per-symbol dispersion — TEST PnL (E1 @ 5×)
ATOM/BTC rows are n=1 — PF=∞ there is a small-sample artifact, not a per-symbol edge.
Threshold sensitivity (E1 @ 5×, TEST)
Direction is consistent; magnitude is noisy — the highest-PF cells have the fewest trades. We run the center cell, not the corner.
—
PAPER RECORD — BACKFILL + LIVE ACCRUAL
Paper record
The summary statistics currently combine labeled historical backfill with trades captured after the live recorder started. Use the Source column to distinguish them. Until the API exposes source-separated statistics, this combined block must not be described as a pure forward holdout.
Paper equity · E1_L1
Backfill left of the line, live right. 5% fixed-fraction, compounded, net of fees.
Paper drawdown
From the running peak of the paper equity.
Combined paper statistics · E1_L1 (resolved paper trades, fees included)
Paper trades
SymbolTierModeSourceEntryEntry px ExitNet %
Held-out backtest vs paper record · E1_L1
x = trade number (not calendar) · both re-based to 100
Same rules, same costs, same profile — backtest held-out vs the live paper record, by trade number (calendar axes would compare 2 years against days). If the live line tracks materially below the backtest line as n grows, the edge is decaying or was regime luck — this chart is designed to show that, not to hide it.
METHODOLOGY & CAVEATS
Current monitor method

For each symbol, the scanner uses the exchange with the most observations so price and base-unit OI come from one venue. A validated-mechanism trigger requires 24-hour OI below −10% and price below −5%. A price-only fallback fires at −12% and is explicitly exploratory. The 24-hour reference accepts a sample within ±3 hours.

The live regime label is only an approximately 7-day BTC-trend proxy because of daemon retention. The historical validation used a 30-day regime. A symbol can remain listed while its trailing 24-hour window still qualifies; this is a state monitor, not a stream of fresh entries.

How to read this
  • Tier distinguishes the 14 validated majors from exploratory symbols.
  • OI+PRICE is the tested mechanism; PRICE-ONLY is a weaker fallback.
  • Profit factor is gross gains divided by gross losses and is unstable at low n.
  • Win rate does not describe the fat left tail; inspect average loss and drawdown.
  • Equity curves assume 5% fixed-fraction sizing and modeled costs.
Data & freshness
  • The current monitor is cached for 5 minutes.
  • The static backtest is served through a 1-hour cache.
  • The paper recorder checks and resolves approximately every 10 minutes; its API view is cached for 2 minutes.
  • This page loads once and does not poll automatically.
Trigger & engine mechanics
    Full caveats (backtest + walk-forward)
      Provenance