# Learning-loop review — 2026-08-08

122 closed trades reviewed, 119 currently open. Proposal only — nothing here is applied automatically (CLAUDE.md §4: the AI review proposes, dan disposes).

## 1. Is overall expectancy holding?

Pooled expectancy is **+1.55% per trade** (n=122), profit factor 1.88, win rate 68.9% (Wilson lower bound 60.2%). At the top level this clears the ~20-trade minimum and the Wilson bound is comfortably above zero, so yes — there's a real, positive edge in the pooled book right now.

But this pooled number should be read cautiously: it blends four strategies with very different risk configs and three accounts (paper-main n=113, Paper02qsr n=3, live-1 n=6). The two non-paper-main accounts are far below the 20-trade minimum individually, so essentially all of this evidence is coming from paper-main, with optimistic simulated fills (no slippage/liquidity limits). Treat the +1.55% as "the paper backtest-adjacent edge looks real," not "the live edge is confirmed."

## 2. Which bucket is the biggest drag?

Restricting to buckets with **≥20 trades** (per the rule):
- By strategy: orb-v1 (n=39), qsr (n=68) qualify; vwap-mr-v1 (n=15) does not.
- Several condition buckets also qualify (SPY>MA200 n=54, relvol<0.8 n=49, RSI 40-50 n=20/50-70 n=21, MA50 -5..0% n=21, exit-reason stop_loss n=32/take_profit n=53), but all of those are net positive or definitionally-outcome buckets (stop_loss/take_profit are labeled by their own result, not a diagnosable "condition").

The clear drag is **orb-v1**: n=39, win rate 41.0% (Wilson low 27.1%), expectancy **-0.08%**, profit factor **0.85**, total P&L **-$102** — the only negative-expectancy strategy bucket with a large-enough sample to act on.

## 3. Entry, exit, or regime problem?

This report doesn't break MAE/MFE out by strategy, so I can't point to an orb-v1-specific MAE/MFE number directly — that's a real gap, flagged rather than papered over.

What we *can* infer from the numbers we have: orb-v1's **payoff ratio is 1.29** (wins are bigger than losses on average) — comparable to, even better than, several profitable buckets. The problem isn't that wins are being given back (which is what a high `avgMfeLosersPct` would indicate); it's that **win rate (41%, Wilson-low 27%) is too low to clear the bar that payoff requires** (breakeven win rate at payoff 1.29 is ~44%). That pattern — decent payoff, insufficient win rate — points to an **entry/signal-quality problem**, not an exit problem.

On regime: I can't test this at all. The market-regime cut only has an "SPY>MA200" bucket (n=54) and "unknown" (n=68) — there is no "SPY<MA200" bucket in the data, meaning we have zero observed bear-regime trades to compare against. This is a data gap, not evidence either way; it should not be used to justify a regime gate right now.

## 4. Proposed parameter change

**orb-v1 · `volumeConfirmMult`: 2.5 → 3.0**

Reasoning: since the payoff ratio is already acceptable (1.29) and the deficiency is specifically win rate (41%, Wilson-low 27%), the lever to pull is entry selectivity, not exit/target management. `volumeConfirmMult` directly gates how strong the volume confirmation must be before taking an opening-range breakout — raising it should filter out the weaker/false breakouts that are likely producing the sub-40% win rate, at the cost of fewer signals. This is a single, small, reversible change to one parameter on a strategy that's currently backtest-only (not live-wired), so it can be tested safely before ever touching capital. Per the anti-self-deception rules, this should be judged only on trades taken *after* the change (not a re-score of the current 39), and needs its own ~20-trade sample before drawing conclusions.

## 5. Notable about open positions

The open book is heavily concentrated in **qsr** (roughly 115 of ~120 open positions vs. only 68 closed qsr trades on record) — the live/paper book is running far ahead of the closed-trade evidence base for this strategy. Since `qsr.maxHoldDays` is `null`, several tranches have been open 8–17 days sitting flat-to-slightly-negative without resolving (e.g., multiple HDB tranches at -1.16% for 8.9–16.9 days, TXN at -0.64% for 16.9 days, ETR/TRGP tranches at -0.6% to -3.0% for several days) — worth watching for stagnation, though this is purely descriptive, not evidence for the parameter change above. Also notable: the two crypto-trend-v1 positions (BTC, ETH) have been open 17.8 days with no calendar time-stop and no unrealized P&L figure reported at all — a monitoring/data-visibility gap worth checking, independent of the strategies reviewed here.