# Learning-loop review — 2026-07-22

11 closed trades reviewed, 13 currently open. Proposal only — nothing here is applied automatically (CLAUDE.md §4: the AI review proposes, dan disposes).

## 1. Is overall expectancy holding?

Overall expectancy is **-0.26%** per trade across 11 closed trades, with a profit factor of 0.61 (below the 1.0 break-even line) and a max drawdown of $109 against total P&L of -$75. Nominally negative, but with only **11 trades** total — well under the ~20-trade minimum — this is noise territory, not a result. The Wilson 95% lower bound on the raw 45.5% win rate is 21.3%, meaning the data can't even rule out a win rate as low as one-in-five. **No conclusion about edge should be drawn yet.**

## 2. Which bucket is the biggest drag?

**No bucket in this report reaches the ~20-trade minimum.** The largest bucket anywhere (overall, and by-strategy since all 11 trades are orb-v1) is 11 trades; every RSI/distance/rel-vol/regime cut collapses to a single "unknown" bucket of 11 (snapshot fields not populated); exit-reason and symbol cuts split it further into buckets of 1–5. Per the anti-self-deception rule, none of these can be treated as an actionable result — they're hypotheses at best (rule 5). Stopping the drag analysis here as instructed.

## 3. Entry vs. exit vs. regime problem

Skipped — no bucket clears the 20-trade bar (see #2), so attributing the drag to a specific MAE/MFE/regime pattern would be reading tea leaves out of a single-digit-to-low-teens sample. (For color only, not for action: the stop_loss exits (n=5) show avg MAE losers -1.06% with no big MFE-losers giveback signal here, and MFE losers overall is +0.41% — too thin a sample to say whether that's an exit or entry issue.)

## 4. Proposed parameter change

**NO CHANGE.** With only 11 closed trades total, all concentrated in one strategy (orb-v1), and every attribution cut fragmenting to <20 (mostly <5) trades, any parameter tweak right now would be tuning to noise and would break the "one change, measured on 20+ subsequent trades" discipline in METRICS.md. 

What to wait for: accumulate closed orb-v1 trades until the overall/strategy bucket reaches ~20, and ideally until the RSI-band / distance-from-MA50 / rel-volume snapshot fields are actually populated (they're all showing "unknown" — worth checking separately whether that's a data-capture gap in the snapshot pipeline for orb-v1, since without those fields the RSI/MA50/rel-vol cuts can never become actionable no matter how many trades accumulate).

## 5. Notable about currently open positions

- **qsr has 11 open legs across 6 symbols**, several stacked as multiple tranches on the same symbol (HDB ×2, NFLX ×2, AEM ×2, NXPI ×2, TXN ×2) — normal for its tranche design, but worth watching as a concentration point.
- **HDB and NFLX are both underwater on both tranches** (-1.12% and -1.27% respectively) with **no calendar time-stop** (`maxHoldDays: null` for qsr) — these could in principle sit open indefinitely if price drifts sideways between stop and target; worth keeping an eye on hold duration once they age past a day or two.
- **TDG and TXN legs are also underwater** (-1.34%, -1.52% ×2) but all positions are only ~0.3 days old, so this is very early and not yet meaningful — just situational color, not evidence for any parameter change per the instructions.
- The two crypto-trend-v1 positions (BTC/USD, ETH/USD) have no unrealized P&L reported and aren't part of the four strategies under review here.