# Learning-loop review — 2026-08-07

113 closed trades reviewed, 94 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 **+1.33% per trade** across 113 closed trades, with profit factor 1.82 and win rate 66.4% (Wilson 95% lower bound 57.3%). The overall sample has cleared the ~20-trade minimum, so this headline number is meaningful — but "overall" is a blend of very different strategies (qsr strongly positive, orb-v1 negative), so it should not be read as "the system is fine everywhere."

## 2. Which bucket is the biggest drag?

Several buckets clear the ≥20-trade bar: `orb-v1` (39), `qsr` (59), `rsi 50-70` (21), `relvol<0.8` (41), `SPY>MA200` (45), exit reasons `stop_loss` (32) and `take_profit` (44).

Among these, **`orb-v1`** (n=39) is the standout drag: expectancy **-0.08%**, profit factor **0.85** (below 1.0 — genuinely losing money), win rate 41.0% with Wilson lower bound only **27.1%**. It's the only strategy-level bucket with both adequate sample size and negative expectancy.

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

The report doesn't give a strategy-specific MAE/MFE breakdown, only aggregate figures (MAE winners −1.84%, MAE losers −3.53%, MFE losers 0.98%), so a fully clean MAE/MFE attribution to orb-v1 specifically isn't possible from this evidence — flagging that limitation explicitly rather than overreaching.

That said, the shape of orb-v1's own scoreboard points toward an **entry problem, not an exit problem**:
- Payoff ratio is 1.29 (wins average bigger than losses) — exits aren't giving back profitable trades, which is what an MFE-driven exit problem would look like.
- Aggregate MFE-losers (0.98%) isn't "materially positive" in the sense the doc warns about (a big giveback like "up 3%, exited -2%") — there's no strong signal that winning trades are turning into losers late.
- The failure mode instead is a **low win rate with a tight, short hold** (41% win rate, Wilson lower 27%, avg hold just 0.4 days) — i.e., a lot of breakouts are being entered that never confirm, consistent with a signal/confirmation (entry) issue rather than mismanaged exits.

This is a hypothesis given the missing per-strategy MAE data, not a certainty — but it points at entry quality (the volume-confirmation breakout signal), not exits or regime (no regime split is available by strategy either).

## 4. Proposed parameter change

**Strategy:** `orb-v1`
**Parameter:** `volumeConfirmMult`
**Current value:** `2.5`
**Proposed value:** `3.0`

Reasoning: orb-v1's low win rate (41%, Wilson lower 27%) combined with a still-healthy payoff ratio (1.29) suggests the strategy is entering too many marginal/unconfirmed breakouts rather than mismanaging winners. Raising the volume-confirmation threshold should filter out weaker breakout attempts at entry, directly targeting the win-rate side of the equation without touching exits/targets (`targetRMultiple`) that currently appear to be working fine. This is a single, small, reversible change — measure the next ~20+ orb-v1 trades post-change before judging.

(No change proposed for any other strategy/bucket — qsr, relvol<0.8, SPY>MA200, rsi 50-70, and the exit-reason buckets all look healthy or are tautological outcome-splits, so touching them now would violate the one-change-at-a-time rule.)

## 5. Notable open-position color (not evidence for #4)

- **qsr dominates open exposure** with ~90 open lines, heavily pyramided into single names via repeated tranches (e.g., ARGX ×6, WMB ×6, SNY ×6, DDOG ×6, AZN/MO/IBKR/VST ×4 each) — expected given `maxBuysPerTickerPerWeek`/`maxTranches`, but it does mean concentration risk is currently clustered in a handful of tickers.
- **VST** is currently open (multiple tranches, ~-0.69% unrealized) and is the same symbol whose *closed* trade history was the worst in the book (2 closed trades, 0% win rate, -9.65% expectancy, -$128 total). Worth watching given the historically poor track record on this name, though this open position itself has no confirmed outcome yet.
- **Long-aging, flat/slightly-negative positions**: several `qsr` positions (HDB, TXN) have been open 16.3 days sitting near breakeven-to-slightly-negative, with `maxHoldDays: null` for qsr meaning nothing will force a resolution. Not a decision input, just a flag that capital is tied up without a calendar backstop.
- **AEM** open 16.3 days is sitting at **+22.67%** unrealized — a large favorable outlier worth watching for how/when it eventually exits (trailing stop vs target), since qsr's trailing-stop exit bucket historically has performed very well (100% win rate, 5.65% expectancy) in closed trades.