# Learning-loop review — 2026-08-12

178 closed trades reviewed, 145 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 **+1.62% per trade** across 178 closed trades, profit factor 1.94, Wilson-lower win rate 63.1%. That n clears the ~20-trade bar, so at the pooled level the answer is "yes, expectancy looks positive and not degenerate."

But two caveats matter before trusting that number:
- **159 of the 178 trades are from `paper-main`** (simulated fills, optimistic). The only live account (`live-1`) has just **10 closed trades** — nowhere near the 20-trade minimum — so we have essentially no confirmed live evidence that this expectancy survives real fills/slippage.
- **swing-dip-v1 has zero closed trades** despite being a live-wired strategy — it's contributing nothing to this evidence base yet.

So: pooled expectancy is holding, but it's really "paper expectancy is holding"; live confirmation is still pending.

## 2. Which bucket is the biggest drag?

Restricting to buckets with ≥20 trades:
- Strategy: **orb-v1, n=48, expectancy −0.04%, profit factor 0.88, win rate 43.8% (Wilson low 30.7%)** — the only strategy-level bucket that's flat-to-losing.
- Exit reason: `stop_loss`, n=46, expectancy −3.92% — but this is tautological (stop_loss exits are losses by definition), so it's a symptom bucket, not an independent "drag" to diagnose the way orb-v1 is.
- Everything else with n≥20 (qsr overall, RSI 30-70 bands, MA50 -5..0%, relvol<0.8, SPY>MA200) is expectancy-positive.

**Biggest drag: orb-v1.**

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

This report does **not** break MAE/MFE out by strategy — only pooled (MAE winners −1.82%, MAE losers −3.68%, MFE losers +0.81%), and that pool is dominated by qsr's 114 trades, so it can't be used to diagnose orb-v1 specifically. I'm not going to force-fit a diagnosis onto numbers that don't isolate this bucket.

Using what *is* available for orb-v1: win rate 43.8% (Wilson low 30.7%) and payoff ratio only 1.21, with an average hold of just 0.3 days. A payoff ratio near 1 with a below-50% win rate — on trades that resolve same-day — points more toward **a signal/entry-quality problem** (too many marginal breakouts triggering) than an exit give-back problem: there simply isn't much time in a 0.3-day hold for a winner to run up and get given back before exit. It doesn't look like a regime issue either — we have no regime breakdown for orb-v1 alone. **Read this as a tentative entry-quality read, not a confirmed one**, since the causal MAE/MFE evidence for this specific bucket is missing from the report.

## 4. Proposed parameter change

**orb-v1.volumeConfirmMult: 2.5 → 3.0**

Reasoning: orb-v1's edge is currently indistinguishable from noise (expectancy −0.04%, PF 0.88, Wilson-low win rate only 30.7%). Since the likely fault is entry signal quality rather than exit placement (see #3), the smallest lever that raises the bar on *which* breakouts get taken — without touching sizing, targets, or stops — is demanding a stronger volume confirmation before entry, per the doc's own guidance ("persistent pattern = enter later / demand more confirmation"). This is a single, small, reversible change. Per rule 4, only trades placed *after* this change should be used to judge it, and it needs its own ~20-trade sample before acting further.

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

- **QSR dominates open exposure** (nearly all open positions) and **has no calendar time-stop** (`maxHoldDays: null`) — several positions are aging 13–21+ days with no resolution mechanism (e.g., HDB open 21.3 days at −3.34% unrealized, TXN 21.3 days at −3.97%). This is consistent with HDB's closed-trade record being genuinely bad (`qsr · HDB`: n=2, 0% win rate, expectancy −4.60%) — worth watching, not acting on (n too small either way).
- A cluster of very recent entries (BTI, EBAY, NKE, ITUB, AZN) are already underwater within 1–2 days of entry (−1.3% to −4.0% unrealized) — too early to mean anything, but worth tracking given the no-time-stop design.
- Two `external`-strategy positions (MARA, ETHUSD) sit outside the four strategies under review and aren't covered by this journal's attribution.

## 6. QSR shadow comparison note (observational only)

The legacy `isTriggered+isBuy/buyZonePct` shadow method and the live `newlyA` method are picking **almost entirely different tickers** — only 3 of the combined ~17 distinct symbols overlap (AEP, SU, TRP). The live method is also firing **far more often** (72 real entries vs. 21 legacy would-buys) over the comparable window. That's a large divergence in both selection and frequency worth flagging qualitatively — it suggests the two signal definitions are picking up substantially different opportunity sets, not just timing the same names slightly differently. The shadow window is still running (ends 2026-08-23), so no outcome comparison can be requested yet, and per the source material this cannot feed into point 4's proposal regardless.