malkhut: Flight9 learnings — markout=quality, queue×flow, depth-for-size

Fable's Flight9/BLUE generalizable features incorporated:

1. Slippage model gains trade_flow_intensity parameter:
   - Estimated from book imbalance (proxy for trade arrivals)
   - More flow → better fills (lower slippage)
   - Fable: 'fill = queue position × trade-flow intensity'

2. Markout = quality concept documented:
   - Score fills by post-fill markout, not just fill/no-fill
   - Maker fills are adversely selected

3. Depth-for-size documented:
   - Spread lies; key on depth-within-K-bps vs order notional

4. Measured fees:
   - BingX maker=2.00bp, taker=5.016bp (over 1,455 fills)
   - BingX commission = NEGATIVE (debit)

5. OB study updated with Flight9 learnings

MALKHUT/docs/OB_MICROSTRUCTURE_STUDY.md

@@ -392,3 +392,55 @@ The OB microstructure directly determines fill quality:
 The PerformanceMatrix tracks `avg_fill_value_score` per (regime, strategy, venue),
 enabling the system to learn: "In this regime, on this venue, this strategy
 achieves the best fill quality."
+
+## 14. Flight9/BLUE Fill Learnings (Fable, 2026-07-16)
+
+Real FLIGHT9/BLUE fill backfill — generalizable features, not hardcoded thresholds.
+
+### Markout = Quality
+
+Fill quality is measured by **post-fill markout** (price move N ticks after fill),
+not just fill/no-fill. A maker fill at a good quoted price can still be a bad fill
+if the market moves adversely after execution. The system scores fills by markout.
+
+**Generalizable:** score fills by post-fill markout. Model fill-CONDITIONAL-on-adverse-flow.
+
+### Fill = Queue × Flow Intensity
+
+Fill probability is driven by **trade-arrival intensity** (the tape), not the static book.
+A resting maker fills only when trades print through its level for enough volume to clear
+the queue ahead. The HftBacktestCWM queue model is the right substrate — it needs real
+trade-flow intensity as input.
+
+**On testnet:** maker fill-rate ~0% (no flow). This is an artifact, not a signal.
+
+### Depth-for-Size, Not Spread
+
+Spread alone LIES: an asset with 1.5bp spread behind ~$977 of depth is unfillable at any
+size. Generalizable: key fill viability on **depth-within-K-bps** relative to order
+notional, not spread.
+
+### Measured Fees (Venue-Parameterized)
+
+| Venue | Maker | Taker | Sign |
+|-------|-------|-------|------|
+| BingX | 2.00 bp | 5.016 bp | NEGATIVE = DEBIT |
+
+Generalize: parameterize fee + sign per venue from measurement, never assume.
+
+### Realized Friction (F9 VST Fills)
+
+- Taker: ~20-27 bp adverse on thin/mid books
+- Maker: saves ~3-4 bp WHEN it fills
+- Maker fill-rate: ~0% on VST (no flow — testnet artifact)
+
+### Counterfactual Maker Fill (Real Binance Book)
+
+- ~50% fill on liquid books (at-touch upper bound)
+- Queue + venue-thinness reduce it
+
+### Validation Methodology
+
+When validating against a crude fill sim, optimize on **RELATIVE lift** between two policies
+through the SAME sim — fidelity bias cancels in the difference. Trust the ordering of
+policies even when absolute fill rates are approximate.