hjnormey/sentiment-engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
318 Commits 1 Branch 0 Tags
97a770da6561641dae090723c864f2e8adab28a3
Commit Graph

8 Commits

Author SHA1 Message Date
Codex
97a770da65 malkhut: online EWMA self-calibrating slippage model 
Flight7 model underestimates by 80% in CWM dynamic book:
  raw predicted: 0.034 bps, actual: 0.180 bps
  Constant error across 22K episodes — no feedback loop.

Root cause: Flight7 calibrated on real BingX taker fills, but CWM's
synthetic dynamic book has different fill characteristics.

Fix: SlippageSelfCalibrator with EWMA feedback loop.
  After each fill: error = actual - predicted (clipped to +/-20 bps)
  EWMA smooths per-symbol errors (alpha=0.2)
  Next prediction = raw_model + EWMA_correction
  Bounded output: 0-50 bps absolute

Convergence (300 eps across 8 assets):
  ETH: 9% error (from 80%)
  SOL: 3.5%
  DOGE: 6.7%
  LINK: 5.7%
  ADA: 9.7%
  BTC: 48.6% (low fill count, converging)
  AVAX: 28.6% (low fill count)
  UNI: 52.5% (low fill count, early outlier)

Truthfulness guarantees:
  - Correction is observable (CALIBRATOR.correction(symbol))
  - Resets between runs (no hidden state)
  - Only uses observed fills, no assumptions
  - Error clipping prevents outlier domination
  - Absolute bounds prevent runaway
 2026-07-20 15:17:16 +02:00
Codex
4926ef6788 malkhut: CHASE mechanics FIXED + Flight9 learnings + TTL enforcement 
1. CHASE mechanics (NOW WORKING):
   - CWM enforces TTL on open orders (auto-cancel when expired)
   - DSL CHASE produces PLACE with metadata={chase: True}
   - Action menu generates chase actions with wait_to_retry_ms TTL
   - OpenOrderState.gains ttl_ms field (0=no expiry, >0=auto-cancel)

2. TTL enforcement (CWM):
   - HftBacktestCWM: auto-cancels orders where age >= ttl_ms
   - MinimalCryptoLOBCWM: same TTL enforcement
   - This is how CHASE works: place→wait→auto-cancel→next step re-places

3. Flight9 learnings:
   - Slippage model gains trade_flow_intensity parameter
   - Book imbalance as proxy for trade arrival rate
   - Markout = quality concept documented

4. CHASE tests: 10 new tests covering TTL enforcement, cancel-retry cycle,
   max retries, DSL CHASE action, CMA codec integration

5. All 800+ tests pass
 2026-07-17 19:30:17 +02:00
Codex
bb229833d3 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
 2026-07-17 16:18:43 +02:00
Codex
8857daedfa malkhut: urgency-driven maker/taker + calibrated slippage + chase + docs 2026-07-17 10:16:55 +02:00
Codex
5c4ccdb1de malkhut: 3.5H instrumented E2E + calibrated slippage + conditional slippage 2026-07-15 19:32:46 +02:00
Codex
c03d914e7a malkhut: conditional slippage (Fable) 2026-07-15 17:05:16 +02:00
Codex
618ad723e3 malkhut(wire): fill quality as PRIMARY optimization target 
Fill quality is MALKHUT's core aim. Wired end-to-end:

1. FillQuality state (state.py):
   - slippage_bps, price_improvement_bps, levels_consumed
   - is_maker_fill, rolling_fill_rate, post_fill_adverse_bps
   - fill_value_score: composite metric for optimization
   - Added to MarketWorldState.fill_quality field

2. HftBacktestCWM.transition() (hft_cwm.py):
   - _compute_fill_quality() computes all metrics per transition
   - Fill quality now tracked for every CWM step
   - Empty book guards added for safety

3. MinimalCryptoLOBCWM.transition() (core.py):
   - Same fill quality computation for deterministic fallback
   - Empty book guards added

4. Reward function (hft_cwm.py):
   - fill_quality_reward = w_fill_probability * fill_value_score (PRIMARY)
   - Bonus for maker fills that improve price
   - Penalty for adverse selection after fill
   - Base reward (PnL, adverse selection, fees) preserved

5. PerformanceMatrix (selector.py):
   - RegimeStrategyScore: 4 new fill quality fields
   - record(): accepts fill_rate, slippage, price_improvement, fill_value_score
   - EMA updates for all fill quality metrics

6. EpisodeResult (cma_trainer.py):
   - avg_fill_value_score, avg_price_improvement_bps, avg_post_fill_adverse_bps
   - Accumulated per-step during _run_episode
   - Recorded to PerformanceMatrix in evaluate_candidate

All 1379+ tests green.
 2026-07-15 15:22:25 +02:00
Codex
8b385cb249 malkhut(cwm): HftBacktestCWM — queue model + 59-test suite 
HftBacktestCWM (cwm/hft_cwm.py):
- PowerProbQueueModel: probabilistic fill per level (pre-computed)
- Level 0 always fills, deeper levels have decreasing probability
- Deterministic fallback when use_queue_model=False
- Same transition/reward/terminal API as MinimalCryptoLOBCWM
- Fallback to deterministic level consumption when hftbacktest unavailable

59 tests (test_hft_cwm.py) covering 15 test classes:
1. Queue model correctness (fill probs, monotonic, bounds, determinism)
2. Determinism & reproducibility
3. CWM interface compatibility (cross, place, cancel, post_only, reduce)
4. Reward function (profit, noop, maker bonus)
5. Edge cases (empty book, zero qty, extreme price, many levels)
6. Position tracking (buy, sell, flip)
7. Fee application (taker fee reduces equity)
8. Counterparty ecology (toxic taker hits book, noop preserves)
9. CWM comparison (Hft vs Minimal agree on noop)
10. Venue propagation (scenario tagging, cross-exchange transfer)
11. PerformanceMatrix venue keying (record, per-venue best, comparison)
12. Risk gate integration (approve, leverage, OOD, kill switch, self-trade)
13. Stress tests (rapid transitions, 20 open orders, cancel all)
14. Full episode integration (single episode runs, policy evaluator)
15. hftbacktest availability check
 2026-07-14 19:46:34 +02:00