Files
StarrivandClaude Fable 5 87f2f35e21 feat: backtesting research framework, market-data ops, and unified logging
Snapshot of the accumulated feature work across several sessions:

- backtesting: single-symbol visual backtest with canonical artifact
  (contract 1.0 -> 1.4): point-in-time replay via production scoring,
  21/63d event study with non-overlapping inference and block bootstrap,
  fixed-window and managed (position-sized, 1R breakeven, trailing-stop,
  close-signal exit) simulations, rejected stateful lifecycle experiment,
  self-contained HTML renderer, cross-sectional factor-ic/score/robustness
  studies over a frozen 148-name universe
- market data: full-market warmup, asset-catalog sync and coverage
  reconciliation with shared rate limiting; cache admin hardening
- logging: unified structured stderr logging across all entrypoints
  (JSONL default, console override), stdout reserved for results
- scoring: entry-timing layer refinements, trade-plan trailing stop echo,
  forward-log A/B fields; model card and agent contract docs updated
- make: command gateway with thin forwarding to backtest.py subcommands
- tests: 555 passing, including managed-exit semantics, lifecycle
  validation, logging guards, warmup/reconcile and rate-limit suites

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-11 22:41:38 +08:00
..