gavindiaz/winning-wallet-finder
1
0
Fork 0
Code Issues Pull Requests Actions 55 Packages Projects Releases Wiki Activity
3,631 Commits 1 Branch 0 Tags
main
Commit Graph

3 Commits

Author SHA1 Message Date
gavindiaz bb3f702cca docs: explain why forward P&L changes daily in AGENT.md 
Added section explaining:
- Forward 30d P&L is a rolling window, changes every day
- Why numbers shift (window moves, sync_floors recalculates p80, git reset restores different backtest.json)
- Judgment rules: 1 week negative = observe, 2 weeks = demote, 3 weeks = kick
- Don't swap wallets based on one day's data - look at multi-week trends
- This is why weekly Friday review exists, not daily
 2026-07-21 09:33:55 +08:00
gavindiaz 3141c392de docs: update 3 docs with critical findings from author's research 
AGENT.md:
- Expanded dimension 4 (niche analysis): ITF tennis/tier-3 esports = real edge,
  5min BTC/scalping = never follow, long holds (>7d) kill compounding
- Added edge disappearance warning (ArbTraderRookie wiped mid-analysis)
- Added 5 new AI agent rules: don't use copy_pnl alone, niche awareness,
  edge fragility, win rate is survivorship-biased, small capital = concentrate

日常使用.md:
- Added emergency exit procedure (host/flatten_positions.py)
- Added edge disappearance detection and response procedure

问题排查与踩坑记录.md:
- Added gotcha 7.11: win rate is survivorship-biased (90.6% shown = 48.3% real)
- Added gotcha 7.12: edge can vanish (wallet wiped from API mid-analysis)
- Added gotcha 7.13: long holds kill compounding (leegunner + lifetime
  but - to copy due to 7.6-day holds)
- Added gotcha 7.14: emergency exit tool (flatten_positions.py)
 2026-07-21 09:10:58 +08:00
gavindiaz ad0e6dd923 docs: add AGENT.md - AI agent collaboration guide for weekly wallet review 
Complete guide for human-AI collaboration on weekly Friday bench review:
- Part 1: Full wallet screening chain (3 layers: watch_sharps -> backtest -> live)
  with copy_pnl vs forward realized explanation (oliman2/leegunner lesson)
- Part 2: Weekly Friday review process (5 steps with exact commands)
  - Step 1: Human pulls 6 data sets from server (exact commands)
  - Step 2: AI analyzes 5 dimensions (forward P&L, copy_pnl, conv_win, seasonal, floor)
  - Step 3: AI outputs structured recommendation
  - Step 4: Human makes final decision
  - Step 5: Human executes on server (promote/demote/pin commands)
- Part 3: Log analysis guide (how to discover wallet niches, floor drift,
  CASH!=CHAIN, RTDS issues from logs)
- Part 4: Data checklist (8 items human must provide from cloud)
- Part 5: AI agent work rules (input format, analysis rules, output format)
- Appendix: Current status snapshot (8 live + 7 bench with forward P&L)
 2026-07-21 08:59:45 +08:00