KALSHIBOT HANDOFF — v1.5 — 2026-04-06
=======================================
Start by reading C:\kalshibot\new_chat.txt for full session history.

CURRENT STATUS:
- paper_trade.py is v1.5, syntax verified, READY TO RUN
- To start: open CMD (not PowerShell), then:
    cd C:\kalshibot
    python paper_trade.py
- Backup at: C:\kalshibot\backups\paper_trade_pre_v1.5.py

V1.5 — 3 DATA-PROVEN CHANGES (from 115-trade overnight analysis):
  1. MIN_ENTRY_PRICE: 0.35 → 0.60  (below 0.60 = 35-46% WR; above = 65-70% WR)
  2. MIN_MOM_SCORE = 8 (new const)  (MOM<8 = 37-46% WR on BOTH bots)
  3. EXP blocked if lag_score==0     (EXP+MOM no-LAG = 20-46% WR, consistent loser)

VALID SIGNALS IN V1.5:
  MOM-only       66% WR  — bread and butter
  LAG+MOM        57-75% WR
  EXP+LAG+MOM    66-100% WR
  Everything else: BLOCKED

PROJECTED: ~65% WR, ~70 trades/day

AUTOMATED BOT (C:\kalshibot\automated\) — READ ONLY, NEVER EDIT:
  Cowork is running a self-iterating version. 10 iterations done.
  Key findings from its 180-trade dataset:
  - EXP-alone = 14.3% WR (confirms our block)
  - EXP+MOM = 78.9% WR (NOTE: different params — full window, no stop loss)
  - Late-phase EXP sniper (final 2 min) = 14.3% WR — disaster
  - Their R/R problem: avg loss 2.4x avg win, needs 70.4% WR just to break even
  - Their stop loss is effectively 0 — we use 25%, much safer
  - Change logs: automated\tracker\CHANGES_LOG.txt and EXPERIMENT_LOG.txt
  - OPEN QUESTION: Should we re-enable EXP+MOM at lower thresholds?
    → Wait for 50+ v1.5 trades before deciding

WHAT TO DO NEXT SESSION:
  1. Start v1.5 and let it gather 50+ trades
  2. Analyze trade_analytics.json — verify actual WR hits ~65%
  3. Check if automated bot has new iteration data worth reviewing
  4. Once 60%+ WR confirmed over 50+ trades → consider raising bet sizes

IRONCLAD RULES:
  - CMD not PowerShell for Python
  - Syntax check before every restart
  - Backup before every major change
  - NEVER touch C:\kalshibot\automated\
  - NEVER make unrequested changes — state findings and wait for instruction
  - No hour blocking (needs weeks of data)
  - Ask "does this improve win rate?" before building anything new