---
name: Manual execution slippage is a known leak
description: User hesitates placing all signals manually; skipped bets cause real P&L drag — motivates the auto-trade plan
type: project
---

The live soccer-betting pipeline fires CLV signals correctly, but the user places bets manually in the Kalshi web UI and **hesitates on some of them** (long-shot prices feel uncomfortable to actually click "buy" on). This causes selection slippage — and because the strategy has a low base rate (~25–30% WR) with lumpy payouts, missing winners hurts disproportionately. Confirmed by the user on 2026-05-23.

**Why this matters:**
- Don't assume "bets in pending_bets.json / signaled by email" == "bets actually placed at Kalshi"
- When reconciling Kalshi account vs internal P&L, expect the user's actual P&L to be *worse* than the tracked theoretical P&L
- This is the explicit motivation for the automated Kalshi order plan at `C:\Users\lirse\.claude\plans\serene-soaring-kazoo.md`

**Quantified 2026-05-31 (the smoking gun):** Pulled real Kalshi settlements vs tracked history. Real account: **174 settled markets, 26 wins → 14.9% real WR.** Tracked paper: **330 bets, 85 wins → 25.8% WR.** So the real win rate is barely HALF the paper rate. On this longshot strategy (clears only via rare 3–5× wins paying for many small losses), dropping 26%→15% WR is exactly what flips +24% paper ROI into a real net loss. The user placed real money on only ~half (174/330) of signals, and the skipped half contained a disproportionate share of winners — you keep the frequent losers (feel safe to click) and miss the rare wins (look least likely). This is why real net loss (~−$100+ incl. ~$37 fees) is far worse than paper +$411, even though sizing-scale alone can't flip the sign.

**Key reframe for the user:** a positive *paper* P&L going *negative* in real life is NOT a sizing issue (scale changes magnitude, not sign) and NOT necessarily edge decay — it's selective/incomplete execution on a strategy that mathematically requires taking ALL signals to capture unpredictable wins. Auto-trading at consistent small size should pull realized WR back toward the paper 25.8% (minus fees), not by improving the edge but by removing the human self-selection into losers.

**How to apply:**
- When user mentions "I lost money today" but tracked P&L is positive, the gap is most likely skipped winners — don't go hunting for a bug in the pipeline first. Verify via real WR vs tracked WR (real settlements: revenue>0 = win).
- When prioritizing work, the auto-trade rollout (Phases 1–3) addresses this directly and the user is motivated to do it
- Don't lecture about discipline — the user knows; the answer is automation, not willpower