---
name: Sub-ask limit orders are adverse-selection traps — take the ask
description: Data shows limit orders below the Kalshi ask fill ~38% and miss ~71% of winners; the CLV edge requires taking liquidity at-ask. Reverses Phase 4 of the auto-trade plan.
type: project
---

Tested 2026-06-01 on live_data/*.csv (418 signals, first CLV in [5,12]). Simulated resting a buy limit N cents below the ask at signal time, filling if any later pre-kickoff ask <= limit.

**Result:** taking the ask (offset 0) = 100% fill, +14.9% flat ROI, +62.4 units. Any sub-ask offset is catastrophic:
- −1c: only 38% fill, ROI ~0%, and **misses 79 of 112 winners** (~71% of all winners land in the UNFILLED bucket). Net collapses to ~breakeven (−63.7 units vs market).
- −2c and lower: fill rate 14%→1%, misses 99→112 winners.

**Why (the mechanism):** This is adverse selection driven by the convergence finding (project_kalshi_convergence / the same session). On the sides we back, Kalshi underprices vs Pinnacle and then drifts UP toward Pinnacle (~+0.6c catch-up) before kickoff. So the price runs AWAY from a low resting bid → no fill → and those are exactly the winners (line moved our way). The ask only comes DOWN to our bid when the side is drifting the wrong way (Kalshi was overpricing) → we fill into losers. The ~1c price improvement on fills cannot offset losing 71% of winners.

**How to apply:**
- Do NOT implement Phase 4's "limit orders below ask to capture spread" for this strategy — it's backwards. The serene-soaring-kazoo.md plan's Phase 4 premise (Polymarket spread-tax) does NOT apply when our own signal predicts the book's direction.
- Auto-trade execution should take liquidity AT the ask (limit-at-ask or market). Entering late/cheap selects into the wrong bets.
- If revisiting execution optimization, the lever is timing (enter promptly at signal, since the edge decays as Kalshi catches up), NOT price-shading below ask.
- Caveat: fill proxy = observable ask path; real resting-bid fills could differ slightly, but the adverse-selection direction is huge (~63-unit swing) and robust.