---
name: Cash-out rules tested — hold to settlement, all leagues, no exceptions
description: Every in-game exit rule (HT always, HT-if-trailing, 75', sell-when-up-2) tests ≤ hold on 453 live_data bets, aggregate and per-league. USL's "+4.9% HT-trailing" and CanPL's "+92% up-2" cells are small-n mining traps. Edge lives at entry; exits only re-pay the in-game spread. Closed 2026-07-08.
type: project
---

User asked 2026-07-08 whether cashing out is profitable by league (USL specifically), then sell-when-up-by-2. Simulated on ALL 453 baseline Kalshi paper bets reconstructible from live_data (entry = first in-band [5,12) pre-KO snapshot @ask; exits sell full position at the logged in-game kalshi bid; hold = settle on ftr). Consistent with the earlier archived study (misc/_cashout_analysis.json, misc/archive/cashout_sim.py) — now confirmed at larger n.

**Aggregate (n=453, HOLD baseline +8.1% ROI):**
- Cash at HT always: **-5.8%** (destroys value — sells the rare winners cheap)
- Cash at HT only if trailing: +9.4% (wash vs +8.1%; insurance costs ≈ what it pays)
- Cash at 75': -5.0%
- Sell when up by 2: **+6.0%** (-2pp, and NEGATIVE even on the 63 triggered bets: +260% sold vs +276% held)

**Key mechanism numbers:** when our side goes up 2, avg in-game bid = 0.812 but 84% of those positions settle as WINS — Kalshi's in-game bids run ~3pp under realized value (thin books, widest spreads). Every exit rule pays that haircut on the biggest positions; the occasional rescued collapse (10/63 blown 2-goal leads) doesn't cover it.

**Per-league is a mining trap:**
- USL HT-if-trailing "+4.9% vs -2.7% hold" = one cell in a 22-league x 4-rule grid; USL is also the THINNEST book (simulated bid fills least trustworthy) and the up-2 rule is negative for USL (-2.4%).
- Sell-up-2 is negative in 19/22 leagues; the 3 positive cells (CAN +92%, B1 +25%, D1 +30%) are 1-2-trigger blown-lead lotteries (CAN: 0 won-through, 2 collapses). Same non-persistence as the league-filter study (corr(train,test) ≈ 0 across leagues).

**Why structurally:** the strategy's entire edge is at ENTRY (pre-KO Kalshi/Poly mispricing vs Pinnacle). In-game we have no informational edge, books are at their widest, and any exit re-pays a spread on the thin side at the worst time, concentrated on the winners that carry the whole longshot P&L. Same family as sub-ask limits (adverse selection) — execution tricks around a thin entry edge all bleed.

**How to apply:** Hold to settlement — all leagues, no exceptions, don't wire any cash-out rule. Don't re-open per-league cash-out cells (they're re-rolls of "did a lead collapse in this league's handful of triggers"). Two legitimate future revisits ONLY: (1) if aggregate HT-if-trailing stays ≥ hold at n in the thousands, it becomes interesting as pure variance reduction (same EV, smaller swings) — aggregate-level decision, never league-level; (2) if Kalshi in-game liquidity fundamentally changes (tight spreads, real depth), re-measure the ~3pp bid-under-realized gap before reconsidering.