---
name: t60 variant — one book in band at T-60min (passive, n>=100 review)
description: User-proposed timing variant: fire when exactly one exchange is in the [5,12) band at the snapshot closest to 60min pre-KO. Backtest n=70 +45.3% z+2.02 (TRAIN/TEST consistent), same-subset xlow control +15.1% — timing, not subset. Logged passively in the forward tracker; primary unchanged.
type: project
---

Proposed by the user 2026-07-19 (during a 0-for-4 morning, deliberately NOT wired as
primary — variant was designed after peeking at forward results, so its z is
unadjusted for selection).

**Rule:** at the pre-KO snapshot closest to T-60min (±45min tolerance), fire when
exactly ONE of Kalshi/Poly has pin-CLV in [5,12) on a side (bet that book's side,
max CLV if both sides qualify). Point-in-time and fully executable — this is the
tradeable cousin of "unique" (whose whole-window definition needs future knowledge)
and of the persistence idea parked in fire_xlow's docstring.

**Backtest (poly_backfill, 975 fixtures):**
- t60: n=70, WR 32.9% vs 23.0% implied (+9.9pp), ROI +45.3%, z=+2.02, TRAIN +49.3% / TEST +39.2%
- Control — strict xlow on the same 676 fixtures that have a T-60 snapshot: n=206, +15.1%, z=+1.50 (+4.4pp WR edge)
- So the concentration comes from the TIMING (cheap vs a sharp late Pinnacle after
  surviving all day un-arbed), not from the fixture subset.

**Status:** logged passively as the `t60` column in forward_test_unique_book.py /
its log CSV (same pattern as xlow_f10). Decision belongs to the n=1000 review of
the PRIMARY rule — if t60's forward sample confirms the concentration, a switch
can be considered THEN with a fresh freeze. Do NOT promote it mid-test; do NOT
bet it live alongside the primary (double-counting the same fixtures).

**How to apply:** when the user asks about t60 or "the 1-hour rule", the numbers
above are the reference; forward t60 stats come from the tracker log. Related:
[[project_xlow_flat_vs_kelly]], [[project_kalshi_only_execution]].