Skip to main content
Read-only snapshot: model data as of 2026-07-16; paper desk record through 2026-10-22. No live data; paper units only.

PAPER DESK -- PAPER ONLY, simulated maker quotes on the recorded Kalshi tape; not a verdict

How in-game works

export 2026-10-22, generated 2026-10-23T19:43:59Z; 2 refusal(s)

Three dates frame this record. Capture history starts 2026-09-24, the first night in the fill-realism table. The first ledger night with paper fills is 2026-10-03. The prospective forward paper series, with schedules written before the games, runs from 2026-10-06 (evidence packet section 5). No calibration verdict yet, for or against the model.

Pipeline (one live paper quote, end to end)

as of: not stated in the receipt
  1. 1.

    State capture ->

    Two local recorders append the venue order book (Kalshi) and the game state feed to per-sport, per-UTC-day shards. Every row keeps its own receipt time, so later steps only see what was known at that instant.

  2. 2.

    Per-sport fit ->

    Once a day, offline, the scoreboard's validated history pair is fit into a frozen calibration per sport (nba, nfl, soccer, tennis). MLB has no daily fit; its in-game calibration is read from scoreboard arms B and C against the mid. Nothing is refit during a game.

  3. 3.

    Arms ->

    Each tick, for each in-window game, every arm turns the freshest book and state into a probability: the venue mid itself, the frozen B/C fits, an MLB blend of the frozen pregame mid with the live state, and an NBA logistic pricer. Stale books or state, a missing pregame mid, or a final game are counted refusals.

  4. 4.

    Live paper quote ->

    An arm that passes its checks writes a hypothetical one-contract maker quote intent per side (yes bid / no bid) with an expiry. No order is sent: there is no venue client and no network access on this path.

  5. 5.

    Fill realism ->

    After the day closes, each intent is replayed against the captured trade prints. A quote is credited only when prints during its life trade through its price, or deplete the modelled queue ahead of it; an unknown queue is never treated as depleted.

  6. 6.

    Markout

    For each credited paper fill, the later mid minus the fill price (fee netted) at fixed horizons, in probability points per contract. A markout is a microstructure diagnostic, clustered by game.

Arms by sport (live recorder)

as of: not stated in the receipt
not available: ingame: docs/evidence/forward/live_shadow/2026-10-22.json missing or unreadable

Quote counts by arm

as of: not stated in the receipt
not available: ingame: docs/evidence/forward/live_shadow/2026-10-22.json missing or unreadable

Refusal reasons

as of: not stated in the receipt
not available: ingame: docs/evidence/forward/live_shadow/2026-10-22.json missing or unreadable

Receipt lag

as of: not stated in the receipt
not available: ingame: docs/evidence/forward/live_shadow/2026-10-22.json missing or unreadable