The machinery,
rehearsed.

The setup

Before taking a first dollar, we ran the payout machine end-to-end on accounts that do not exist: 300 synthetic clients, 900 payout requests, eight simulated weeks (2026-06-07 to 2026-08-02), all of it decided by the published payout policy applied literally — the 24-hour window, the automatic +2% on a miss, and refusals only under the published clauses.

The run is deterministic: seed 20260802 drives one seeded stream in one fixed order, so the same seed reproduces this dataset byte-identical — every aggregate on this page is re-derived from that stream at build time, not typed in. A rehearsal you cannot reproduce is an anecdote.

The 24-hour scorecard

The published rule: a payout request settles within 24 hours of approval, or the system adds 2% to the amount by itself — no ticket, no chasing. In the rehearsal we applied that rule to ourselves, and the misses cost the simulated house real arithmetic:

The 33 late rows were not tidied out of the record: each one carries its elapsed time and its +2% permanently, which is exactly what the real ledger will do to us when a real request runs late.

Per rail

Four rails, the same 24-hour clock on each. The simulated volumes are synthetic; the rail list and the rule they run under are the real ones.

Refusals, by published clause

57 of 900 requests were refused — and every refusal cites one of the published clauses, in its exact published wording, because the simulator has no way to emit any other reason. That constraint is the rehearsal: a refusal ground that is not on the published list cannot be used, in the sim or in production.

The fifth clause shows zero by design, not by luck: a rule 10 refusal comes out of the 72-hour trade-correlation review, a path this rehearsal does not model. Saying so here beats a suspiciously tidy distribution over five.

The ledger format, rehearsed 1:1

The rehearsal wrote its rows in the public ledger’s exact column shape — SETTLED · ACCOUNT · AMOUNT · RAIL · ELAPSED · OUTCOME — so the format ships tested. Below, a fixed sample: every 45th of the 900 requests, in request order, uncurated. The handles are synthetic; no sim-trader is a person.

Amounts include the automatic +2% where the 24-hour target was missed. Refused rows stay in the table with their clause — a ledger that only contained successes would be marketing.

What this proves — and what it does not

What a rehearsal on synthetic accounts proves:

• The ledger format — columns, statuses, refusal citations — ran end-to-end against 900 requests before any client saw it.
• The penalty arithmetic is wired, not aspirational: the +2% was computed and attached automatically on all 33 misses, totalling $569.14.
• Every status a real request can be in — settled, settled late, refused under a published clause — existed and rendered before the first client exists.

And what it does not prove:

• Nothing on this page is a payment. No money moved, no trader was paid, no request was real. Speed and refusal rates in a simulation carry no evidence about production.
• The real ledger is on /payouts/, reads $0.00, and stays there until the first real settlement at rulebook v1.2. That number is the honest one, and this page exists so the rehearsal can never be mistaken for it.