#draft-v0.2 full-grid confirmation (feature-conditioned)

Confirmation evidence for the accepted LandingChancePPM = 43000. Not a tuning pass — no parameter was varied.

#Why this run exists

The 2,000,000-round confirmation observed only 10 full grids. A rate estimated from 10 events has a Poisson interval roughly twice as wide as the acceptance band it is being checked against, so it could place the estimate near the band but not inside it with any confidence. Resolving that from base rounds would need hundreds of millions of them, because 99.36% of rounds never enter the feature at all.

Conditioning on the trigger fixes it. Every sample here is a triggered feature played through the canonical RunHoldAndWinToCompletion — the same implementation production steps through one request at a time — with trigger counts weighted by the exact enumerated trigger-count distribution. The estimate stays unbiased for the real game while every sample is spent where the event lives.

#Manifest

Reproduce:

cd /Users/admin/kiro/backend
go test -tags hwdesign ./libs/mathengine/games/hold_and_win_reference \
    -run TestFullGridConfirmation -v -timeout 3600s

#Result

Total RTP and feature RTP use the exact base RTP (75.1160%) and the exact trigger probability (1 in 155.4035), both re-enumerated in this run and unchanged.

#Relationship to the round-level confirmation

The two runs are independent and consistent.

The round-level figure rests on 10 events, so its own Poisson interval is roughly [0.030%, 0.127%] — which comfortably contains 0.090053%. The conditioned estimate is about 12× more precise and is the figure to quote; the round-level figure is retained because it is what the whole-round run actually observed, and dropping it would hide the basis of the earlier claim.

Both are inside the accepted band, and the conditioned figure is closer to the 0.10% target than the round-level one. This does not materially contradict the accepted band, so 43,000 PPM stands unchanged.

#Correction recorded

An earlier version of the draft-v0.2 results table reported the unconditional full-grid frequency as "1 in ~127,700 base rounds". That was wrong: it multiplied the 1-in-1,277 conditional figure by 100 instead of by the 155.4035-round trigger interval. The three figures are now always reported together, so a conditional rate cannot be mistaken for an unconditional one:

conditional        0.0783%  (round-level)   /  0.090053%  (conditioned)
per trigger        1 in 1,277               /  1 in 1,110
per base round     1 in ~198,500            /  1 in 172,570

#Limitations

• Conditional figures are Monte Carlo, reported with standard error and interval. Base RTP and trigger probability are exact.
• The conditioned run measures the feature through RunToCompletion. Reference-side equality with step-by-step execution is proven separately on the golden corpus; live-runtime parity is still outstanding.
• DRAFT / NOT FOR CERTIFICATION.

Resolved 2026-08-09 (backend cd4c093). The limitation recorded above was accurate when this run was produced. Live-runtime routing and reference↔runtime parity now exist: the spin orchestrator opens the feature on a paid trigger, advances it one respin per request, and its complete state is compared with RunHoldAndWinToCompletion at three landing chances. The original wording is left in place because it is what this report observed; this note is the correction, not a rewrite.