#Vault Breakers: Super Score — Seeded Simulation Evidence, draft-v0.1

SUPERSEDED by draft-v0.2. Retained unchanged. This document records the APPROVED unbalanced reference baseline of 2026-08-19, including the two Bonus Buy prices that deliberately did not balance. Nothing in it is rewritten by later tuning.

DEV ONLY · NOT FOR PRODUCTION · NOT FOR CERTIFICATION. Every figure in this document is a MONTE CARLO ESTIMATE. The exact results — one snapshot's regular RTP, its hit rate, the scatter-class trigger probability and the Super Scatter instant-prize expectation — live in ../exact/draft-v0.1.md and are never re-derived by sampling.

#Commands

cd /Users/admin/kiro/backend
go test -tags vbdraft -count=1 -v -timeout 180m \
    ./libs/mathengine/games/vault_breakers_super_score -run Draft

#Run A — whole-game confirmation

#Shard boundaries and seeds

#Results — all estimated

Target 96.00% lies inside the interval, 0.76 standard errors from the estimate.

#RTP decomposition

The free-spins term measured by RunBatch is one number, 8.9272%, because the atomic executor settles a sequence as a lump. It is split here by the regular/charge ratio measured directly in Run C on runtime-shaped sequences (55.34% / 44.66%), which is the same rule the atomic path applies. The split is derived, the total is measured, and neither is counted twice.

#Win bands

#Exceedance

The band edges fall exactly on these thresholds, so these are read off the histogram rather than interpolated.

#Run B — distribution, percentiles and charge behaviour

#Percentiles — all estimated

P50 and P75 are zero because 81.3% of rounds pay nothing. That is the shape of a high-volatility count-anywhere game, not an artifact.

#Base-game charges

Free-spin charges are settled inside the feature and are not itemised here.

#Charge value distribution — 1,124,401 drawn

Every observed frequency agrees with the configured weight — a direct check that the pool the engine draws from is the pool the config declares.

#Feature frequency

Run A reports free spins contributing on 0.005652 of rounds — slightly below the trigger rate above, because a triggered sequence can pay nothing.

#Run C — feature-conditioned: the free-spins split, retriggers and both buys

Sequences are played the way the runtime plays them: one spin per request against the chosen variant, retriggers resolved from the engine's own verdict via mathengine.ResolveOutcomeTrigger, and the whole-cycle ceiling ending generation rather than clamping at the end.

#Results — all estimated

The regular parts agree to within 0.06% — as they must, since the two modes differ only in their charge pool. That agreement is itself a check that the variant pool override touches nothing else.

#Buy prices

The design's reference prices do not survive contact with this game's free spins, and the design itself requires recomputation against the tuned model.

The reason is structural, not a tuning error. In this design free spins differ from base spins in exactly one way — the Super mode's charge pool — so Standard free spins are 15 ordinary spins plus retriggers, worth 15.52×. A 15-spin feature cannot be worth 100× a stake in a 96% game. The reference game's free spins are its main event because its free-spin reels are charge-dense; ours are the base reels, because FreeSpinsExecutor ignores a free-spins-specific reel set and using one would make simulation and runtime disagree.

#Reconciliation and tolerance

Per-round σ is 16.112286, dominated by the charge mechanic rather than by the Super Scatter prize. That matters because it sets what any sample size can resolve:

The ±0.10 pp band proposed on 2026-08-19, before the variance was measured, needs roughly 1.04 billion rounds. The recorded band for Draft v0.1 is therefore the interval this run actually supports:

whole-game RTP = 95.8775%  ±0.3158 pp   (95% CI, 100,000,000 rounds)
target         = 96.0000%
deviation      = -0.1225 pp  =  0.76 standard errors
verdict        = target inside the interval

The original proposal assumed the 50,000× instant prize was the dominant variance term and that computing it exactly would buy the precision. Measurement says otherwise. Reducing the interval further needs either an order of magnitude more sampling, or decoupling charge values from the round's RNG stream so the sum can be Rao-Blackwellised — a change with its own consequences for recorded vectors.

#Unresolved

1. The free-spins regular/charge split is derived from Run C's ratio, not measured inside Run A. The atomic executor reports a sequence as one number.
2. Charges collected inside free spins are not itemised anywhere; only base-game charges are.
3. RunBatch's PerFeature.HitCount counts rounds where a feature contributed win, not rounds where it triggered. The two differ for free spins and both are reported above.