#Magic Vault 5×3 — Evidence Manifest, draft-v0.1

DEV ONLY · NOT FOR PRODUCTION · NOT FOR CERTIFICATION. This manifest binds a set of Draft results to one config and one backend commit. It is not a freeze manifest and confers no lifecycle status.

The results below were APPROVED as Draft math on 2026-08-10. That approval accepts the measured numbers; it does not advance the lifecycle state, and this manifest is still not a freeze manifest.

#Bound identities

Recompute both hashes before trusting this manifest:

git -C /Users/admin/kiro/backend rev-parse HEAD
shasum -a 256 /Users/admin/kiro/backend/libs/mathengine/games/magic_vault_5x3/magic-vault-5x3.yaml

The config SHA-256 is computed from the exact bytes the binary embeds (magic_vault_5x3.RawYAML()), so a hash recorded here describes the config that actually ran.

#Documents

#Backend artifacts

#Reproduction commands

cd /Users/admin/kiro/backend

# Contract, exact results, golden vectors, carry-over, max-win witnesses
go test -race -count=1 ./libs/mathengine/games/magic_vault_5x3

# Studio preset and anti-drift
go test -race -count=1 ./apps/math-studio/internal/domain/packs -run MagicVault

# Seeded evidence runs
go test -tags mvdraft -count=1 -v ./libs/mathengine/games/magic_vault_5x3 \
    -run 'DraftSimulation|FeatureConditioned' -timeout 40m

#Results bound by this manifest

Exact and estimated are separate columns throughout. Nothing sampled is labelled exact.

Total RTP = exact base RTP + exact P(trigger) × estimated E[award | trigger].

#Seeds and sample sizes

Shard seeds and boundaries are listed in the simulation report.

#Method boundaries

• The whole game is not enumerable. ExactCalculate refuses it (stochastic_feature), and that refusal is asserted rather than worked around. There is no exact whole-game RTP and no exact whole-game max-win distribution.
• Exact base results are produced by the same engine that executes the game. They are internally exact; they are not independent verification.
• The max-win claim is a reachability witness plus a containment argument, corroborated by 3,000,000 executed outcomes. It is not an exhaustive enumeration.
• Internal tests and simulations are not certification.

#Unresolved drift and open items

1. No runtime parity corpus. magic-vault-5x3 is not registered in libs/mathengine/parity; recording new vectors was out of Draft scope. Reference↔runtime parity for this game is unproven and is a prerequisite for any Final.
2. All-simulated total sits below the acceptance band as a point estimate (95.5731%), at −1.33 SE from the reconciled 95.9998%. Its own SE (0.3209pp) is wider than the band, so it is reported as a compatibility check rather than as the acceptance measurement.
3. Full-grid rate disagrees between runs at low count: 0.1075% (27 events, whole-game) vs 0.0743% (743 events, feature-conditioned). Use the latter.
4. WILD 3- and 4-of-a-kind are structurally unreachable under the shared LinesEvaluator, so WILD carries a 5-entry only. Tested and documented; changing it is a shared-engine decision, not a pack decision.
5. Base target deviates from the design brief's 55.44% (achieved 55.9961%), because the inherited conditional mean did not apply. Fully recorded in the changelog.

#Status

draft-v0.1 — DEV ONLY. NOT FOR PRODUCTION. NOT FOR CERTIFICATION.

Draft math approved 2026-08-10. Lifecycle state remains draft and remains mutable.

Not frozen, not published, not routed, no feature flag, no deployment, no operator enabled. Tuning, parity vectors, runtime handoff and any Final version require separate explicit approval. The unresolved items above — chiefly the absent runtime parity corpus — are prerequisites for a Final and were not closed by Draft approval.