Methodology & evidence
Recommendations are deterministic outputs of build-versioned formulas. Uncertainty is presented as data, not hidden behind a confidence score.
Evidence labels
Verified means the formula or control flow was inspected for the supported build. Measured means it is calibrated from observations. Experimental means important runtime assumptions remain open.
Mousey evidence
mousey.rank-budgetVerifiedMousey receives one point per rank, or two after the Darts jackpot.
mousey.slot-gatesVerifiedPlayer-visible slots unlock at ranks 1, 2, 3, 5, 7, 10, 15, 20, and 25.
mousey.skill-formulasVerifiedAll nine Mousey curves and the Fire level cap were inspected in the current build.
mousey.magic-orderVerifiedMagic rolls are mutually exclusive in Fire, Gold, Arcane, Poison order.
mousey.xp-calibrationMeasuredXP source rates and uncertainty bounds come from Rank 35 A/B measurements.
mousey.gunner-ratioMeasuredThe high-progression Gunner/player throughput ratio is calibrated from observed runs.
- Fresh-run source-specific validation remains incomplete.
mousey.gold-source-modelExperimentalGolden and Alchemic formulas are verified, but source weighting still needs runtime calibration.
- Manual, Auto Throw, and Gunner gold weights require isolated measurements.
mousey.frictionless-runtimeMeasuredThe displayed curve is verified; empirical XP throughput showed no gain in the tested high-rate state.
- The empirical result is not proven universal across progression states.
mousey.external-chance-modifiersVerifiedThe mapped Zenith multiplier and additive Sushi chance use build-versioned formulas.
mousey.shadow-modelExperimentalShadow is an analytical expectation using configurable duration, damage, and source assumptions.
- Duration, source rates, and controlled Arcane/Poison runtime tests remain incomplete.
Zenith catalog evidence
zenith.catalog.build-23705722VerifiedAll 40 identities, grid positions, one-gem level costs, cycle gates, and five-level tooltip curves were recovered from the installed build and verified on 2026-07-18. Runtime stacking is tracked separately and remains Research pending, so the catalog does not claim an optimal purchase path.
Compatibility
Local calculation
Optimization runs in a Web Worker. The current public slice does not send calculator configuration to a backend and does not require an account.
