Pack Odds Verdict

Verify that the odds a mystery or blind pack ADVERTISES are the odds it actually DELIVERS.

A fairness and disclosure tool for anyone selling collectible packs, blind boxes, randomized bundles or loot-box-style products. Its job is what a disclosure regime asks for: prove published odds match reality. Given published per-tier odds and a pack design, it runs a SEEDED, REPRODUCIBLE Monte Carlo simulation and returns a verdict -- PASS, FAIL naming every tier that diverged and by how much, or a REFUSAL when the trial count cannot support the tolerance requested.

IT DOES NOT TELL YOU WHETHER YOUR ODDS ARE FAIR. It tells you whether your odds are what you SAID they were.

FIVE SILENT FAILURES, each producing a pack that looks fair and is not, with no error anywhere.

1. PUBLISHED ODDS ARE NOT THE REALIZED ODDS. A seller declares per-slot probabilities, then the real build applies rules on top -- no duplicates within a pack, a guaranteed rare in a flagship slot, a pity counter -- and each rule shifts the true distribution away from the published one. Nobody recomputes. MEASURED: three completely defensible rules push a published 4% epic rate to a REALIZED 36%.

2. EXPECTED VALUE QUOTED FROM THE MEAN OF A SKEWED DISTRIBUTION. Pack value is long-tailed; a few huge hits carry the average. "Buyers get $12.88 of value" and "the median buyer gets $7.54 and 60% of buyers get less than the advertised average" are BOTH TRUE about the same pool. So distributionReport() always returns mean, median, p10/p90 and the share of buyers below the mean TOGETHER -- there is no function in this package that returns "the" expected value as a single number.

3. AN UNSEEDED SIMULATOR CANNOT AUDIT ANYTHING. If a fairness run cannot be reproduced exactly from a recorded seed, its result is an anecdote. Every simulation records its seed, and the same design, seed and trial count reproduce the exact same verdict on any machine.

4. PERCENTILE METHOD DISAGREEMENT. Nearest-rank, linear interpolation and half a dozen named variants disagree on small samples. This package picks exactly one -- linear interpolation between closest ranks, numpy's method="linear" default / R type 7 -- names it in the source, and pins it with a hand-verified, numpy-cross-checked test.

5. FLOATING-POINT ACCUMULATION DRIFT over hundreds of thousands of trials. Money is tracked in integer minor units during accumulation, never as a running float total, and every summation function REFUSES a bare float input.

IT REFUSES rather than guesses: a tolerance the trial count cannot support returns a refusal naming the trials required, because a fairness verdict computed at a resolution the sample cannot carry is a confident number about noise.

VERIFIED: 107 tests, measured by running the suite.

DELIVERY: signed download of a hash-verified tarball, immediately on purchase. Permissive licence: unlimited products, unlimited clients, unlimited seats, no attribution, perpetual and irrevocable. One restriction, do not republish the source as source.

Interface

What you call, and what comes back. Types and signatures only — the implementation ships with the source.

  export function dollarsToCents(dollars: number): number;
  export function centsToDollars(cents: number): number;
  export function sumCents(centsList: readonly number[]): number;
  export function validatePublishedOdds(odds: PublishedOdds): void;
  export function validatePackDesign(design: PackDesign): void;
  export function drawTierFromOdds(odds: PublishedOdds, rng: SeededRng): string;
  export function drawPack(design: PackDesign, rng: SeededRng, pity: PityCounters): string[];
  export function buildPool(tiers: Tiers, rng: SeededRng): PoolCard[];
  export function designPool( price: number, n: number, cvr: number, p: readonly number[], m: readonly number[], ): DesignPoolResult;
  export function poolStats(tiers: Tiers, price: number): PoolStats;
  export function varCostFraction(cfg: VariableCostConfig): number;
  export function pyRound(x: number): number;

  export type PublishedOdds = readonly PublishedOddsEntry[];
  export type TierName = (typeof TIER_NAMES)[number];
  export type Tiers = Record<TierName, Tier>;
  export type VerdictStatus = "PASS" | "FAIL" | "INSUFFICIENT_TRIALS";