# BetFury "Provably Fair" — Independent Audit

**BetFury advertises a 99.02% RTP on its in-house Original games and calls them "provably fair."**
This repository holds the complete evidence behind the ProvablyFair.org investigation of three of
those games — Limbo, Dice and Plinko: **240,000 real bets**, of which **180,000 instrumented rounds
are verified one-by-one against BetFury's own cryptographic commitments**, plus a 60,000-bet
exploratory capture published as corroboration.

| Game | Measured RTP | Advertised | Deviation | |
|------|:---:|:---:|:---:|---|
| **Limbo** | **97.326%** | 99.02% | **−1.69 pts** | win rate 88.478% vs the displayed 90.018% — **12.58σ below the game's own odds** |
| Plinko | 99.008% | 99.02% | −0.01 pts (−0.02σ) | consistent with advertised |
| Dice *(control)* | 99.393% | 99.02% | +0.37 pts (+0.92σ) | consistent with advertised |

**Limbo's sample house edge is ~2.7× the advertised rate.** Dice and Plinko — measured the same
way, at the same time, on the same account — land on their advertised mathematics, so the rig that
finds Limbo short finds the controls on target: the deficit is specific to Limbo, not a measurement
artifact. The shortfall is serially independent (a skewed payout distribution, not a streaky RNG),
and it appeared twice: an earlier 60,000-bet exploratory run at a 2.00× target returned **97.890%**
with one instant bust every **29** bets against one-in-**51** expected.

📄 **Read the full report** at [ProvablyFair.org](https://provablyfair.org) — verdicts, the
statistical treatment, and what "provably fair" would actually require.

## Why you can trust these numbers

Every instrumented row is a real, unaltered BetFury outcome. BetFury commits to each round by
publishing `SHA-256(randomSeed)` where `randomSeed` is `<result>_<random>` — the outcome in
plaintext. We recorded that seed and hash for every bet; **`SHA-256(seed)` reproduces BetFury's own
published hash on all 180,000 rounds**, so nobody can claim the log was fabricated — the hashes are
BetFury's. (That the commitment contains the *finished result*, with no player input and no
derivation formula, is also precisely why the system is not provably fair — Part 1 of the report.)

```
$ node scripts/verify.mjs
limbo   60000 rounds | SHA256(seed)==hash: 60000/60000 | result-in-seed: 60000/60000  ✓ PASS
plinko  60000 rounds | SHA256(seed)==hash: 60000/60000 | result-in-seed: 60000/60000  ✓ PASS
dice    60000 rounds | SHA256(seed)==hash: 60000/60000 | result-in-seed: 60000/60000  ✓ PASS

TOTAL   180000 rounds | 180000/180000 hash-verified | 180000/180000 result-in-seed
```

## Reproduce every figure

Requires only Node.js — no dependencies, no network.

```
$ node scripts/analyze.mjs
══════════ LIMBO  (n=60000) ══════════
INTEGRITY  hash-verified: 60000/60000 ✓ ALL | index ordered: yes | stake: 0.001
HEADLINE   money RTP: 97.326%  vs advertised 99.020%   z=-11.81
           95% CI [97.045%, 97.607%]
           win@1.1x: 88.478% vs BetFury's shown 90.018%   z=-12.58
[…]
══════════ SUMMARY — measured RTP vs advertised 99.02% ══════════
  limbo     97.326%  z= -11.81
  plinko    99.008%  z=  -0.02
  dice      99.393%  z=   0.92  (control)
```

Two z-scores appear for Limbo because they test different claims: **−11.81** tests the money RTP
against the advertised 99.02%, and **−12.58** tests the observed win rate against the 90.018% win
chance BetFury's own bet panel displays. They are separate statistics — both computed by the
script, both fatal to the advertised figure.

```
$ node scripts/exploratory.mjs
══════════ LIMBO EXPLORATORY  (n=60000, target 2.00x) ══════════
HEADLINE   money RTP: 97.890%  vs advertised 99.020%   z=-2.77
BUSTS      1.00x rounds: 2072 (3.453%, 1 in 29)  vs 1176 expected under advertised (1.960%, 1 in 51)   z=26.4
IMPLIED R  0.97512  (95% CI 0.9736-0.9766)  vs advertised 0.9902
AGREEMENT  instrumented run implies R 0.97415; difference 0.91 sigma -> consistent ✓
```

The exploratory capture (15–16 July, before the instrumented runs) recorded results, payouts and
timestamps but not the seed/hash fields, so it is **not pooled** with the verified 180,000 — it is
corroboration. Fitted on its own, it points at the same payout constant as the instrumented run.

## Method

- **Pre-committed.** Fixed `n = 60,000` per game, declared before the first bet — the
  `data/*.precommit.json` manifests are timestamped seconds ahead of each game's first recorded
  round. One analysis, no extending, no cherry-picking.
- **Single stake, single currency, no bonus bets.** `0.001 USDT` per bet, full record logged per
  round in capture order with index and timestamp.
- **A built-in control.** Dice applies its edge in the *payout* (a true 50% win chance paid at
  1.9804× instead of 2× — confirmed on every row), so a correct measurement must land on 99.02% —
  and it does. Plinko's full paytable is embedded in the data (16-row Medium:
  120.4/41/10/5/3/1.5/1/0.5/0.3 mirrored; fair-model mean 99.02%). Money RTP (paid ÷ staked) is the
  single, mechanism-agnostic metric used everywhere.

## Repository layout

```
data/
  limbo-60000.jsonl              60,000 instrumented rounds — seed, hash, result, stake, payout
  plinko-60000.jsonl             60,000 rounds — incl. the full 16-step ball paths
  dice-60000.jsonl               60,000 rounds — raw rolls (the control)
  limbo-exploratory-60000.json   60,000 exploratory rounds at 2.00× (no seed/hash — corroboration)
  *.precommit.json               dated fixed-n pre-commitment per game
  analysis-summary.json          machine-readable results
  siq-rng-certificate-2022.pdf   BetFury's SIQ RNG certificate (local copy)
scripts/
  verify.mjs                     re-hash every round → matches BetFury's committed hashes
  analyze.mjs                    reproduce the RTP / distribution / independence analysis
  exploratory.mjs                reproduce the exploratory-run corroboration figures
```

Account identifiers and balances were stripped from the public data; nothing else was altered.

## Scope

Part 1 of the report (the system is not verifiable by players) is a statement about
*verifiability*, not a claim that any single result was altered. BetFury's 2022 SIQ certificate
(included) covers a submitted `Math.random()`-based PRNG sample and states that the RGS and live
games are the subject of separate reports; it contains no payback-ratio testing. Part 2 is a direct
measurement: in these captures Limbo paid below its advertised RTP; Plinko and Dice paid as
advertised. All findings are scoped to Limbo, Plinko and Dice on betfury.com, captured
15–16 July 2026, and every figure is reproducible from this repository.

---

Prepared by [ProvablyFair.org](https://provablyfair.org) — the certification standard for provably
fair gambling.