Every captured round re-derives exactly from the rules (5,000 rounds, 0 mismatches), the balance never drifts, and the observed RTP, hit rate and Coin Burst frequency match the exact model. One statistical flag on reel 1 did not reproduce in 5 fresh sessions, so it's recorded as a watch item, not a bug.
The verifier (tools/verify.js) was written from the rules document, not from the game code. The exact model goes through all 27,000 stop combinations.
| Re-test session | Reel 1 p | Reel 2 p | Reel 3 p |
|---|---|---|---|
| Main capture (seed 2026) | < 0.001 | 0.415 | 0.253 |
| Fresh session, seed 11 | 0.036 | 0.272 | 0.649 |
| Fresh session, seed 23 | 0.283 | 0.086 | 0.272 |
| Fresh session, seed 57 | 0.681 | 0.971 | 0.991 |
| Fresh session, seed 101 | 0.889 | 0.641 | 0.472 |
| Fresh session, seed 404 | 0.732 | 0.037 | 0.501 |
| Result | Check | Area | Actual |
|---|---|---|---|
| Pass | PW §1 loads with the start balance and the lowest play amount | Behaviour · desktop | Passed in Playwright |
| Pass | PW §8 − and + step through every play amount and disable at the ends | Behaviour · desktop | Passed in Playwright |
| Pass | PW §8 controls are locked while a round plays | Behaviour · desktop | Passed in Playwright |
| Pass | PW §7 balance after every round = before − play + win (20 rounds, mixed amounts) | Behaviour · desktop | Passed in Playwright |
| Pass | PW §3–§7 every round the server returns re-verifies independently (100 rounds) | Behaviour · desktop | Passed in Playwright |
| Pass | PW §7 a play larger than the balance is refused with no round and no debit | Behaviour · desktop | Passed in Playwright |
| Pass | PW §8 history lists the last rounds, newest first, with the right wins | Behaviour · desktop | Passed in Playwright |
| Pass | PW §1 nothing on screen uses gambling wording ("bet", "wager", "cash") | Behaviour · desktop | Passed in Playwright |
| Pass | PW §1 loads with the start balance and the lowest play amount | Behaviour · phone | Passed in Playwright |
| Pass | PW §8 − and + step through every play amount and disable at the ends | Behaviour · phone | Passed in Playwright |
| Pass | PW §8 controls are locked while a round plays | Behaviour · phone | Passed in Playwright |
| Pass | PW §7 balance after every round = before − play + win (20 rounds, mixed amounts) | Behaviour · phone | Passed in Playwright |
| Pass | PW §3–§7 every round the server returns re-verifies independently (100 rounds) | Behaviour · phone | Passed in Playwright |
| Pass | PW §7 a play larger than the balance is refused with no round and no debit | Behaviour · phone | Passed in Playwright |
| Pass | PW §8 history lists the last rounds, newest first, with the right wins | Behaviour · phone | Passed in Playwright |
| Pass | PW §1 nothing on screen uses gambling wording ("bet", "wager", "cash") | Behaviour · phone | Passed in Playwright |
| Pass | M-1 Every captured round re-derives from the rules (lines, wilds, Coin Burst, cap, rounding, balance) | Maths | 5,000 rounds, 0 mismatches |
| Pass | M-2 Balance chains from round to round with no drift | Maths | Each round starts at the previous round's end balance |
| Pass | M-3 Verifier self-test: it catches a planted 1-cent overpay | Maths | 1 mismatch found in the tampered capture (R01003) |
| Pass | S-1 Observed RTP is consistent with the exact model | Statistics | 90.28% vs 94.781% (z -1.70) |
| Pass | S-2 Hit rate is consistent with the exact model | Statistics | 38.98% vs 38.72% (z 0.37) |
| Pass | S-3 Coin Burst frequency is consistent with the exact model | Statistics | 41 in 5000 vs 1 in 125 (z 0.16) |
| Watch | S-4.1 Reel 1 stops are uniform | Statistics | chi² 62.86 on 29 df, p < 0.001 · not reproduced in 5 re-tests |
| Pass | S-4.2 Reel 2 stops are uniform | Statistics | chi² 29.98 on 29 df, p = 0.415 |
| Pass | S-4.3 Reel 3 stops are uniform | Statistics | chi² 33.63 on 29 df, p = 0.253 |
| Note | Q-1 Cap before or after rounding? (§9) | Open question | Observed: cap first, then round down. The highest possible win is 25.6× play, so the 500× cap is never reached |
| Note | Q-2 RTP (§9, not stated) | Open question | Exact RTP 94.781% from all 27,000 stop combinations |
The game was played in Chromium (desktop and Pixel 7 emulation) with Playwright, which also captured each round's payload. Every payload was re-checked by the independent verifier. A separate capture of 5,000 rounds fed the statistics. Statistical flags were re-tested on fresh sessions before any call was made. Everything here can be reproduced with npm run all.