Audit a cymba game without trusting cymba

Export the PAR sheet

Open the Game Configs page on your dashboard, find the config, and click Download PAR Sheet. The PDF bundles the full config, theoretical RTP, and five deterministic verification fixtures.

Alternatively, use the API for a JSON bundle that also includes inline spec HTML:

# Bash
curl -s -H "Authorization: Bearer cy_live_..." \
     https://cymba.dev/api/v1/par-sheet/my-config-id \
     > par-sheet.json

Recompute theoretical RTP independently

Your math person follows §5–§16 of the math model (bundled in the PAR sheet as spec_full_html) using standard slot-math techniques. For configs with ≤ 100M combinations they enumerate exactly; otherwise they Monte Carlo. The output is compared to theoretical_rtp.total_rtp in the PAR sheet.

# Pseudocode (any language)
totalCombinations = product( len(strip) for strip in config.reels.strips )
EV = 0
for combo in 0..totalCombinations-1:
    stops   = mixRadixDecompose(combo, stripLengths)
    grid    = buildGridFromStops(stops, config)
    payout  = resolveMath(grid, config) / linesPlayed   # per math model §5
    EV     += payout / totalCombinations

rtp = EV * 100
assert |rtp − par_sheet.theoretical_rtp.total_rtp| < tolerance

Tolerance: bit-identical for exact enumeration; max(2σ, 1% relative, 0.1% absolute) for Monte Carlo at reasonable sample counts.

Replay the verification fixtures

Five deterministic triples in the PAR sheet. Auditor reproduces each spin using their own implementation of §2 (RNG) + §5–§13 (math) from the PAR sheet bundle. Every result_hash must match byte-identically.

# For each fixture in par_sheet.verification_fixtures:
outcomes = ProvablyFairOutcomeSource(
  fixture.inputs.server_seed,
  fixture.inputs.client_seed,
  fixture.inputs.nonce
)
result = resolveSpin(
  par_sheet.config,
  outcomes,
  round_id = fixture.inputs.round_id   # reuse the published round_id
)

# Canonical encode per protocol §17, then HMAC over the result (minus result_hash itself):
payload = canonicalEncode( result without result_hash )
expected_hash = HMAC_SHA256(key=fixture.inputs.server_seed, data=payload)

assert expected_hash == fixture.expected.result_hash
assert configHash(par_sheet.config) == fixture.expected.config_hash

If any hash differs: either the auditor's implementation deviates from the spec, the config was tampered with, or the engine has a regression. All three are detectable.

Bind past spins to the PAR sheet

Every spin response carries config_hash. To prove a past spin used the config described in the PAR sheet, compare:

assert spin_response.config_hash == par_sheet.config_hash

Mismatched hashes mean the spin was generated by a different config than the one in the PAR sheet — cymba can't quietly swap configs without the auditor noticing.

Verify the provably-fair commit

For real player spins (not fixtures), the customer logs server_seed_hash + nonce per spin. After POST /api/v1/seed/rotate reveals the prior seed:

assert sha256(revealed_server_seed) == previously_committed_server_seed_hash

# Then replay each spin (as in step 2) using the now-revealed seed.

Commitment proves cymba locked in the seed before the bet was placed — it could not have chosen outcomes after seeing the player's stake or behaviour.