Proof
not syncedEvery reputation score comes from a deterministic replay of a frozen dataset. Recompute it on your own machine, hash the output, and confirm it byte-for-byte equals the latestDatasetHash committed on Mantle. Same inputs, same Brier scores, same consensus — provable off-chain and on-chain.
Scores aren't a black box. generate-frozen.mjs replays a fixed window of agent signals and outcomes with no randomness and no network calls. Run it twice, run it on any machine — you get the exact same CSV every time.
The SHA-256 of the frozen dataset is written on-chain. If a single byte of the track record changed, the hash would diverge — and this page would flip to not synced.
Run the deterministic generator. It rebuilds the frozen dataset from the committed agent signals — pure computation, no network.
node data/datasets/generate-frozen.mjs
Take the SHA-256 of the regenerated CSV. This is the canonical fingerprint of the entire track record.
shasum -a 256 data/datasets/frozen.csv
Read latestDatasetHash from the SibylLedger on Mantle and confirm it equals your locally computed hash.
16 golden vectors pin the math. The Foundry suite and TS replay assert identical Brier scores and consensus on both sides.
If the hash matches, the on-chain reputation is provably the same one this app shows. No trust required.
node data/datasets/generate-frozen.mjs # then: SHA-256(data/datasets/frozen.csv) == on-chain latestDatasetHash
Fixed (prediction, outcome) → Brier inputs with hand-checked expected outputs. Both the Solidity scorer and the TS replay must reproduce them exactly.
The on-chain SibylLedger scorer is fuzzed and unit-tested in Foundry — fixed-point Brier, inverse-weighting, per-agent caps, FLAT dead-band.
The off-chain replay is asserted against the same golden vectors and against on-chain reads, so the dashboard can never drift from the contract.