Pragma · Leaderboard
Leaderboard scores are reproducible against a pinned methodology version. When a score is objectively wrong given that version, we correct it — in public, on a fixed SLA, with a tamper-evident record. When a score is simply lower than someone hoped, that is not a bug.
Past editions stay frozen at the methodology version they ran against, so scores remain comparable across runs. A correction amends the disputed record on its own edition; it does not retroactively rewrite other editions.
Every dispute and ruling — regardless of outcome — is published at /leaderboard/disputes. Each record carries an audit-chain hash; once Specshift Cloud anchors a published edition, those hashes are written to immutable, write-once storage, so anyone can verify the public record matches the original submission. No silent corrections. No silent rejections.
One dispute per company per edition, with a per-IP fallback of five submissions per hour, to keep the ledger signal-rich.