30 frontier model cards were inspected to chart how labs report benchmarks. The data, drawn from a curated registry, shows when each organization first publishes a benchmark—marked by an orange diamond on the timeline—while subsequent cards from the same group appear as gray ticks, keeping the cumulative staircase flat. Cards without a publication date are omitted from the visual timeline but still contribute to overall totals, illustrating reporting saturation rather than score saturation.

Benchmark counts are tallied per model card: a single card that lists AIME in multiple configurations counts the same as one that lists it once. This prevents lengthy appendices from inflating a vendor’s share. The system also distinguishes shared standards (six vendors) from house‑style reports (one vendor), offering a clear view of industry consensus.

Daily evidence and attention volumes are tallied separately, with each bar representing an independent count. Releases that merely re‑announce existing data are excluded, ensuring the ledger reflects fresh contributions. The dashboard, though currently unavailable, aggregates source mix, categories, events, and fetch health to provide a comprehensive snapshot of frontier benchmark activity.