M4 · results explorer

Benchmark results explorer

Inspect collected benchmark runs across workloads, targets, and cache states. Filter by workload or target using the controls below — the URL hash updates for sharing. No performance conclusions are drawn here; this is an inspection surface for raw run data.

No accepted corpus claim. Results shown are individual run records from the local KV store. Statistical comparison requires the M1 pilot corpus and preregistered stopping rules.

Workload × target matrix

Sample counts per workload and target. Cells show the number of collected runs, not performance conclusions. Empty cells indicate no data collected.

Loading matrix…

Run records

Individual run records with correctness status, workload, target, and captured-at timestamp. Click a row to inspect the full JSON record.

Loading runs…

Aggregate counts

Loading summary…

What this shows

Individual benchmark run records from the local KV store, with workload, target, cache state, and correctness status.

What this does not show

Performance conclusions, timing comparisons, or competitive declarations. Those require the M1 pilot corpus with preregistered stopping rules and statistical analysis.

Data source

Runs are stored via POST /v1/runs in Deno KV. The explorer fetches from GET /v1/runs with URL hash-backed filters for shareable views.