Methodology
How reports are scored
Reports rank services with the Stackbroker composite score (version v0) — practical routing factors, not abstract reputation. Missing evidence is shown honestly and scored at a neutral midpoint rather than fabricated. The full routing methodology — factor definitions, neutrality invariants, and the versioned changelog — lives at /methodology.
| Factor | Weight |
|---|---|
| Downstream task success | 20% |
| Schema conformance | 15% |
| Cost per successful task | 15% |
| p95 latency | 15% |
| Failure-mode legibility | 10% |
| Provenance / receipt quality | 10% |
| Idempotency / replay safety | 5% |
| Policy fit | 5% |
| Freshness / source confidence | 5% |
Published reports snapshot the ranking at publish time and are immutable for reproducibility. A report shown as a preview is computed live from the current evidence base and may change until published.