Findings
Full evaluations of named deployed systems, published quarterly. Every completed evaluation is published, whether or not the result favours the party that commissioned or funded the work.
The first evaluations are in progress. Nothing is listed here until the subject has had its right of reply and that window has closed.
When an entry appears it will carry the system evaluated, the operator, the execution window, the access tier, the sampling design and its denominator, the Quantile Report with intervals, and links to the full report, the harness, the notebook, the results file, and the test set where that can be released.
Four tiers, four cadences.
| Type | Cadence | What it is |
|---|---|---|
| Findings | Quarterly | Full evaluation of a named deployed system. Complete method, data, uncertainty, right of reply, notebook. |
| Notes | Monthly | Short empirical analyses. A single figure, a replication, a partial result, a methodological observation. |
| Methods | As developed | Harnesses, reference datasets, protocols, and the Quantile Report specification. Open licensed and citable. |
| Reviews | Opportunistic | Assessment of someone else's published evaluation. Did the study support its conclusion, and what did its access permit? |
Uncertainty on every figure. The access tier stated. The harm threshold fixed before analysis. The notebook and results file published alongside. The subject given at least twenty-one days of reply, published unedited. Corrections issued as prominently as the original.