Editorial criteria and provenance
This reference separates three questions: is a method useful for a concrete task, where is its source instance, and has its effect been measured?
Admission criteria
Each criterion scores 0 for absent, 1 for partial or abstract, and 2 for specific and complete. Admission requires at least eight of ten, with the first three each scoring two.
| Criterion | What the review checks |
|---|---|
| Scenario | A concrete task and known conditions |
| Mechanism | An instruction or operation the reader can apply |
| Contrast | Bad and good examples for the same task and facts |
| Observable result | A result or artifact the reader can inspect |
| Boundaries | Conditions where the method needs adaptation or should stop |
These are editorial criteria, not measured success probabilities. Duplicate mechanisms merge; weak or incorrect methods are revised and reassessed, or withdrawn. Old identities remain reachable through migration pages.
Source states
A located source has a repository, immutable commit, file, section and line range. It proves an observed instance. It does not establish the first inventor. Earliest provenance is recorded separately.
Source unconfirmed means the original text has not been located. A method can retain this label while meeting the same usefulness criteria. Finding a source does not excuse bad advice: examples and factual claims are reviewed independently.
Teaching examples and experiments
Every pattern shows independently constructed teaching examples. They are not historical model outputs. Actual experiments identify their fixtures, model/configuration, run count and available outputs. Historical reports without raw outputs are labelled accordingly, and their numerical claims are not promoted to current evidence.