concept: rubric.md
How the rubric grades a terse artifact
rubric.md is the scoring half of the spec -- a fixed set of exactly 7 criteria a candidate a2a.terse.v1 artifact is graded against. Each criterion targets a distinct failure mode the style is designed to prevent, from missing structure to non-portable YAML to unsourced claims.
| criterion | what it catches |
|---|---|
structural_fidelity | A required top-level key is missing, empty, or mistyped. |
terseness | Redundant keys, prose duplicating structured fields, over-long steps. |
io_contract | An input/output field is untyped, or a step/eval entry references something undeclared. |
self_eval_tail | The eval array is missing, empty, or trivially unfalsifiable. |
cross_platform | A YAML construct that parses differently across standard 1.2 loaders. |
determinism | Ambiguous ordering, locale-dependent coercion, or hidden external-state dependence. |
traceability | No version marker, or derived content presented without citing its source. |
An artifact does not need a numeric score to be useful here -- the rubric is meant to be read as a checklist first, a scoring instrument second. Browse the live rows →