git:20260916.cb7fd6c to git:20260916.9d75634

+ | `report` | Read the target repo's `.docgrad/scorecard-latest.md` and reprint it, plus a per-round score trend drawn from `.docgrad/history.jsonl`. If the files do not exist, tell the user to run improve/loop first (a plain audit is report-only and writes nothing). **Check for branch divergence first**: `git rev-list --count HEAD..docgrad/converge` (when that branch exists) > 0 → put a warning at the top of the report: "the converge branch is N commits ahead of this branch, the trend below may be incomplete". The report header always states its data source as `<branch> @ <short-sha>`. **Legacy rows** (no `schema`) are shown as a separate "1.x rounds" block and are never joined to schema-2 rows: draw one break before the first schema-2 row, "v2 changed what is measured; rows above are 1.x and cannot be compared". Inside the legacy block the 1.x rules apply unchanged, kept as the legacy rules: (a) `rubric_hash` differs from the previous round → draw a break line noting "the ruler changed here, scores before and after cannot be compared directly"; (b) `corpus_hash` differs from the previous round → draw a break line noting "the corpus scope changed here: `files_total`, `claims_total`, the freshness denominator and the pollution denominator all moved, so scores either side cannot be compared"; (c) the measure fingerprint — read as `measure_hash`, or as `thresholds_hash` when that is the name present — differs from the previous round → break, same wording as (a); (d) the judge fingerprint — read as `judge_hash`, or as `judgement_hash` when that is the name present — differs from the previous round → break, same wording as (a); (e) rounds with no `economy` key are from the five-dimension era before v1.0.0 → draw `—` for that dimension and note "the rounds below are five-dimension; targets-met and overall scores cannot be compared with newer rounds"; (f) a field missing from a legacy row is unknown and draws no break — this covers every legacy shape: no fingerprints at all, only some fingerprints, old names, and new names without `schema` (rows written between v2.0.0 E1 and this change). This is also where rubric.md §Version history's "report has to map them" is honoured: the name mapping in (c) and (d) is that mapping. **Schema-2 row violations** are each printed as "row N violates history schema 2: <what>": a missing `docgrad` object; a missing `measure` object; or any of these required `docgrad` keys missing — `version`, `rubric_hash`, `measure_hash`, `judge_hash`, `corpus_hash` (the key set `docgradMeta()` emits today). The required-key check applies only when the `docgrad` object is present — a missing object is one problem, not six. Each violating row gets exactly one line listing all of its problems, separated by `; `, for example "row 2 violates history schema 2: missing docgrad" or "row 4 violates history schema 2: missing corpus_hash; missing version". A key not in the list is ignored; the violating row is still shown, but no break decision is inferred from it. **Comparison baseline**: every schema-2 break decision compares a row with the previous schema-2 row that is not a violation — a violating row is skipped as a baseline. A key that is present with value `null` is a value, not a violation (`corpus_hash` is null without a config, and `version` can be null per lib.mjs `docgradMeta`); null compares equal only to null. **Measure trend**: one series per measure id, number first and verdict beside it; break the measure trend when `docgrad.measure_hash` or `docgrad.corpus_hash` differs from the baseline, naming which one moved. **Judge series** (optional to draw) is labelled "judge — not comparable across rounds", and never averaged or summed; break it when `docgrad.judge_hash` or `docgrad.rubric_hash` differs from the baseline — a judge-series break alone does not break the measure trend. **No overall score is ever computed.** When `.docgrad/ledger.jsonl` exists, also report cumulative coverage and which claims are still `fail` or `stale` |