Every recording yields a run record naming its commit, the tree each report's run measured, its totals and what was absent, dropped or unknown, and a coverage summary joins it only bound to the commit and tree it was measured on

What it means

majordomus evidence record returns, beside what it merged into the ledger, the recording as one typed run, an EvidenceRunRecord:

• its id: the CI run it was (ci:<provider>:<run>:<attempt>), or the local commit and time;
• its origin, its commit and the weakest tree over the reports it was given;
• the measurement of each report;
• totals by outcome and by runner;
• which reports were absent and why, what the reports held that no claim can name (dropped), and the tests they named that this repository does not have (unknown);
• the executions, each of which maps back to the record's id.

--run-record <file> writes the same record. --coverage <summary> joins the summary scripts/rust-coverage --summary-json wrote as an EvidenceCoverage, which cannot be built without a full commit, carries the tree its run measured and the floors the threshold files state, and is current only for its own commit on clean trees.

How it works

evidence::record in apps/majordomus-cli/src/evidence/record.rs builds the record from the same values it merged: the totals through ledger::outcome_counts, the one count the ledger's own summary uses, and the dropped list the crate adapter produced, so the recording's answer and the record never disagree. The types are in apps/majordomus-cli/src/evidence/run.rs and apps/majordomus-cli/src/evidence/coverage.rs; CommitId refuses an abbreviated or upper-case id when it is parsed and when it is deserialised.

In CI, scripts/ci/evidence-collect reads its manifest's totals, tree and absences from the run record rather than computing them a second time from the ledger's rows.

How to see it

majordomus evidence stamp --producer suite --report suite.tsv --out suite.provenance.json
majordomus evidence record --suite suite.tsv --provenance suite.provenance.json \
  --run-record run.json --format json | jq '.run_record | del(.executions)'
bash test/run.sh 505_a_recording_carries_what_its_run_measured
bash test/run.sh 506_a_crate_binary_that_ran_nothing_is_not_a_pass

What it does not cover

There is no index of run records and no bound on how many are kept yet, and a run record holds one result per test binary, not per test. Nothing reads a run record to judge a claim: a later surface lets one withhold proven through the one monotone rule, and none does yet.

Why it exists

What a recording knew as a run was thrown away: the ledger keeps one execution per test, and a coverage summary carried no commit, so a number about some tree nobody could name. A typed run keeps what the run measured together and binds coverage to the commit it describes. ADR 0087 (proposed) records the model.