#636 What the reads of #628's merge with #625 left (#612)

A follow-on to #628 (issue 612, "The eval job reads a named set, and a gate set's report is aggregates only"). It fixes the two low findings from reading the code that #628's merges with #625 and #624 changed, after #628 landed as 78f6836e.

What changes

• models is held as required in eval-aggregate/1. It adds:

  • a counter-example, examples/invalid/eval-aggregate.no-models.json;
  • a case in test_the_gate_passes_a_good_aggregate_report_and_refuses_a_bad_one that refuses a gate report with no models;
  • the mutant "a gate set's report need not name the models that answered", on the schema.

  eval-report/1 already had the same three, from #625.

• Record 0164 gains a Boundaries row. The row names the answering models, each case's usage in the ledger, and the six mutants that hold them. A row is only added, as check_decisions requires. The Decision row "The ledger" stands unedited: it lists fewer fields than the ledger holds, and says nothing false.

The delta reads

Read	Result
Argon, on the merge with #625 (from the mechanical merge, conflicts as git left them, to the resolution)	A clean read: 4 of 4 parts, 0 findings
Claude review, on both merges	No correctness bug; the two low findings fixed here
Claude security, on both merges	Nothing to fix. A maxItems on models was suggested as an optional backstop, and is not taken: the CLI names each model, not the model's own text
Muse, Astra	Not run (quota); owed after 2026-11-09

Checks

Check	Result
SIM_SEEDS=40 make verify	OK, 1677 tests (at 4e5757e1)
Mutants: the new one and #625's on eval-report	2 of 2 killed
gitleaks	No leaks

Reads of this PR (54c74fb4): Argon clean (0 findings); Claude review and security, no blocking finding, one low (the mutant filed under decision 0162's block) fixed in 4e5757e1. Muse and Astra: not run (quota), owed after 2026-11-09.

🤖 Generated with Claude Code