Audit Candidate Eou Set

Audit Candidate EOU Set

Audit $path (a candidate-set artifact under foundry/self-evolution/candidate-sets/, schema schemas/candidate-set.schema.yml — ECP-0013).

Optional input (ECP-0017 / Rule 96)

If foundry/captured-workflows/cw-{app_id}.yml exists with all four human_approval gates populated, load it. The Set Value Coverage Test runs against its domain_values block.

Required reading

1. foundry/constitution.yml
2. foundry/registry.yml
3. foundry/governance.yml
4. foundry/meta-eous/audit-candidate-eou-set.yml

Stop conditions

Halt and report before running tests if:

• $path does not resolve to a readable YAML file.
• The file does not contain a candidates array, or candidates is empty.
• foundry/constitution.yml or foundry/registry.yml does not exist — required for Authority and Non-Overlap tests.

Tests

Run each test against every candidate in the set.

Boundary Test

Each candidate must have one distinct success criterion. A candidate whose success criterion subsumes another candidate's fails this test. Flag candidates with compound or overlapping success criteria.

Non-Overlap Test

No candidate may duplicate an existing EOU in foundry/registry.yml or another candidate in this set. Check by purpose, target_object, and success criterion. Flag overlapping pairs and recommend merge or rejection.

Minimality Test

Each candidate must fail the following substitution checks — if it passes any, it should be converted instead of created as an EOU:

• Can this be a rule (static constraint, no execution required)?
• Can this be a validator (schema constraint, machine-checkable)?
• Can this be a regression case (captures a known failure)?
• Can this be a stop condition on an existing EOU?
• Can this be a checklist item inside an existing EOU's steps?

Authority Test

Generating EOUs in the candidate set must not have authority_level set to approve, publish, or mutate_active. Flag candidates claiming more authority than write_candidate or write_inactive without constitutional justification.

Operational Value Test

Each candidate must satisfy at least one of:

• prevents_failure: names a concrete failure mode it blocks
• improves_decision: names a judgment it makes explicit
• exposes_hidden_judgment: surfaces an implicit step that is currently untracked
• improves_traceability: adds observable accountability to a previously opaque step

Reject candidates that exist only for completeness or that duplicate the purpose of an existing artifact.

Counter-Generation Test

Each kept candidate must include arguments_against — the strongest case for not creating it. Candidates missing this field are not ready for specification.

Set Composition Test

If the candidate set includes EOUs that generate outputs, it must also include EOUs (or identify existing ones) for auditing and approving those outputs. A generation-only set without a corresponding audit path is incomplete.

High-Stakes Test

Candidates touching finance, health, legal, safety, content about minors, public claims, publication, or active governance must have responsibility.approver set to a named human role and escalation.require_human_when non-empty. Flag candidates in these domains missing human ownership.

Set Value Coverage Test (ECP-0017 / Rule 96)

Skip if no captured_workflow is loaded with complete human_approval.

For each domain_value of priority ≤ 3 in the loaded captured_workflow, verify that at least one candidate in minimal_recommended_subset operationalizes the value via its distinct_success_criterion (the value's id must appear in the criterion text). Emit findings in audit_outcome.notes naming each unserved value and its priority. Unserved priority-1 values escalate the verdict to FAIL; unserved priority-2 or priority-3 values escalate to REVISE.

Verdict thresholds

Output

Write the audit report to foundry/audits/candidate-set-audits/{set_id}.audit.yml:

set_id:
audit_date:
verdict:  # PASS | REVISE | FAIL
candidate_recommendations:
  keep:    # list of candidate IDs — pass all tests
  merge:   # list of (id_a, id_b, reason) pairs
  defer:   # list of (id, reason) — valid but out of scope
  reject:  # list of (id, reason) — fails tests
  convert: # list of (id, convert_to: rule|validator|regression_case|stop_condition|checklist)
findings:
  - candidate_id:
    test:
    severity:  # critical | high | medium | low
    description:
    required_action:
required_revisions_before_specification:
  - # list of changes needed before any candidate can proceed to eou-specify

Constraints

• Do not modify the candidate set file — write only to the audit output path.
• Do not apply verdicts retroactively to an existing audit report — write a new dated report.
• Do not set verdict: PASS if any Authority or High-Stakes test fails, regardless of other results.
• Do not promote any candidate to lifecycle_stage: active — that is the role of $eou-promote after human approval.
• Treat arguments_against as required, not optional — candidates missing this field fail the Counter-Generation Test.
• A PASS verdict does not authorize specification — candidates still require human review before proceeding to $eou-specify.