tkersey/review-fold

Review Fold

Mission

Turn witnessed falsification pressure into one immutable classified-bug artifact.

review or failure source + Goal Contract + current Construction
-> counterexample-set/v1
-> no mutation authority

$review-fold owns fact classification, disposition, and quotienting. $actuating owns evaluation against the current Construction, successor selection, orchestration, review credit, and closure. The source owner retains its own receipt semantics.

Minimal law

claim != observed fact
observed fact != liability
liability != accepted scope
accepted scope != selected repair
Counterexample Set != mutation authority

Counterexample Set

artifact:
  schema: counterexample-set/v1
  artifact_id:
  goal_id:
  semantic_author: review-fold
  created_at:
  predecessor_refs: []
  supporting_refs:
    - review-campaign-started-event-ref
    - review-attempt-or-failure-event-ref
  payload:
    subject:
      construction_ref:
      repository:
      artifact_digest:
      review_contract_digest:
    classes:
      - class_id:
        boundary_key:
        law_ref:
        discrepancy: excess | deficit | incoherence | partiality | misbinding
        owner_boundary:
        severity: critical | high | medium | low
        status: accepted | rejected | blocked | follow-up
        observed_facts: []
        evidence_refs: []
        finding_refs: []
        witness:
        falsifier_ref:
        applicability:
        quotient_basis:

For CAS-derived evidence, supporting_refs cites the current review_campaign_started event and each exact terminal attempt or transport event used by the Set. Actuating resolves those Evidence Ledger events and requires their campaign_id and request identity to match the campaign derived from this Set's existing subject tuple. Do not duplicate that derived campaign identity as a fifth Counterexample subject field. A non-review falsifier needs no campaign reference.

A current clean source may produce an empty classes list. One class represents one stable Counterexample to one governing law at one boundary. Review attempt IDs, commits, publication epochs, filenames, and proposed patches are provenance, not class identity.

For CAS evidence, each finding_refs entry is the sha256: digest of the exact canonical compact CAS finding-row bytes. Preserve the enclosing CAS receipt as attempt, tuple, request, and verdict provenance; none of those transient identities defines the Counterexample class or substitutes for the canonical-row digest.

Dispositions

• accepted: current evidence establishes an in-scope falsification.
• rejected: evidence shows the claim is false, stale, already satisfied, preference-only, or not a liability.
• blocked: validity, ownership, applicability, or current subject identity remains unknown.
• follow-up: valid evidence lies outside the accepted Goal.

Rejected classes require rejection evidence. Accepted classes require Actuating to evaluate the current Construction before mutation. A suggested repair remains source prose and never enters the selected construction merely because a reviewer proposed it.

Procedure

Before the first native Ledger command in this workflow, load $ledger and complete $ledger ensure once. Reuse Actuating's current adapter gate. When invoked standalone, require ledger --version to be at least 0.11.0 and verify that ledger --source actuation --help exposes only append|prepare|state|project|doctor|path before materialization.

1. Bind the source to the current Goal, Construction, subject digest, static Review Contract digest, and source-owner receipt. A failing test, incident, compatibility failure, or other non-review falsifier requires no review campaign. A CAS-derived set additionally binds its originating campaign, whose Review Contract digest must match the static digest in the Counterexample subject. Never fabricate a campaign for local evidence or make review_contract_digest optional.

2. Separate each claim, observed fact, and suggested repair.

3. Decide whether the fact is a current liability under an accepted Goal law.

4. Identify the governing law, stable boundary, discrepancy, owner, witness, falsifier, applicability, and evidence.

5. Quotient duplicate and same-class findings. One class may cite many finding rows and may recur across Construction successors. When a class recurs, the new Set's predecessor_refs must include the prior Set that most recently carried that class.

6. Assign exactly one disposition to every class.

7. Materialize canonical JSON with the current six-command artifact adapter:

   ledger --source actuation --repo <repo> --goal <goal-id> \
     append --input <counterexample-set.json>
   

8. Return the materialized Counterexample Set to Actuating. Do not propose or execute a repair.

Use review-fold.valid.example.json as a shape example, never as evidence or authority.

Guardrails

• Do not choose the review backend, lens, architecture, repair, work node, or next action.
• Do not count clean attempts or decide review convergence or closure.
• Do not turn style, speculation, or suggested patches into code.
• Do not accept scope expansion without source authority.
• Do not define class identity from transient implementation or review IDs.
• Treat Ledger materialization and validation as structural artifact work only. A pass grants no mutation, repair selection, review credit, or completion.