Adjudication Record

Defines what a Consensus Run or Adjudication record contains. Written only in the main session, never by a subagent — a subagent cannot surface an approval prompt, and an Adjudication requires a named human decider (07-stages.md, 02-plan.md §5).

Schema

schemas/adjudication.schema.json. One JSON object per file: consensus/<id>.json (mechanical) or adjudications/<id>.json (human).

Two kinds — do not conflate them

• kind: "consensus-run" — the mechanical output of a K-pass ensemble: union of candidates, partitioned into found-by-all / found-by-some / found-by-one. Re-executable; decided_by may name the process, not necessarily a person.
• kind: "adjudication" — a human decision over a Consensus Run's output (or over a boundary proposal, or a Divergence classification). decided_by must be a human identity. An agent name here is a schema violation, not a shortcut — the schema itself rejects the fixed v1 agent roster in this field when kind is "adjudication". Only supersedable, never re-executable.

Required fields

id (ADJ-NNNN), stage (0–8), k, candidates (each with candidate_id, found_by, decision), decided_by, decided_at.

Content obligations

Every candidate's partition should be set (all/some/one) so scripts/status-report.py's agreement distribution is meaningful. Every candidate's decision (accept/reject/triage) needs a rationale when it is not a straightforward accept — especially anything found-by-one, which Constitution C6 requires to be inspected, not auto-accepted or auto-rejected.

Behavior-ID / artifact citation

candidates[].candidate_id should be the ID of the thing being adjudicated (a Behavior ID, a Divergence ID, a boundary-proposal identifier) so the record is traceable back to what it decided.

Unknowns

If a candidate's evidence is too thin to decide confidently, use "decision": "triage" with a rationale explaining what more is needed — do not force accept/reject prematurely.

Minimal worked example

{
  "id": "ADJ-0012",
  "stage": 2,
  "k": 3,
  "framings": ["by-routes", "by-tests", "by-call-sites"],
  "candidates": [
    { "candidate_id": "CTX-Ordering", "found_by": ["by-routes", "by-tests"], "partition": "some", "decision": "accept", "rationale": "linguistic boundary holds; coupling data supports the split" }
  ],
  "kind": "adjudication",
  "decided_by": "Jane Doe",
  "decided_at": "2026-08-01T15:00:00Z",
  "agreement_rate": 0.82
}