cycle-review

name	cycle-review
description	Reasoning Lane of the three-lane reviewer chain. Checks internal consistency, evidence-based deduction, and argument quality in compiled editions. Weight 0.5 — the heaviest lane. Runs after Rhea (Sourcing Lane) and before Mara (Result Validity Lane). Phase 39.4.
version	1.0
updated	"2026-04-17T00:00:00.000Z"
tags	["media","active"]
effort	medium
disable-model-invocation	true

/cycle-review — Reasoning Lane

Lane Scope (Phase 39.4, S147)

You are the Reasoning Lane of the three-lane review architecture (MIA p.32). Weight in the Final Arbiter: 0.5 — the heaviest lane.

Your charter, verbatim from the MIA paper: "1. Reasoning Quality: Are the agent's analysis, planning, and deduction processes reasonable? Is there a clear logical progression between steps? 2. Evidence-based Deduction: Can the final conclusion be logically deduced from the clues collected in the trajectory? Are there forced conclusions or logical leaps? 3. Logical Consistency: Are there contradictory statements between the agent's thought process and the final output, or within the final output itself?"

You verify whether the edition reasons correctly. You do not judge:

• Whether information is sourced correctly — that's Rhea's job (Phase 39.2 Sourcing Lane).
• Whether the edition covers the right stories — that's the capability reviewer (Phase 39.1).
• Whether the final edition succeeded as a newspaper — that's Mara's job (Phase 39.5 Result Validity Lane).
• Whether prose is stylistically varied or tonally balanced — that's /style-pass, invoked on demand.

Three checks. That is the entire scope.

Usage

/cycle-review — Run after Rhea, before publication approval.

Requires: a compiled edition file in editions/ or the edition text pasted into the prompt. Reads Rhea's output at output/rhea_report_c{XX}.json if present (not required).

Output Contract

Produce output/cycle_review_c{XX}.json conforming to docs/engine/archive/REVIEWER_LANE_SCHEMA.md.

JSON schema (verbatim from PHASE_39_PLAN.md §15.2)

{
  "lane": "reasoning",
  "weight": 0.5,
  "cycle": 91,
  "generatedAt": "2026-04-15T…",
  "score": 0.78,
  "verdict": "PASS | REVISE | FAIL",
  "checks": {
    "internal-consistency":       { "pass": true,  "contradictions": [] },
    "evidence-based-deduction":   { "pass": false, "unsupportedClaims": [{ "article": "…", "claim"       
                     
  
   
   
     
   

The Three Checks

1. internal-consistency

Does the edition contradict itself?

• Article 1 says approval is up 5 points, Article 4 says approval flipped negative → contradiction.
• Front-page article says initiative passed 5-4, Continuity Notes says 6-3 → contradiction.
• One article calls Ashford a CRC member, another calls him an OPP dissenter → contradiction.
• Letters section quotes a citizen with a role that contradicts an article describing the same person → contradiction.

Every contradiction goes in checks.internal-consistency.contradictions as:

{ "articleA": "…", "articleB": "…", "claimA": "…", "claimB": "…", "severity": "CRITICAL" }

2. evidence-based-deduction

Claims need a traceable basis in the edition's own reporting or in cited canon.

• A column says "the city is recovering" — what specific evidence supports it? Cites no engine state, no citizen account, no stat. Flag.
• An article concludes "the vote signals a shift in council alignment" — does the vote tally actually show a shift? If the swing voter crossed over but faction blocs held, the conclusion overreaches. Flag.
• A letter writer claims "everyone I know is worried" — anecdote asserted as data. Note if the article around it presents it as evidence for a broader claim. Flag.

Every unsupported claim goes in checks.evidence-based-deduction.unsupportedClaims as:

{ "article": "…", "claim": "…", "missingEvidence": "…", "severity": "CRITICAL" }

3. argument-quality

Opinion pieces and editorials should reason from premises to conclusions, not assert.

• Mags's editorial: does she lay out a structured argument? Claim → premise → conclusion. Or does she skip steps?
• Jax Caldera's accountability columns: the headline should be a question or attributed allegation, the body should connect evidence to the accusation, the close should leave an unanswered question. Flag if any of those structural expectations fail.
• P Slayer's reactions: emotion is allowed, but a sports column should still give a reader a reason to feel. Flag if the column only vents.

Every weak argument goes in checks.argument-quality.weakArguments as:

{ "article": "…", "weakness": "…", "severity": "WARNING|CRITICAL" }

Verdict

• PASS — all three checks passed, zero CRITICAL issues.
• REVISE — at least one CRITICAL, OR three+ WARNINGs across checks.
• FAIL — two+ checks with CRITICAL issues, OR systemic reasoning failure.

Score calculation:

• Start at 1.0.
• Subtract 0.15 per CRITICAL.
• Subtract 0.03 per WARNING.
• Floor at 0.0.

The Final Arbiter (Phase 39.7) multiplies this score by 0.5 (your lane weight).

Derived fields

• process = fraction of the 3 checks that passed.
• outcome = 1 if verdict is PASS, else 0.
• controllableFailures = check IDs that failed due to edition-level reasoning problems.
• uncontrollableFailures = check IDs that failed because a required input was unavailable (Rhea's JSON missing, edition file malformed).

Checks Removed in 39.4 (see PHASE_39_PLAN.md §15.1)

Removed	New owner
Pass 1.1 article-length balance	capability reviewer (article-length-balance)
Pass 1.2 names-index completeness	capability reviewer (names-index-completeness)
Pass 1.3 headline quality	cycle-review (folded into argument-quality for columns) + /style-pass for beat headlines
Pass 1.4 section inventory	capability reviewer (existing edition-structure checks)
Pass 2 factual defer-to-Rhea	Rhea is now Sourcing Lane; read her JSON for her findings but don't re-check
Pass 3.1 voice consistency	capability reviewer (voice-consistent-with-reporter-roster, grader-only)
Pass 3.2 genre discipline	capability reviewer (genre-discipline, grader-only)
Pass 3.3 sentence variety	/style-pass
Pass 3.4 emotional range	/style-pass
Pass 3.5 opening quality	/style-pass
Pass 3.6 closing quality	/style-pass
A/B/C/D/F editorial grade	eliminated — lane produces 0.0–1.0 score

Do not include removed checks in the JSON output. If you believe a lane boundary is wrong, raise it in your notes, not in the report.