Conflict Detector

Conflict Detector

Use this skill during Step 6 when conflicts appear.

Conflict Types

1. Definitional conflict
2. Scope conflict
3. Obligation conflict
4. Enforcement/sanction conflict
5. Recency conflict

Output

Use references/conflict-report-template.md.

Resolution priority:

1. Legal hierarchy
2. Jurisdictional relevance
3. Recency
4. Original-text support

Before marking a conflict resolved or unresolved, apply the deterministic helper in citation_auditor.conflict_resolution:

from citation_auditor.conflict_resolution import resolve_conflict

resolution = resolve_conflict(claim_a, claim_b, sources, target_jurisdictions, profile="general")

• If resolution.outcome == "resolved", use preferred_source_id in the conflict report and include the helper's top rationale in provisional_resolution.
• If resolution.outcome == "unresolved", keep the conflict tagged [Unresolved Conflict] and record the score delta/rationale in residual_uncertainty.
• Do not override the helper with narrative intuition unless a jurisdiction-specific legal rule is cited from a primary source; if that happens, record the rule and source id in the conflict report.

Use a matter-specific profile when the conflict fits one of these patterns:

If none fits cleanly, keep profile="general" and mark close-score results [Unresolved Conflict].

Severity Handling

• High severity (obligation/sanction): attempt source expansion first.
• Low severity (definition/recency): mark [Unresolved Conflict] when unresolved.

Persistence — output/unresolved-conflicts.json

When a conflict is marked [Unresolved Conflict] (i.e., it cannot be resolved after applying the resolution-priority rules and any source-expansion attempt), write a record to output/unresolved-conflicts.json.

Entry Schema

{
  "conflict_id": "CF-001",
  "first_detected": "2026-03-05",
  "last_seen": "2026-03-05",
  "times_encountered": 1,
  "type": "obligation",
  "severity": "high",
  "jurisdictions_involved": ["NL", "AU"],
  "domains_involved": [1],
  "description": "Conflicting obligations on loot-box disclosure between NL KSA ruling and AU ACCC guidance",
  "claim_a": {
    "summary": "NL: loot boxes are games of chance requiring full odds disclosure",
    "source_id": "NL-A1",
    "citation": "[A1] KSA, Decision 2024-07-01, p. 12",
    "grade": "A"
  },
  "claim_b": {
    "summary": "AU: loot boxes are not gambling under Qld Racing Act s. 4",
    "source_id": "AU-P2",
    "citation": "[P2] Queensland Racing Act 2002, s. 4",
    "grade": "A"
  },
  "resolution_attempted": true,
  "resolution_outcome": "unresolved",
  "residual_uncertainty": "Whether AU federal law may impose separate disclosure requirements",
  "next_steps": "Check Australian Consumer Law Schedule 2, Competition and Consumer Act 2010",
  "status": "open"
}

Write Rules

1. Before writing, check if an entry with the same jurisdictions_involved + domains_involved

   • type already exists in the file.
   • Duplicate found: update last_seen and increment times_encountered. Do not create a new entry.
   • No duplicate: append a new entry.

2. Set status to "open" when first written. Change to "resolved" only when a subsequent research session provides a definitive primary-source resolution.

3. Update output/unresolved-conflicts.json immediately when the conflict is detected — do not wait until Step 9.

Read Rules (at session start)

When the main agent loads this file at session start:

• Surface any "open" entries whose jurisdictions_involved and domains_involved overlap with the current query's scope.
• Prepend a brief note in the Step 6 analysis output: "Note: {N} pre-existing unresolved conflict(s) from prior sessions are relevant to this query. See conflict IDs: {list}. These are included in the analysis below."