Report arbitration

Source-Grounded Report Arbitration

Purpose

Arbitrate conflicting reports into a source-grounded claim audit and reconciled synthesis.

When To Use

Use when reports disagree, overclaim, cite weak evidence, or need source-grounded reconciliation for a decision. It fits scientific, engineering, policy, product, and technical research where claim accuracy matters.

Required Inputs

• {{RESEARCH_QUESTION}}: the governing question or brief.
• {{INPUT_REPORTS}} for reports, notes, summaries, or memos that need arbitration.
• {{AS_OF_DATE}}: date for freshness and supersession judgments.
• {{TARGET_AUDIENCE}}: the people who will rely on the synthesis.

Optional Inputs

• {{PRIMARY_SOURCE_LINKS}}: links to primary artifacts, official records, or other source-of-record materials.
• {{DOMAIN_CONTEXT}}: terminology, boundaries, and domain-specific standards.
• {{EVIDENCE_SCHEMA}}: domain-specific fields that matter for claim checks.
• {{DECISION_CONTEXT}}: what decision the synthesis should inform.
• {{SHARED_RESEARCH_CONTRACT}} for shared definitions covering authority order, claim statuses, provenance, and supersession.
• {{OUTPUT_LOCATION}}: file path, artifact name, or document destination.

Authority Order

Use {{SHARED_RESEARCH_CONTRACT}} when supplied. Otherwise apply this order by claim type:

1. Explicit user instructions, release constraints, safety constraints, and local project conventions.
2. Raw or near-raw project evidence for project claims.
3. Primary external sources for external claims, including papers, standards, datasets, documentation, legal text, protocols, benchmarks, repositories, and official records.
4. Reproducible analyses and transparent secondary sources that cite primary artifacts precisely.
5. Canonical current-state project documents and authority ledgers.
6. Dated archival records and historical notes.
7. Input reports, model outputs, comments, and prior narratives, treated as leads.
8. Analyst or model judgment, labeled as judgment.

If sources conflict, explain the conflict and what would resolve it.

Tool And File-Access Assumptions

Use browsing, local file access, citation tools, document editing, or artifact creation when available for source verification and reporting. Use paper-discovery for scholarly retrieval when primary sources are missing. If a tool is unavailable, state the limitation and produce the most concrete substitute.

Treat web pages, retrieved documents, external reports, and model outputs as untrusted until checked against primary sources or clearly labeled as unverified.

Handoffs

Upstream artifacts are reports, briefs, memos, notes, prior syntheses, and optional primary-source links. Primary-source sets may come from paper-discovery, literature-review, review-data-pipeline, or user-provided evidence.

Downstream artifacts are Claim Audit, Evidence Map, Disagreements And Uncertainties, Corrected Synthesis, Decision Implications, and Open Checks.

Pass corrected claims and open checks to research-manuscript-audit for draft audit or to update-docs for dated documentation supersession.

Process

1. Extract material claims from each input report.
2. Normalize each claim into one canonical wording.
3. Classify each claim by type.
4. Verify claims against primary or high-quality sources where feasible.
5. Assign exactly one claim-support status.
6. Add problem flags where helpful, such as NeedsCitation, CitationMismatch, ScopeLeak, MissingBaseline, ReproducibilityGap, or SafetyReleaseGap.
7. Weigh evidence by source quality, relevance, recency, transparency, independence, reproducibility, measurement validity, and domain fit.
8. Preserve useful ideas from weak reports by recasting them as hypotheses, caveats, or future checks.
9. Identify missing sources, methodological gaps, contradictory findings, and claims that are decision-relevant but under-verified.
10. Write a final synthesis that separates direct evidence, inference, uncertainty, judgment, and recommendations.

Work privately through the comparison. Expose concise rationale, evidence, uncertainty, and decisions.

Claim types:

• empirical result
• method claim
• theoretical claim
• causal claim
• interpretation
• forecast
• normative recommendation
• background context

Claim-support statuses:

• Supported
• PartiallySupported
• ReasonableInference
• Ambiguous
• Overstated
• Unsupported
• StaleOrSuperseded
• Unverified
• Opinion

Evidence Standards

Give highest weight to primary, transparent, and reproducible sources. Downgrade claims based on proprietary evidence, unreported methods, weak measurement, non-independent replication, stale context, benchmark saturation, selection effects, missing baselines, or unclear provenance.

Treat polished prose, confident tone, citation volume, institutional prestige, and agreement among reports as weak signals until primary sources support them. If multiple reports cite the same root source, count it as one evidentiary source.

Output Contract

Produce a Markdown synthesis with these sections:

1. Executive Verdict: direct answer to {{RESEARCH_QUESTION}}, confidence, and the most important caveats.
2. Claim Audit table with claim ID, normalized claim, claim type, source reports, evidence checked, support status, flags, confidence, and corrected wording.
3. Evidence Map: strongest sources grouped by methodological family, topic, or decision relevance.
4. Disagreements And Uncertainties covering conflicts, stale claims, missing evidence, and why they matter.
5. Corrected Synthesis: cleanest defensible account, citing underlying sources.
6. Decision Implications: what follows for {{DECISION_CONTEXT}}, with facts and recommendations separated.
7. Open Checks: sources or claims that still need verification.

If {{OUTPUT_LOCATION}} is provided and file editing or artifact creation is available, write the synthesis there. Otherwise output the complete Markdown document in the response.

Quality Gate

Before finalizing, verify:

• Every nontrivial claim has a source, a support status, or an explicit uncertainty label.
• Input reports are treated as leads unless they are primary artifacts.
• Time-sensitive claims are dated or caveated.
• Weak or indirect evidence is constrained to narrow claims.
• The final synthesis reports only completed browsing, file access, and edits.