enforce-npov

# Enforce NPOV ## Overview Use this skill to apply the repository's editorial constitution before any durable follow-up advances. In MVP it is a doc-only workflow: assess due weight, attribution, unsupported inference, and contradiction handling, then route the result to the correct policy or escalation lane. ## Classification - **Mode:** Doc-only workflow - **MVP status:** Active - **Execution boundary:** Policy review and handoff only. Do not silently rewrite contested content, approve original research, or bypass explicit escalation. ## When to Use - A draft or proposed update needs neutral-point-of-view review - Evidence could be overclaimed, under-attributed, or framed with the wrong due weight - A change introduces unsupported inference or original research risk - Conflicting sources or existing pages require an explicit policy outcome - Governance must decide whether content can proceed, needs revision, or must be escalated ## Contract - Input: evidence-backed draft text, synthesis notes, or a proposed wiki plan - Decision model: classify the outcome as `allow`, `revise`, or `escalate` according to NPOV and contradiction rules - Output: a policy finding with due-weight notes, attribution gaps, contradiction handling, and next action - Handoff rule: revision requests return to the governed upstream lane; escalations produce open questions or conflict records instead of forced merges ## Assertions - Claims must remain attributable to cited evidence with appropriate due weight - Unsupported synthesis or original research is blocking - Contradictions trigger explicit handling rather than silent overwrite behavior - The skill never authorizes direct wiki writes on its own - Policy ambiguity routes to escalation rather than guesswork ## Procedure ### Step 1: Review the evidence-backed claim set Read the proposed draft, synthesis, or plan alongside the cited evidence and any relevant existing pages or summaries. ### Step 2: Check neutral presentation Evaluate whether the framing, emphasis, and attribution reflect the available evidence instead of the editor's preferred narrative. ### Step 3: Check for unsupported inference Look for synthesis leaps, omitted attribution, or claims that exceed what the cited evidence can support. ### Step 4: Classify the policy outcome Use one of three outcomes only: - **allow**: policy concerns are satisfied for the next governed step - **revise**: attribution or weighting gaps must be corrected upstream - **escalate**: contradiction or ambiguity requires explicit arbitration ### Step 5: Route the follow-up Return revision notes, open questions, or conflict-log recommendations without forcing a merge or publication path open. ## Boundaries - Do not treat stylistic preference as a policy violation without evidence - Do not resolve contradictory sources by deleting one side silently - Do not approve content that introduces unsupported original research - Do not bypass policy logging or open-question follow-up when ambiguity remains ## Verification - [ ] Due weight and attribution are explicitly reviewed - [ ] Unsupported inference is separated from evidence-backed synthesis - [ ] Contradictions produce a clear revise or escalate outcome - [ ] The result does not authorize direct wiki writes by itself - [ ] Remaining ambiguity is preserved for governance follow-up ## References - [`AGENTS.md`](../../../AGENTS.md) - [`docs/architecture.md`](../../../docs/architecture.md) - [`docs/ideas/wiki-curation-agent-framework.md`](../../../docs/ideas/wiki-curation-agent-framework.md) - [`raw/processed/SPEC.md`](../../../raw/processed/SPEC.md) - [`schema/page-template.md`](../../../schema/page-template.md) - [Manual of Style](../references/manual-of-style.md) — prose register, heading conventions, vocabulary standards