{"id":"content-credibility-engine","relativePath":"content-credibility-engine.md","title":"Content Credibility Engine","markdown":"# Content Credibility Engine\n\nMeta Museum treats content generation and syndication as a credibility system, not a volume system.\n\nThis document defines four operational layers:\n\n1. Trust layer: provenance, authorship, traceability, rights.\n2. Originality layer: semantic novelty and source-grounded synthesis.\n3. Distribution layer: channel orchestration, SEO metadata, syndication.\n4. Consistency layer: cadence, voice control, and review discipline.\n\n## Scope and compatibility\n\nThis policy extends existing repository rules in:\n\n- `CLAUDE.md`\n- `docs/roadmap.md`\n- `docs/linked-art/LinkedArtModel1.0-Reference.md`\n\nIf any conflict appears, Linked Art fidelity, rights safety, and \"cite or refuse\" remain non-negotiable.\n\n## 1) Trust layer requirements\n\nTrust metadata is persisted in:\n\n- `provenance/ledger.json`\n- `provenance/source-map.yaml`\n\nRequired metadata per publishable artifact:\n\n- `artifactId` stable identifier\n- `contentHashSha256` over canonical source bundle\n- `author` and `reviewers`\n- `createdAt` and `publishedAt`\n- source list with URL, provider, retrieval timestamp\n- citation coverage evidence\n- rights/reuse disposition\n\nMinimum enforcement:\n\n- No publish without at least one source reference.\n- No publish with unresolved rights status.\n- No publish if citation coverage fails policy threshold.\n\n## 2) Originality layer requirements\n\nOriginality records are persisted in:\n\n- `semantic-core/originality-index.json`\n\nBaseline novelty policy:\n\n- compute embedding-space novelty against recent published corpus\n- classify as \"meaningfully original\" when cosine distance is above `0.18`\n- require one \"unique insight\" note per artifact (human-written or curator-approved)\n\nGuardrails:\n\n- template-only outputs are rejected\n- uncited paraphrase outputs are rejected\n- all generated claims must map to source evidence or be removed\n\n## 3) Distribution layer requirements\n\nDistribution control artifacts:\n\n- `distribution/schedule.yaml`\n- runtime queue database at `distribution/queue.db` (gitignored)\n\nChannel set (initial):\n\n- Web\n- LinkedIn\n- Medium\n- Email\n- API feed/syndication\n\nRequired pipeline behavior:\n\n- per-channel publish caps and minimum cadence checks\n- snippet extraction and summary variants\n- SEO metadata generation (title, description, canonical URL, tags)\n- rights-safe media checks before enqueue\n- queue-worker orchestration across `web`, `linkedin`, `medium`, `email`, and `api` channels\n\n## 4) Consistency layer requirements\n\nVoice and cadence controls:\n\n- `generation/style-profile.md`\n- `distribution/schedule.yaml`\n\nBaseline policy:\n\n- minimum 2 outputs per week across channels\n- maximum 1 output per day per channel by default\n- style profile versioned with last-updated reviewer\n\n## 5) Monitoring and audits\n\nMonitoring baseline persisted in:\n\n- `monitoring/metrics.json`\n\nTracked dimensions:\n\n- trust score\n- originality score\n- distribution reach\n- engagement velocity\n\nWeekly audit checks:\n\n- citation/link integrity\n- novelty drift\n- rights/compliance regressions\n- broken outbound links\n- engagement velocity minimum threshold\n- trust/originality score regression thresholds vs configured baseline\n\nAutomation:\n\n- Local run: `pnpm credibility:audit`\n- Gate run (non-zero on failure): `pnpm credibility:audit:check`\n- Weekly GitHub Action: `.github/workflows/credibility-audit.yml`\n  - schedule: every Monday\n  - checks: validation drift, eval relevance threshold, markdown broken-link scan, engagement/trust/originality eval-threshold alerts\n  - artifact: `artifacts/credibility-audit/latest.json`\n\n## 6) Implementation notes for this repository\n\n- OpenTelemetry is the default tracing substrate for this workflow.\n- Postgres is the system of record where queue/provenance services graduate from file-backed prototypes.\n- Redis remains the short-lived cache layer for scoring and queue coordination.\n- FastAPI service extensions are allowed where they fit current C2/C4 service boundaries.\n- Additional AI framework dependencies (for example LangChain/LlamaIndex) are optional and must pass the standard \"ask before acting\" dependency gate in `CLAUDE.md`.\n\n## 7) Definition of done for credibility-ready content features\n\n1. A test exists that fails when citations or rights metadata are missing.\n2. A test exists that fails when originality metadata is absent for generated artifacts.\n3. Queue/schedule policy is enforced by code, not just documentation.\n4. Metrics are emitted and trace-linked for generation, review, and publish operations.\n5. Human approval is explicit before publication actions.\n","sections":[{"level":2,"heading":"Scope and compatibility","anchor":"scope-and-compatibility"},{"level":2,"heading":"1) Trust layer requirements","anchor":"1-trust-layer-requirements"},{"level":2,"heading":"2) Originality layer requirements","anchor":"2-originality-layer-requirements"},{"level":2,"heading":"3) Distribution layer requirements","anchor":"3-distribution-layer-requirements"},{"level":2,"heading":"4) Consistency layer requirements","anchor":"4-consistency-layer-requirements"},{"level":2,"heading":"5) Monitoring and audits","anchor":"5-monitoring-and-audits"},{"level":2,"heading":"6) Implementation notes for this repository","anchor":"6-implementation-notes-for-this-repository"},{"level":2,"heading":"7) Definition of done for credibility-ready content features","anchor":"7-definition-of-done-for-credibility-ready-content-features"}],"html":"<h1 id=\"content-credibility-engine\">Content Credibility Engine</h1>\n<p>Meta Museum treats content generation and syndication as a credibility system, not a volume system.</p>\n<p>This document defines four operational layers:</p>\n<ol><li>Trust layer: provenance, authorship, traceability, rights.</li></ol>\n<ol><li>Originality layer: semantic novelty and source-grounded synthesis.</li></ol>\n<ol><li>Distribution layer: channel orchestration, SEO metadata, syndication.</li></ol>\n<ol><li>Consistency layer: cadence, voice control, and review discipline.</li></ol>\n<h2 id=\"scope-and-compatibility\">Scope and compatibility</h2>\n<p>This policy extends existing repository rules in:</p>\n<ul><li>`CLAUDE.md`</li><li>`docs/roadmap.md`</li><li>`docs/linked-art/LinkedArtModel1.0-Reference.md`</li></ul>\n<p>If any conflict appears, Linked Art fidelity, rights safety, and &quot;cite or refuse&quot; remain non-negotiable.</p>\n<h2 id=\"1-trust-layer-requirements\">1) Trust layer requirements</h2>\n<p>Trust metadata is persisted in:</p>\n<ul><li>`provenance/ledger.json`</li><li>`provenance/source-map.yaml`</li></ul>\n<p>Required metadata per publishable artifact:</p>\n<ul><li>`artifactId` stable identifier</li><li>`contentHashSha256` over canonical source bundle</li><li>`author` and `reviewers`</li><li>`createdAt` and `publishedAt`</li><li>source list with URL, provider, retrieval timestamp</li><li>citation coverage evidence</li><li>rights/reuse disposition</li></ul>\n<p>Minimum enforcement:</p>\n<ul><li>No publish without at least one source reference.</li><li>No publish with unresolved rights status.</li><li>No publish if citation coverage fails policy threshold.</li></ul>\n<h2 id=\"2-originality-layer-requirements\">2) Originality layer requirements</h2>\n<p>Originality records are persisted in:</p>\n<ul><li>`semantic-core/originality-index.json`</li></ul>\n<p>Baseline novelty policy:</p>\n<ul><li>compute embedding-space novelty against recent published corpus</li><li>classify as &quot;meaningfully original&quot; when cosine distance is above `0.18`</li><li>require one &quot;unique insight&quot; note per artifact (human-written or curator-approved)</li></ul>\n<p>Guardrails:</p>\n<ul><li>template-only outputs are rejected</li><li>uncited paraphrase outputs are rejected</li><li>all generated claims must map to source evidence or be removed</li></ul>\n<h2 id=\"3-distribution-layer-requirements\">3) Distribution layer requirements</h2>\n<p>Distribution control artifacts:</p>\n<ul><li>`distribution/schedule.yaml`</li><li>runtime queue database at `distribution/queue.db` (gitignored)</li></ul>\n<p>Channel set (initial):</p>\n<ul><li>Web</li><li>LinkedIn</li><li>Medium</li><li>Email</li><li>API feed/syndication</li></ul>\n<p>Required pipeline behavior:</p>\n<ul><li>per-channel publish caps and minimum cadence checks</li><li>snippet extraction and summary variants</li><li>SEO metadata generation (title, description, canonical URL, tags)</li><li>rights-safe media checks before enqueue</li><li>queue-worker orchestration across `web`, `linkedin`, `medium`, `email`, and `api` channels</li></ul>\n<h2 id=\"4-consistency-layer-requirements\">4) Consistency layer requirements</h2>\n<p>Voice and cadence controls:</p>\n<ul><li>`generation/style-profile.md`</li><li>`distribution/schedule.yaml`</li></ul>\n<p>Baseline policy:</p>\n<ul><li>minimum 2 outputs per week across channels</li><li>maximum 1 output per day per channel by default</li><li>style profile versioned with last-updated reviewer</li></ul>\n<h2 id=\"5-monitoring-and-audits\">5) Monitoring and audits</h2>\n<p>Monitoring baseline persisted in:</p>\n<ul><li>`monitoring/metrics.json`</li></ul>\n<p>Tracked dimensions:</p>\n<ul><li>trust score</li><li>originality score</li><li>distribution reach</li><li>engagement velocity</li></ul>\n<p>Weekly audit checks:</p>\n<ul><li>citation/link integrity</li><li>novelty drift</li><li>rights/compliance regressions</li><li>broken outbound links</li><li>engagement velocity minimum threshold</li><li>trust/originality score regression thresholds vs configured baseline</li></ul>\n<p>Automation:</p>\n<ul><li>Local run: `pnpm credibility:audit`</li><li>Gate run (non-zero on failure): `pnpm credibility:audit:check`</li><li>Weekly GitHub Action: `.github/workflows/credibility-audit.yml`</li><li>schedule: every Monday</li><li>checks: validation drift, eval relevance threshold, markdown broken-link scan, engagement/trust/originality eval-threshold alerts</li><li>artifact: `artifacts/credibility-audit/latest.json`</li></ul>\n<h2 id=\"6-implementation-notes-for-this-repository\">6) Implementation notes for this repository</h2>\n<ul><li>OpenTelemetry is the default tracing substrate for this workflow.</li><li>Postgres is the system of record where queue/provenance services graduate from file-backed prototypes.</li><li>Redis remains the short-lived cache layer for scoring and queue coordination.</li><li>FastAPI service extensions are allowed where they fit current C2/C4 service boundaries.</li><li>Additional AI framework dependencies (for example LangChain/LlamaIndex) are optional and must pass the standard &quot;ask before acting&quot; dependency gate in `CLAUDE.md`.</li></ul>\n<h2 id=\"7-definition-of-done-for-credibility-ready-content-features\">7) Definition of done for credibility-ready content features</h2>\n<ol><li>A test exists that fails when citations or rights metadata are missing.</li></ol>\n<ol><li>A test exists that fails when originality metadata is absent for generated artifacts.</li></ol>\n<ol><li>Queue/schedule policy is enforced by code, not just documentation.</li></ol>\n<ol><li>Metrics are emitted and trace-linked for generation, review, and publish operations.</li></ol>\n<ol><li>Human approval is explicit before publication actions.</li></ol>","updatedAt":"2018-10-20T01:46:40.000Z","checksum":"d9aa323586701afb7fcd729d276b832b7f4b17e3a81bf441938bca7888c6a843","checksumPrefix":"d9aa32358670","anchorCount":8,"lineCount":145,"rawUrl":"/api/docs/content?path=content-credibility-engine.md","htmlUrl":"/docs?doc=content-credibility-engine.md","apiUrl":"/api/docs/content?path=content-credibility-engine.md"}