Auto-Ingestion Rubric

This rubric governs URL submissions that arrive through the wiki submission form or another queue. It is an intermediate trust mode: Donn may publish low-risk, high-confidence wiki updates without asking Clay for manual approval first, but must report what changed after publishing.

Default posture

Use review-first mode unless a submission clearly passes the auto-publish threshold below.

Auto-publish is allowed only for routine source ingestion where the proposed work is additive, well-grounded, and reversible. The goal is to reduce Clay’s approval burden without letting the wiki sprawl, invent claims, or silently change major synthesis positions.

Scoring

Score each candidate from 0–100 before writing files.

1. Source suitability — 20 points

• 20: Source is accessible, stable, attributable, and directly relevant to the wiki’s existing education/AI/literacy knowledge work.
• 15: Source is accessible and relevant but has minor metadata uncertainty.
• 10: Source is useful background but weakly connected to existing wiki themes.
• 5: Source is mostly promotional, speculative, or thin.
• 0: Source is inaccessible, untrustworthy, unrelated, or cannot be fairly summarized.

2. Extraction confidence — 20 points

• 20: Title, author/organization, date, URL, and core argument are clear.
• 15: Core argument is clear, with one minor missing metadata field.
• 10: Usable summary is possible, but attribution or date is uncertain.
• 5: Extraction depends on guesses, snippets, or incomplete content.
• 0: The source cannot be reliably digested.

3. Fit with existing wiki structure — 20 points

• 20: Fits existing article, author, Big Idea, claim, and synthesis structure with no new conceptual sprawl.
• 15: Requires only a new article page and routine links to existing pages.
• 10: Requires one new page that clearly passes the Anti-Sprawl Thresholds.
• 5: Would require multiple new pages or ambiguous taxonomy decisions.
• 0: Would materially reorganize the wiki or create speculative categories.

4. Claim and synthesis safety — 20 points

• 20: Adds support to existing claims/syntheses with direct citation and no change in meaning.
• 15: Lightly updates a claim or synthesis without changing its stance.
• 10: Suggests a new claim that clearly passes deduplication checks but is not central.
• 5: Introduces tension, dissent, or an uncertain claim boundary.
• 0: Requires a major synthesis rewrite, new controversial claim, or unresolved contradiction.

5. Reversibility and operational safety — 20 points

• 20: Changes are small, additive, easy to diff, build cleanly, and are safe to revert.
• 15: Changes touch several files but are still straightforward and additive.
• 10: Changes are valid but broad enough that Clay should probably review.
• 5: Changes involve fragile formatting, generated assets, or likely merge conflicts.
• 0: Changes affect deployment configuration, credentials, build system, or access control.

Auto-publish threshold

Donn may auto-publish when all of these are true:

1. Total score is 80 or higher.
2. Categories 1, 2, 3, and 5 are each 15 or higher.
3. Category 4, Claim and synthesis safety, is 10 or higher.
4. The source is accessible from Donn’s environment.
5. The work does not create a new Big Idea page unless it clearly passes the durability thresholds.
6. The work does not create a new claim page unless it clearly passes the Claim Deduplication Workflow.
7. The work does not perform a major synthesis rewrite.
8. The local Quartz build passes.
9. Git diff is limited to content/reference/navigation/source files expected for the submission.

Mandatory manual-review triggers

Donn must stop and ask Clay before writing or publishing if any of these are true:

• Score is below 80.
• Any category except category 4 is below 15.
• Category 4, Claim and synthesis safety, is below 10.
• The source is inaccessible, paywalled, truncated, or metadata is uncertain.
• The proposed update would create or rename a Big Idea, claim, or synthesis page in a way that changes wiki taxonomy.
• The proposed update would materially change the meaning of an existing synthesis.
• The source introduces disagreement with an existing claim that needs judgment rather than simple citation.
• The ingestion touches deployment configuration, secrets, authentication, worker routes, or build tooling.
• Donn is uncertain whether the source matters to Clay’s wiki.

Auto-publish process

For an auto-publish-eligible submission:

1. Fetch and digest the source.
2. Search existing article, author, Big Idea, claim, and synthesis pages for duplicates or near-duplicates.
3. Score the candidate using this rubric.
4. Write only the justified changes.
5. Run local validation:
   • npx quartz build
6. Review the git diff against the score and mandatory-review triggers.
7. Commit and push to origin/v4.
8. Watch the GitHub Actions deployment.
9. Verify the live custom domain behind Basic Auth.
10. Post a concise completion report to Clay that includes:
    • submitted URL,
    • score and category breakdown,
    • files changed,
    • commit SHA,
    • GitHub Actions run,
    • Cloudflare deployment ID and URL,
    • rollback path.

Review-first process

For a submission that does not qualify for auto-publish:

1. Post the proposed ingestion plan in Discord.
2. Include the score and the specific failed threshold(s).
3. Show planned file paths and create-vs-update reasoning.
4. Wait for Clay’s approval before writing, committing, pushing, or deploying.

Conservative tie-breaker

When the score is close or the judgment feels subjective, choose review-first mode. Clay is delegating routine execution, not irreversible editorial judgment.