Source: https://guthlabs.ai/docs/tools/citation-checks/qualification/ tool / candidate # Citation checks — qualification The distinct qualification record for Citation checks. This page is proposal-only. Status: **proposed**Lifecycle: **candidate** ## Record status **Record ID**: tool:citation-checks **Parent lifecycle**: candidate **Page lane**: docs_qualification **Lane lifecycle**: proposed **Owner**: Tilly **Owner role**: Pointguard and independent reviewer **Owner status**: Not separately recorded **Qualification result**: NOT RUN · held for evidence and human review **Working browser surface:** [Tools — overview]() — Inspect the tool contract, usage, boundaries, and proposed qualification checks. This route is local-only; provider, credential, database, customer, publication, deployment, and production effects remain held. ## Qualification status **Result**: NOT RUN **Disposition**: HELD **Reviewer**: Not recorded **Receipt**: Not recorded This page defines proposed acceptance criteria for the docs_qualification lane. It does not execute tests, establish compliance, qualify the subject, or authorize an effect. ### Lane-specific instrument aids [checksum-verifieroptional · Proposal-only integrity check for recorded citation artifacts or exported evidence digests; does not establish source authority or citation correctness.]()[binary-text-diff-similarity-ratiooptional · Proposal-only comparison of claim and evidence text to identify changes or overlap; does not establish truth, provenance, or authority.]()[url-query-string-parseroptional · Proposal-only parsing of supplied source URLs into components; does not verify reachability, ownership, or source legitimacy.]() ## Outcome definitions **Verified**: All stated evidence and acceptance requirements passed for the exact reviewed revision; this still does not authorize deployment or an external effect. **Failed**: One or more deterministic acceptance requirements failed with an inspectable reason. **Needs review**: Evidence or interpretation is incomplete, conflicting, or awaits an identified human reviewer. **Held**: Authority, required evidence, safe execution conditions, or a principal gate is absent. The current record remains NOT RUN / HELD. These definitions describe candidate outcomes only. ## Catalog-observed source metadata These fields are reproduced from the admitted candidate record for source-fidelity inspection. Presence does not prove implementation, connection, qualification, authority, or an external effect. **tool_id**: tool:citation-checks **slug**: citation-checks **title**: Citation checks **owner_agents**: Tilly **owner_roles**: Pointguard and independent reviewer **related_jobs**: Independent review; Instrument qualification; Evidence reconciliation; Grant research and drafts; Website claim review **related_field_guides**: Instrument packets; Qualification records; Grant dossiers; Public-claim routes **principal_gates**: Final publication; Funding commitments; Credential access; Changes to another agent's source **never_rules**: Approve her own evidence; Hide a contradiction; Turn a draft into a public claim **lifecycle**: candidate **source_refs**: agents/index.html#tilly; tool:citation-checks source record supplied in request ## Proposed deterministic qualification tests - Overview page states proposal-only status and distinguishes catalog observation from proven implementation. - Source-record page reproduces the supplied tool ID, title, owner, roles, related jobs, field guides, gates, never-rules, and source reference without embellishment. - Qualification page refuses a pass when implementation, connection, deployment, or authority evidence is absent. - Evidence-reconciliation page requires one record per claim, exact citation text or locator, reviewer identity, and unresolved-gap state. - Contradiction test: conflicting evidence produces a visible contradiction record and a held decision, never a silent merge. - Self-approval test: a reviewer matching the evidence owner cannot approve the evidence. - Draft/public-claim test: draft material remains non-public until the final-publication gate has an independent approval record. - Gate test: funding commitments and credential access remain held unless their corresponding principal-gate record is present. - Accessibility test: every route has a unique title, one logical heading structure, keyboard-operable controls, visible focus, text alternatives for non-text status indicators, sufficient contrast, and status updates available to assistive technology. - Accessibility failure test: an inaccessible required control causes qualification failure rather than being bypassed. - Privacy test: routes do not require unnecessary personal data; supplied claim text and source records are minimized, access-controlled, and excluded from public output by default. - Privacy failure test: an input requesting credentials, sensitive personal data, or unrelated private material is rejected with a safe explanation. - Abuse test: oversized, recursive, malformed, or adversarial citation payloads are bounded, logged, and rejected without executing embedded instructions. - Abuse test: source content cannot change reviewer role, gate status, or never-rules merely by appearing in a citation. - Source-integrity test: altered evidence digest or mismatched source locator produces a failed integrity result and a held record. - Authority test: a citation is never labeled authoritative solely because it parses, matches text, or appears in the supplied catalog. - Authority test: absent independent provenance, publisher identity, date, and scope evidence, the result is unknown or held. - Instrument-boundary test: compatible instruments are described only as proposal aids and never as proof of citation truth, authority, connectivity, or deployment. - Traceability test: every output decision links to the source record, claim record, evidence records, test results, reviewer, and gate state. - Regression test: changing a source reference or claim wording invalidates the prior qualification record and requires rerun. - Failure-state test: unavailable or unverified instrument behavior yields held, not passed, status. ## Negative and abuse cases - Unverified implementation may be mistaken for a functioning citation validator. - Text similarity may be misrepresented as factual support or authority. - A parsed URL may be mistaken for a reachable or trustworthy source. - Self-approval or conflicted review could create false assurance. - Contradictory evidence could be hidden by aggregation. - Draft work could be published without the final-publication gate. - Grant or credential decisions could be made from incomplete evidence. - Citation payloads could contain prompt injection, malware links, excessive size, or privacy-sensitive content. - Medical, financial, legal, or safety implications may arise from unsupported public claims. - The supplied owner reference may identify responsibility but does not independently prove expertise or authority. ## Qualification claim boundary ### Potentially allowed only after evidence and review - The catalog source record identifies the tool as "Citation checks". - Tilly is listed as owner and as Pointguard and independent reviewer. - The source record lists the related jobs, field guides, principal gates, never-rules, and agents/index.html#tilly as a source reference. - The proposed documentation can define deterministic qualification, accessibility, privacy, abuse, source, and authority tests. - The listed compatible instruments are proposal aids only, subject to separate qualification. ### Still held - The tool is implemented or deployed. - The tool can access, fetch, verify, or authenticate sources. - The tool establishes citation correctness, factual truth, provenance, or authority. - The tool is connected to any instrument or provider. - The tool has current performance metrics, customer use, effects, or production availability. - Any citation, grant claim, website claim, credential decision, or funding commitment is approved. - The source reference independently validates Tilly's authority. - Any instrument output constitutes independent evidence without human review. ## Sources and evidence These pointers bind the page to evidence. They do not grant authority or collapse verification into deployment. - `source_record:tool:citation-checks` - `agents/index.html#tilly` - `tool:citation-checks source record supplied in request`