Evidence Cards and Provenance Rails: Scope, Traceability, and Validity

The current reviewed series contains thirteen articles, 115 numbered sections, 69 charts, thirteen evidence cards, and thirteen provenance rails. Each article places cohort rules, measured findings, comparisons, interpretation, and limitations in the reading column. Its evidence card records corpus size, matched count, coverage window, and extraction date. Its provenance rail records producer, run date, named databases, and primary-source classes. These elements make figures traceable. They do not establish that a source is complete, a cohort definition is valid, or an interpretation follows from the measurements. Those claims require separate review.

1. Inspection scope and counting rules

This audit covers the thirteen numbered article bodies in the reviewed series. It counts explicit numbered section headings, chart blocks, evidence-card blocks, citation lists, and provenance blocks in their publication records. The count measures document structure. It does not grade the scientific quality of each claim.

2. Findings remain in the reading column

The reading column carries the full argument. It starts with the cohort and counting rules. It then presents the descriptive census, composition checks, controlled comparisons, supported interpretation, limitations, and primary citations. Charts stay beside the paragraphs that define their denominators. The evidence card also stays in this column because its row counts and coverage dates qualify the findings.

3. The provenance rail owns source identity

The page separates provenance blocks from the main document and renders them in a side rail. The rail records who produced the article, when the run occurred, which databases contributed, and which primary-source classes support the work. This separation prevents source identity from displacing findings. It also prevents charts and evidence cards from shrinking into metadata panels.

The structure follows the central W3C provenance distinction among entities, activities, and agents. An article is an entity. A dated research run is an activity. Academica is the producing agent. Named databases and primary records identify source entities. The rail does not expose private connection details or executable query text.

4. Traceability and validity are separate claims

An evidence card can establish which corpus view supported a figure, how many records were searched, how many matched, the covered dates, and the extraction date. It cannot establish four different validity layers at once. Corpus validity concerns completeness, updates, duplicates, and missing fields. Matching validity concerns cohort definitions and classification rules. Comparative validity concerns denominator alignment and confounding. Interpretive validity concerns whether the prose follows from the measured result.

5. The GLP-1 review is the worked example

The GLP-1 adverse-event review contains seventeen numbered sections, fifteen charts, one evidence card, one citation list, and one provenance rail. Its evidence card records 368,915 matched reports from 11,175,581 archive rows through 30 June 2026. The body then tests reporter qualification, geography, serious flags, concomitant drug burden, administration terms, clinical terms, and indication composition.

The card proves the scope and extraction point of the matched cohort. It does not prove causal drug-event relationships or comparative treatment safety. The article addresses those limits through reporting-regime comparisons, term-level profiles, explicit noncausal language, and a final limitations section.

6. Publication metadata supports reruns without exposing infrastructure

Readers receive source classes, coverage dates, match counts, and extraction dates. Internal reruns retain executable definitions outside the browser document. This boundary protects connection details while keeping each public figure attached to a stable scope statement. Reproduction still requires source access, the complete matching rule, software behavior, and the same extraction boundary.

7. Supported interpretation

The reviewed series now uses one publication contract across thirteen articles. Findings and their qualifications remain in the reading column. Source identity remains in a provenance rail. Evidence cards expose scope and time boundaries without exposing executable infrastructure. This structure improves traceability and consistency. Scientific validity still depends on source quality, cohort construction, comparison design, and bounded interpretation.

8. Limitations

Block presence cannot prove that prose and figures agree. Primary citations can identify authoritative records without proving corpus completeness. A recorded extraction date cannot prevent later source revisions. Query-free public documents require controlled internal preservation of matching rules for full reproduction. The thirteen-article scope excludes other published content outside this reviewed series.