🧪 Populated matrix — covid-19 vaccine myocarditis/pericarditis. This condition now carries real literature. Node badges are computed from the adjudicated (curated) records only, exactly as the gating code does. A second tier of screened records (from the automated literature pipeline) is shown inside each node clearly marked “needs review” and does not change any status. Claim C3 (epidemiologic excess risk) is computed as CONTESTED — the age-stratified VSD excess (Goddard) and the all-ages null (Klein) both stand; that tension is preserved, not resolved. Other framework nodes with no records remain in their “not yet searched” state. This is a research-transparency view, not a verdict that the vaccine caused any individual case.
The Matrix is deliberately built so it is structurally difficult to present a one-sided case: every node shows supporting and refuting records side by side, “no data found” is a tracked state distinct from “refuted,” a node’s status is computed from its records (not set by hand), and a condition summary cannot be read without its coverage indicator. It is a research-transparency instrument, not a verdict.
Coverage — how complete is the search, before reading any conclusion
Claims — six pre-defined evidentiary questions
Node grid — the observation slots that feed each claim
supports records support
refutes records refute
mixed both
no data searched, none found
not searched default
How the framework enforces discipline
- Symmetric search. Every supporting query is paired with a mirror “null / refuting” query in the same session; both hit counts are logged even when the mirror returns zero.
- “No data” is a real result. A searched-but-empty node is recorded with its exact strings and date — distinct from “refuted” (absence of a study is not evidence against an effect) and from “not yet searched.”
- Computed status. A node’s supports/refutes/mixed/no-data status is a function of its literature records, not a field anyone can set directly.
- Pre-registration lock. A node’s hit rule and falsifier rule become read-only once the first record is attached; later changes require a new versioned node with an audit trail.
- PRISMA accounting. Records identified → screened → excluded (with reason) → included are tracked per node, so a summary reads like a systematic review rather than an advocacy brief.
Full spec: evidence-matrix-feature-spec.md; search protocol: literature-mining-strategy.md. The live write/audit tool is a separate, database-backed build. See also the Mechanistic Pathway Matrix (mechanism families per combination) and the vaccine–AE association database index. The former applies the M1–M12 mechanism families to every vaccine–adverse-event combination on the site.