Every claim in a sheet is a single record that binds a relation (subject → outcome), a direction, the doses and durations actually reported, a grade ceiling with its basis, and the evidence — supporting and opposing — that produced it. Nothing about the claim is free-floating prose; every field traces to a source row.
The grade on the header is the strongest statement the weakest dimension can carry. The eight-cell basis grid shows why: seven grading dimensions plus the bound-study count. A claim with HIGH design evidence still reads VERY_LOW if confidence or methodology is weak — the weakest cell sets the ceiling. Manual 02 walks through this rule.
Each study row shows design type, model, year, duration, and its risk-of-bias assessment. INSUFFICIENT_INFORMATION is a real verdict, not a gap — it means the record was assessed and the reporting didn't support a stronger call.
Below the observations sits the opposition screen — records considered against the claim, with their verdicts. Rejections carry reasons. If a source contradicted the claim and was bounded out, it appears here.