3 · Proposed requirements — twelve, ranked, none filed

How these were ranked

Three inputs, in this order: how many platforms ship the capability (the survey’s own signal), how much a safety reviewer’s working day depends on it, and how well it fits what this portfolio is already good at. A capability on seven platforms that would require a different product ranks below one on four platforms that drops into the existing architecture.

Sizes are rough: S one session · M a few sessions · L a requirement with sub-issues · XL a multi-month workstream. Goal chips point at the standing goal each proposal would sit under.

Deduplication, stated plainly

Last night’s goal atlas proposed 32 candidate requirements from the internal landscape. This survey ran against the external one and landed on the same ground in seven places. Where that happened the atlas has priority and these pages defer rather than propose twice: C3 Kaplan-Meier (matrix A17), C4 exposure-adjusted rates (A19), C5 graded lab shift (A20), C6 concomitant-medication track (folded into P5 below, which proposes the superset), C7 static export (folded into P10), C8 performance budget (D10), A3 data-mapping surface (D1), A6 one-click report (folded into P10), A7 project snapshots (adjacent to B9, noted in P10).

What that leaves is the part the external survey found and the internal one could not: the review-workflow layer, which is P1, P2 and P7, and four chart families the migration inventory never contained because safetyGraphics never had them either.

The proposals

If only three get filed

The ranking is a full list; this is the recommendation.

P1 — review state and change-since-last-review. The strongest signal in the survey, filed nowhere, and the concrete answer to what the app goal means by “replacement”. Start with the renderer-side half, which is small.
P4 — mean change from baseline with confidence intervals. The cheapest genuine chart gap: two FDA Core figures, an unchanged data contract, and a display every surveyed platform has and we do not.
P7 — participant-level safety alerting. The one piece of external parity that lands on the autonomy goal instead of the chart goal, and the demo that ties this project’s two halves together in September.

P8 (auto-drafted narratives) is the highest-variance item on the list. It is the most demonstrable thing here and the one with a documented failure mode that matters — the literature is clear that the primary risk in clinical summarisation is quiet omission of true abnormalities, worst for rare high-stakes events. Worth filing as a requirement with that constraint written into the acceptance criteria, or not at all.