D0027 — Where the data-coverage numbers come from

[obotclaw[bot]]

Data coverage per visit is the thirteenth safety-census number and the only one of the thirteen that could not be built. It is also the first item of the chart requirement you approved. Both stop in the same place, and answering this releases two items from the ranked ten: the census rebuild, which is second, and the coverage chart, which is third. Nothing else in the census is waiting — all four steps of the rebuild are merged.

Two things are missing and both could be closed by guessing. No standard domain says which visit a lab result belongs to. And "expected", which you defined as the participants whose time on study reaches that visit, needs a scheduled day per visit that no standard domain carries. A guessed coverage figure looks exactly like a measured one on the page, so every option was measured on both studies we have rather than argued.

What the measuring found, in three lines: the visit label is in the raw lab data on both studies and is dropped by a mapping specification we do not own — but a study's own mapping can pass it through, and our demo study already does, so this needs nothing from anyone else's package. Attributing a lab result to a visit by date matches nothing at all on the bundled study, and 12.5% — chance — when matched to the nearest visit. And your definition of expected, with the scheduled day inferred, produces coverage of 1,220% at one visit and no denominator at all at two more, because that study's visit calendar and its time-on-study column disagree for every participant in it.

The five questions, with the recommendation on each:

• DC1 (D0027.1) — where a visit label comes from. Recommended: the study says so, in its own lab mapping, and the metric stops rather than guessing when it is absent.
• DC2 (D0027.2) — what defines the day a visit is expected on. Recommended: two declared forms, the visit table or a study-declared schedule, taking the visit table when both are there, and never an inferred schedule. This is a variation on the answer you already gave, which is why it is being put back to you rather than built.
• DC3 (D0027.3) — what happens to a result from someone the denominator says is not there. Recommended: count only the expected, and publish the outsiders as their own number — the same shape you approved for the death count's two disagreeing sources.
• DC4 (D0027.4) — a study that supplies neither input. Recommended: publish nothing, and name the missing input.
• DC5 (D0027.5) — what orders the visits. Recommended: the study's declared order when there is one, the middle of the observed dates when there is not, and the day it ordered on printed on the row. Never alphabetical.

Every figure, every option's real cost, what the one surviving inference would take to be wrong, and four things written down before this that turned out not to be true, are on the page: https://jwildfire.github.io/obot.roadmap/reports/decisions/2026-08-22-data-coverage-inputs/

---

Drafted by 👯🤖 Claude Code (Claude Opus 5, worker W0121) for 🧭🤖 obot-navigator. Not reviewed by @jwildfire.