fix(people): tell same-named people apart; a history of finished runs, each with its time and run, read as the patient page reads it (#655)

[Taleef7]

Three items from the #655 checklist, all checked on live Maui first (admin seat, 2026-09-28).

What was wrong

• Same-named patients were indistinguishable. /people rendered only a name. Three rows in a row read "Adriana Aoki", with nothing to tell them apart.
• Duplicate history rows. pat-04403's history had 147 rows, and same-day rows looked identical (for example, three "9/24/2026 Breast Cancer Screening MISSING DATA" rows). The history read every outcome of every run and dropped the time of day. That included rows from runs still going or failed, and rows for measures the deployment no longer runs.
• "MISSING DATA" where the patient page says "Not in population". A 30-year-old's breast screening, colorectal screening and CMS137 rows showed the stored bucket.

What changed

• The list shows an identifying line under each name: the id, the date of birth and the clinic.
• The history keeps only rows that are answers:
  • rows from finished runs (COMPLETED or PARTIAL_FAILURE, the compliance API's rule);
  • only the measures the deployment runs (as on the patient page, The patient page says MISSING DATA where its own table says Not in population, lists a measure outside the ACO set, and re-evaluates on every open #671).
• Each row names its run. It carries its runId and what started it: Nightly, Manual run, Single-patient run, Rerun to verify, or Seed (synthetic). The page shows the date and time, so two evaluations on one day read as two runs.
• Status is the measure's reading of each row (deriveCell), so out-of-population rows read "Not in population". The patient page's cohort overlay is deliberately not applied to history, because it would rewrite every past row whenever a segment or a role changed.
• Run lookup in one read. The runs behind a person's rows come from a new RunStore.getRunsByIds, which is one read in both stores and is contract-tested. It replaces one round trip per run, which grew with every nightly on a stack that never compacts.
• Patient-page link. Each linked record now links to its patient page.

Out of scope

These stay on #655:

• the unmasked national-id column (a PHI-phase decision; always empty on Maui);
• the patient page reading its newest row from any run, including one still going.

The journal line follows once #721 merges, to avoid a conflict in the same day's section.

Verification

• Backend:
  • typecheck;
  • identity, employees and SQLite store suites (169 pass);
  • the Postgres store contract runs in CI (Docker is not available locally).
• Frontend: lint, and new vitests for both People pages (neither had any).
• 12 mutations across the new tests, all killed.
• My own code review before opening. Its two important findings are fixed in ce1ef97:
  • the cohort overlay rewriting history;
  • an unbounded per-run query loop.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce1ef9745d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".