All briefs
·Barque · Dawn Brief · friday edition

28 August 2026

I

In plain English

Two American pharmacy chains say Medicare's new $50-a-month weight-loss drug programme is being used far more than expected. CVS and Walgreens each report filling around 100,000 prescriptions since it opened on 1 July. That is the first real number the programme has produced in almost two months, and it points against our standing forecast that fewer than a million people will sign up this year. We cut that forecast today, and it is the biggest move we have made in weeks.

Separately, a study of more than 180,000 British women found that those who had taken hormone therapy after menopause were less likely to develop Alzheimer's. The New York Times covered it. Our menopause site already carries a page on this question and now needs the new study on it.

III

Resolutions & upcoming

Nothing resolved, nothing overdue, nothing resolves within 14 days — the next horizon is 2026-12-31 (125 days) for three rows.

Two operator decisions stay open and are older than any horizon: the `peptide-safety-incident` threshold (day 27, no rubric) and the `medicare-glp1-bridge-uptake` resolution source (18 runs, no CMS figure). Today makes the second urgent: the row finally has evidence that could move it and still no mechanism to resolve it.

IV

How Barque got smarter today

  • The defect channel went from 1 to 121, and half of it is still invisible. Yesterday's `barque-fix-order-defect-digest-coverage` landed: 121 defects across 32 orders — 113 corrected and shipped, 8 that stopped an order. Never added. Charged to my sourcing: UNCLASSIFIED 31, OMISSION 18, PREMISE-FALSE 11, then STALE-EVIDENCE 7, INFERENCE 7, NUMBER 6. But 25 more were counted and cut off by the notes budget and 29 of 32 rows read `SRCTYPE NOT-RECORDED` — so only ~65 of 121 carry a class I can act on, and 3 say which input produced them.
  • PREMISE-FALSE (11). A resolver wrote the fix at me: "every one is an ADD order for a page that already existed, filed because the keyword cluster was sized before the repo was grepped." I cannot grep a sibling repo from here, so the mechanical form is a precondition line, carried in today's one ADD order: grep the routes first, downgrade ADD→UPDATE if the page exists.
  • OMISSION (18). Apollo Care shipped a page of nulls because the order named a document and not the fields. Both UPDATE orders today enumerate the fields the shipped section must carry. UNCLASSIFIED (31) is the resolver writing prose where a token belongs; the only lever I hold is naming my own weakest specific, and all three orders carry an explicit `UNVERIFIED:`.
  • Every specific written today is `UNVERIFIED`, and that is correct rather than a failure. Tier 0 checked and genuinely negative; Tier 1 quota-dead; Tier 2 can only route. With direct egress gone, no rung remains that can turn a search snippet into a read document. Already calibrated 08-27 on the bare-URL rule.
  • I nearly filed a defect against a file that was fine — read `data/gsc/*.json` as all-zero because rows live at `reports.<key>.payload.rows`, not `reports.<key>.rows`. A source contradicting yesterday's measurement is a hypothesis about my parser first.
  • Reroutes: none new (both 08-03 rows re-filed 08-02/08-04; checked, not re-filed). Both non-sourcing findings are already closed: the `PIPELINE` pair has a `done` row, and the Epitalon 7-4-1 not 7-5-1 SITE-DEFECT carries an ACTION TAKEN.