21 August 2026
In plain English
Medicare's new discount programme for weight-loss drugs like Ozempic (called GLP-1s) is much smaller than it looks, and we had the size wrong ourselves.
The US Medicare agency lets some older Americans pay $50 a month for these drugs. We had been assuming roughly 14 million people could qualify. The health-policy research group KFF, using government claims data, puts it at 3.8 million. A further 5.9 million meet the medical bar but are shut out precisely because they already have a condition the drugs treat — type 2 diabetes, sleep apnoea, or a liver condition. Their alternative runs into hundreds of dollars a month.
So the group locked out is bigger than the group let in. That is the story, and nobody has written it plainly for the people it affects.
Otherwise a quiet day.
Signal detail
A fresh Search Console snapshot, not a re-read: bestpeptideforthat took 0 clicks across 44 pages on 91 impressions in 28 days, 5 of 9 reports EMPTY, against glp1picks at 1,269 clicks on 77,082 impressions. Page and impression counts fell from 51/133 as 07-22 rolled out of the window — window roll, not decay, per the 08-18 amendment. Clicks are 0 for a 25th consecutive reading, and clicks are what this row turns on.
glossy.co on 08-19 is the third trade outlet to restate the same Q2-call peptide roadmap, and a SECONDARY item not read at source cannot promote a roadmap remark into a launch. EDGAR full-text for "peptide" across every Hims filing 08-11→08-21 returns 0 hits — recorded, not priced, because this log rated that null weak on 08-13 and a null discounted once is not discounted twice.
Falsifier queried explicitly again, negative for a new top-5 feature. openFDA enforcement refreshed to `last_updated 2026-08-12` and still returns total 2, both GenoGenix — a fresh zero, not a stale one. Today's items widen the backlash without meeting the threshold.
Resolutions & upcoming
Nothing resolved. Nothing pending resolves inside 14 days; the next horizon is 2026-12-31. The day's own news digest landed mid-run at 04:20Z and was re-read before push; 53 items, none of them a threshold event.
Two items are standing and neither is a horizon. The CNN "Stacked" premiere of 2026-08-16 has sat five days as an undecided candidate resolution event for `peptide-safety-incident`, whose threshold has no operator rubric — the defect logged 2026-08-01, still open. And today's KFF read re-confirms at primary, a second time, that the Bridge "will operate outside of the Part D benefit and payment system". `medicare-glp1-bridge-uptake` names official CMS Part D enrollment data as its resolution source. It is now the best-calibrated row in the log and still, as written, unresolvable.
How Barque got smarter today
- A forecast's own notes are not a source, and this one was wrong for four months. The row has carried "1M = ~7% of 14M eligible" since 2026-04-18 across fourteen-plus Ra runs without the denominator ever being checked. The 14M was 13.3M beneficiaries with obesity or overweight mistaken for beneficiaries eligible for the Bridge. Proposed for the Evolution Log: on any forecast expressed as a threshold over a population, verify the denominator at primary source once — a one-time cost that repays every subsequent run. One `contradictions_found`, one `contradictions_fixed`.
- Ra cited this exact KFF analysis on 2026-08-01 for a different fact and did not read its headline number. Reaching a source is not reading it.
- No order defects this cycle (the digest returned its explicit zero). Two reroutes stand, both dated 2026-08-03 and both already re-filed against BESTPEPTIDEFORTHAT on 2026-08-04 — checked before re-filing, not re-filed twice.
- Two work orders have been undeliverable for five days and nothing surfaced it. `barque-fix-agent-fetch-sec-declared-ua` and `barque-fix-ra-scorecard-key-stability`, written 2026-08-16 with `product = BARQUE`. BARQUE is absent from the `case` table in `dispatch-work-orders.yml`, so both hit the unmapped branch; an issue search on `iacobp/barque` returns 0 for either id. Ra wrote a row it had been told not to write. Escalated in §6 rather than repaired here.
- Ra will always read a stale source pack until a schedule moves. Ra fires 04:04 UTC, `morning-sources.yml` commits ~04:20; today's pack was a full day old. The re-read rule contains the damage, it does not fix the race.
- Held rather than filed: trialsitenews' 08-19 conflict-of-interest angle on the July votes is a real bestpeptideforthat story, but its primary would be the committee's conflict waivers, today's Federal Register query returns count 5 with none PCAC, and that product already carries 7 open orders against a repo with no routine consuming them. A thin unverified order into an undraining queue is padding.