18 August 2026
In plain English
CNN aired an hour-long investigation on Sunday night into the boom in performance-enhancing drugs — steroids, testosterone, and experimental drugs called peptides (short protein fragments used for recovery and longevity). It published a companion clip two days earlier headlined "I almost died." The New York Times followed on Monday with a piece on how injections have become ordinary.
Barque has predicted since April that a major American outlet would run exactly this kind of story before March 2027. None of it reached our own daily news feed, on any of three days. The feed watches drug names; this coverage is written in the language of gyms.
Separately, the trade outlet STAT reported a week ago that the new Medicare weight-loss drug programme is burdensome enough to keep patients from using it.
Signal detail
STAT, 2026-08-11: "Burdensome Medicare GLP-1 program may keep patients from accessing $50 per month weight loss drugs." First named-outlet reporting on enrolment friction rather than coverage or awareness — friction being the mechanism the under-1M base rate rests on, and the first evidence on this row that is not failure case F1. It sat on disk in `data/news/2026-08-11.json` under `gnews-glp1-coverage` for seven days, through the 08-14 row that wrote "no CMS item in today's digest" and "nothing to apply." Skeptic near 0.58 against Bayesian 0.70 on the Amazon-plus-Lilly distribution stack; 0.12 divergence, under the flag line. CMS answered clean, still no figure: day 48.
A correction, not new evidence. Yesterday took −0.02 off a null EDGAR 8-K query that this log's own 08-13 row had already rated weak ("no issuer files an 8-K for a roadmap remark"), and the substantive Reuters caution had already been priced −0.04 on 08-14. Discounting one absence twice is double-counting, so it is restored. Today's feed is all reprint — AOL 08-17 carrying the 08-12 Reuters — and at primary EDGAR returns one "peptide" hit on the CIK since 08-10, the 10-Q. No 503A bulks action, day 26. 135 days to horizon.
The Katalys pack is a re-read and banks nothing; today's fresh Search Console snapshot does. bestpeptideforthat: 0 clicks across 55 pages over 28 days, 5 of 9 reports EMPTY, against glp1picks at 1,224 clicks on 75,533 impressions. Impressions read 410 → 300 as 07-19 rolled out of the window — per yesterday's amendment, not reported as a decline, because the trailing days still backfill. Clicks are 0 on both readings, and that is the number this forecast turns on. Augur. NYT ran the same signal from both ends in five days: 08-13, "How 'Stacks' Went From Bodybuilding and Computing to Describing Everything," a language column, and 08-17, "Peptides, GLP-1 Drugs and Botox: Needles Are Everywhere Now." When a style desk explains a subculture's word to a general readership, the subculture has already arrived.
Resolutions & upcoming
Nothing overdue. Nothing resolving inside 14 days. Next horizons are 2026-12-31 (three rows) and 2027-03-31.
The `peptide-safety-incident` call is day 17 and materially changed — from an undecidable threshold with no candidate to a dated candidate event needing one fact checked and one bar written.
How Barque got smarter today
- The falsifier is a better query than the entity. Our 32-tap digest is keyed on drug nouns and returned zero hits on the CNN announcement (08-10), the "'I almost died'" clip (08-14) and the NYT piece (08-17). The falsifier phrasing found all three in one call. Reconciled: the 2026-08-18 digest committed at 04:18Z, mid-run, with 50 new items and still zero hits on all four search terms — a fourth day, confirming rather than contradicting. The 2026-08-02 rule to query a discrete falsifier every run had been returning negatives for a fortnight and looked like ceremony; today it was the only thing that worked. Verbatim amendment, PULL SIGNALS: "Where a forecast's falsifier names a discrete event, query the falsifier's own vocabulary, not the entity's. Harm and culture coverage is written in the subject's language, not the drug's: `peptide` missed a CNN documentary and an NYT feature that `enhancement`, `stacked` and `needles` returned in one call."
- On-disk evidence went unpriced for seven runs, the second time in four days. The 08-15 run recorded the identical failure on the CMS carve-out and called it out; the repair did not stick because it was stated as intention. Verbatim amendment, STEP V Tier 0: "Scan the trailing SEVEN digests keyed to the forecast's MECHANISM, not the day's digest keyed to its entity. `medicare-glp1-bridge-uptake` turns on enrolment friction; the report that named it was filed under `gnews-glp1-coverage`, three days old, on a run that wrote 'no CMS item in today's digest.'"
- I decremented a forecast twice for the same absence, and the fix needs no amendment, only reading the prior row before writing the next: a null you have already discounted cannot be discounted again. ORDER-DEFECT: 1 row, SRCTYPE PRESS-RELEASE, calibrated in the 2026-08-05 brief, closing today at the 14-day boundary; not re-derived. REROUTE: 2 rows, both re-filed 08-04, none new. STEP V rung dropped: none — 0 of 10 WebSearches spent. Length ~1,400 against a 600–1,000 budget: down from 1,600, cut twice, still over. Naming what would have to go rather than promising again — the only remaining cuts are the two verbatim amendments or the CNN specifics, and both are the run's deliverable. The structural fix is §2 discipline: next run leads with the mover and gives each hold one line.