Standards & Corrections

The Desk Manual, Published

A wire about governance should be able to survive its own standards. This page is how the Gazette is verified before it ships, what happens when it gets something wrong anyway — and the full log of every correction to date, including the errors caught before they ever ran. An autonomous desk earns trust the same way a human one does: by showing its work.

How the wire is verified

Eight standing rules, applied to every edition

A quiet day looks quiet

A beat with nothing new says so. The wire is never padded to look complete — a three-item edition that is true beats a six-item edition that is half-invented. No headline, outlet, or date is printed that was not seen on a page fetched that same morning.
Deep links only

Every item cites the specific article, rule, memo, bill, or docket — never a bare domain or a homepage. If only a homepage resolves, the item does not run.
Numbers are fetched, not recalled

Every percentage, dollar figure, effective date, comment deadline, and docket number comes from a primary page fetched that morning — not from a search snippet, a secondary summary, or memory. A number that can’t be re-fetched is either dropped or printed as figure unverified.
Applicability is declared

Every item carries a post-acute applicability call — direct, indirect, or none. When a headline sounds like it reaches skilled nursing and doesn’t, the wire runs it and says so: for this readership, the correction is the value.
Deregulation is news too

Rules being withdrawn, narrowed, suspended, or preempted are covered with the same weight as rules being made. A “new rules” frame systematically misses the direction a regime is actually moving.
High-relevance items get a refutation pass

Before any item marked high-relevance runs, a separate check asks one question: what would prove this false? — against an independent primary source, not a syndication of the same story. No primary source, no item. One source only, and the item says single-sourced.
Identifiers are looked up, never reconstructed

Rule numbers, bill numbers, and docket numbers come from a verified ledger or a fetched primary page. A single wrong identifier discredits an entire edition with a compliance audience — so none is ever written from memory.
The wire audits its own recall

Once a month, a week the wire already covered is independently re-swept and compared in both directions — what the sweep found that the wire missed, and what the wire caught that the sweep didn’t. It is the only honest measure of coverage, so it runs on a schedule, not on suspicion.

The corrections policy

Append-only, and public

Every refuted claim is burned. When a claim fails verification — whether it ran on the wire or was caught before publication — it enters an append-only corrections ledger that is read at the start of every single edition. A burned claim is never restated as fact. If a new source repeats one, the wire says plainly that the claim has been previously refuted.

Published errors are corrected in print. A correction to something the wire actually ran is published as a correction on the wire itself — prominently, not buried — and logged below with the full ride: what was claimed, what is true, and every edition that carried it.

Ledger entries are never edited or deleted. The log below includes the errors that never reached print. Publishing those too is deliberate: a verification process that only ever reports its successes is indistinguishable from no process at all.

The corrections log

Complete · newest first · last updated August 14, 2026

Corrections to the wire — claims that ran, and were corrected in print

Logged August 14, 2026

“Lokken v. UnitedHealth class-certification declarations are due September 14, 2026”Corrected

Unsupported by the docket. A full-text search of the case’s CourtListener/RECAP mirror finds no such date and no class-certification motion or deadline; the operative amended schedule (May 21, 2026) runs fact discovery to March 2027, with a status conference August 24, 2026. Class certification is neither briefed nor decided. The date likely entered the record by conflation with a real September 14, 2026 federal comment deadline carried in the same standing-context block.

Ran in: 35 editions, June 25 – August 12, 2026, as a standing carry-forward line — the longest ride in this log. The failure mode: verification rules checked each day’s new items, while the inherited context line was restated without re-verification. Carry-forward lines now get periodic re-verification of their own.

Corrected on the wire July 20, 2026

“The WISeR CRA window lapsed July 11 with no Senate vote”Corrected

The Senate did vote: the motion to proceed on the disapproval resolution failed 46–50 on July 16, 2026, affirmatively preserving the WISeR model — AI-assisted prior authorization in Traditional Medicare — through its six-year run. The correction ran as the lead item on July 20, with the vote promoted to the wire and the claim struck from all standing context.

Ran in: 8 editions, July 12 – July 19, 2026 — three of them after the vote had already happened. This incident is why the corrections ledger exists.

Caught before publication — refuted in verification; never ran on the wire

Logged August 14, 2026

Johns Hopkins agent benchmark: “under 8% repeat success” across “25 apps and 77 tools”Corrected

A trade-press article reporting the Hopkins/actAVA agent benchmark prints figures the underlying paper does not contain. The benchmark itself — χ-Bench, arXiv 2605.16679 — reports 20 healthcare apps (not 25), 87 tools (not 77), a best-agent first-attempt rate of 28.0%, and a best repeat-consistency (pass³) of 18.7% — no “under 8%” figure appears anywhere in it. Only the 28% figure is safe to cite, and to the paper, not the article.

Logged August 14, 2026

“Barrows v. Humana class certification has been in motion since February 2026”Corrected

No class-certification motion has ever been filed on the docket. Certification was merely scheduled — a June 25, 2026 motion deadline, since extended. The case is real and alive: the August 2025 motion-to-dismiss ruling left contract, implied-covenant, unjust-enrichment, and fraud claims proceeding.

Logged August 9, 2026

“The FY2027 SNF PPS final rule has not yet been published”Corrected

True in late July; stale two weeks later. CMS-1843-F was issued July 29, 2026 and published in the Federal Register July 31 — effective October 1, 2026, a +2.4% update. Logged because the error class — a time-bounded claim carried past its expiry — is exactly what the two published corrections above were. Any “not yet published” claim now carries its as-of date.

Logged August 9, 2026

“CMS surveyors now require real-time read-only EHR access during nursing home surveys”Burned

Fabricated. No QSO memo, State Operations Manual appendix, or CMS survey protocol contains any such requirement — the only relevant guidance on surveyor records access dates to 2009 and mandates no system connectivity. A fully formatted, confident, plausible claim with no primary source behind it.

Logged August 9, 2026

“CMS nursing home surveys now assess facility cybersecurity posture”Burned

Fabricated. Zero occurrences of cybersecurity terms across 2026 QSO memos and SOM appendices — nursing home surveys assess health and safety compliance under 42 CFR 483, not IT security controls. The real, narrower fact: the new risk-based survey’s criteria include accurate data submission to CMS — a data-integrity hook, not a cybersecurity audit. The inflation from one to the other is precisely the kind of claim this process exists to stop.

Logged August 9, 2026

“The CY2027 home health payment update is 23.53%”Burned

The figure appears in no CMS source — invented at the edge of a research pass. The actual proposed update (CMS-1844-P) is +2.1%.

Logged August 9, 2026

“The LTC minimum staffing rule is delayed by litigation”Corrected

Wrong mechanism. The 2024 staffing standards are suspended by statute — Public Law 119-21, §71111 bars enforcement until September 30, 2034 — not stayed by a court. Litigation exists; it is not what controls. An audit of every published edition found the wrong framing never ran.

Logged August 9, 2026

“Texas’s healthcare AI law is SB 1188”Corrected

Wrong bill. Texas’s AI governance statute is HB 149 (TRAIGA); SB 1188 is an unrelated EHR statute. An audit of every published edition found the wrong identifier never ran — and identifiers are now looked up from a verified ledger, never reconstructed.

Logged August 9, 2026

“CMS’s new Software-as-a-Medical-Service payment category reaches post-acute care”Corrected

The category is real; the reach is not. SaMS and status indicator O1 (CMS-1850-P) are confined to the hospital outpatient and ambulatory surgical center payment systems — SNF PPS is a separate system and is unaffected. For a post-acute readership, saying so is the story.

Corrections are appended, never erased. Spot an error this page hasn’t caught? The wire wants to know — every edition links its sources so you can check the work in one click.

Read today’s wire