statusDetail described the sentence only as "An unconditional discharge imposes no imprisonment, fine, or probation but leaves the conviction in place," without Judge Merchan's own stated reason for structuring it that way. Read on its own, the line can land as "the crime didn't really matter," which is not what the record supports. Merchan said the opposite.

Ten Corrections entries (dated 2026-08-14 and 2026-08-15) described the findings behind them as coming from an "independent fairness review," "independent readability review," or "a reviewer built specifically to distrust euphemism and false balance", while the Masthead's own "Review record" section states plainly, describing the same underlying passes: "These passes were run by the author against the author's own work... This is not independent review." A reader who opened Corrections without first reading that Masthead disclaimer would reasonably take "independent" at face value, which contradicts what the archive says about itself two clicks away.

The "withheld-new-entrant-278" row (the archive's own "most careful page, on purpose") rested on a single source: a June 2025 letter from Sen. Adam Schiff, a Democratic critic of the administration, labeled "Tier 2" even though this archive's own glossary defines Tier 2 as peer-reviewed research or a nonpartisan institution, and its own precedent (a bipartisan Khanna/Massie legislator letter elsewhere in the archive) classifies a lawmaker's letter as advocacy, not Tier 2. That made this the one row on the archive's most sensitive page thinner than the sourcing bar the rest of the site holds itself to. checkEntry()'s advocacy-not-alone rule requires a Tier 3 source never stand alone, but financialTies.json sits outside checkEntry()'s coverage entirely, since it only walks entries.json.

The "Southern border apprehensions fell sharply" claim already carried three estimates (the 85,218/92%-decline figure via FactCheck.org, the administration's own framing of it as its central achievement, and a negative-finding estimate on the absence of a causal-attribution study). A fairness review flagged that this claim reads thinner than the same entry's enforcement-cost claims (majority non-criminal arrests, four deadly task-force shootings with two named victims, sustained protest), which run longer and carry more specific detail. That review's premise that this claim was "a single bare statistic beside a one-line DHS quote" was itself inaccurate (it already had three sourced estimates) but the underlying observation, that the claim lacked absolute/comparative magnitude next to a headline percentage, held.

The summary read "A US-brokered framework produced a ceasefire in Gaza and the release of remaining living hostages, after two years of war." That was stated as flat archive fact, with none of the hedge this same entry already carries two fields down: its own "contested" field states "Credit for a mediated agreement is inherently shared among mediators and parties. This entry records the US role as decisive per analysts, not as sole cause," and its own effect estimate attributes the identical claim to a source ("US pressure on both parties is widely credited by regional analysts as decisive in reaching the framework"). The summary is the first and most-read text on the entry and asserted the stronger, unattributed version of a claim the entry's own body explicitly disclaims, a matter of characterization (credit-sharing among multiple mediators, Qatar, Egypt and Turkey, and both parties to the war), not settled fact. `checkEntry()`'s CAUSAL check only scans `effect.claim`, so nothing mechanical catches an inconsistency between an entry's summary and its own body.

The Financial Ties view documented Trump's own holdings and the Trump-family crypto/federal-crypto-policy overlap, but carried nothing on Elon Musk (named in the DOGE entry as its public face) despite SpaceX and Tesla holding federal contract and regulatory relationships with agencies (FAA, NHTSA) that DOGE's government-wide reach touched.

The fake-elector effect claim ("Certificates from fake electors were submitted in seven contested states") documented the scheme as established fact (the House Select Committee's report, Eastman's upheld disbarment, and state grand jury indictments) but carried no estimate stating the electors' own defense theory, that the slates were 'contingent,' meant to preserve litigation options pending court challenges. Every other comparably contested entry in this archive (DACA, sanctuary jurisdictions, DEI, the transgender military ban, Title X) pairs a disputed finding with the accused party's own characterization as a labeled estimate; this was the one exception.

The "Standing indicators" strip on Overview had no figure for Cabinet-level appointments or turnover, despite tracking judges seated, border enforcement, and other personnel-adjacent figures.

The "Controls below the WCAG AA/AAA target-size minimum" row read "Not yet re-measured against the built app". No longer true. It had in fact been measured twice: a desktop pass (0 violations, CI-enforced) and a separate manual mobile pass at 390px (also 0 violations), neither reflected in the copy.

Indicators' and Markets' term-comparison tables, and Personnel's yea/nay vote bars, colored figures using the site's actual `--color-cost`/`--color-benefit` tokens purely by arithmetic sign or vote direction, a falling unemployment rate or a `nay` vote rendered in the same red used elsewhere for net-cost incidence, silently contradicting "color marks standing only, not good or bad" on pages that state that exact rule. Separately, the shared "New here?" onboarding banner's dismiss button landed at the bottom-left of the card instead of pinned near the heading whenever its text wrapped on a narrow screen (a flex-wrap/justify-between interaction), and every view's specific explanatory subtitle was deleted from the DOM entirely on mobile (`hidden md:block`). Dossiers' inline source labels (e.g. "GAO report, August 6, 2026") were plain, non-interactive text, contradicting that page's own banner claim that every number links to its source.

The `ice`, `edu`, and `s232` promise rows showed only the linked entry's short `status` field. Each entry's own effects data documents a fact that complicates or contradicts the promise's own wording: the majority of surged-area arrests being non-criminal-conviction (ice), halved staffing with documented civil-rights and aid-servicing backlogs (edu), and a net manufacturing-job loss from the tariffs meant to bring jobs back (s232). None of it reached the Promises page. The page also gave no account of why some promises read "Kept" and most read "In progress," and native vs. linked promises were distinguishable only by a subtle date-format difference.

The archive had no scale-ranked view of its own entries and tracked only 4 campaign promises (wall, aca, nowars, ukraine24), all sourced entirely from within their own entries. 96 entries total.

The `futureact` entry's contested field joined two independent clauses with a semicolon into one long sentence. The `unemploy2019` entry's contested field referenced "the Tier 1/2 literature cited here", the archive's own internal source-tier shorthand, used inline in prose with no glossary link and no other instance of it anywhere in the 96 entries' free-text fields.

The presidential comparison table covered debt added, debt per year, net jobs, and judges confirmed, four axes, none of them a measure of executive power itself.

Every entry's "contested ground" disclaimer (the field recording what's genuinely unresolved, including where the archive itself could be wrong) rendered after all of that entry's claims and estimates, for all 92 entries that carry one.

Four separate stat displays (the new elapsed-term comparison (caught pre-publish), the "Trump II so far" cards on the vs.-Presidents page, the "Standing indicators" ticker at the top of the Overview view, and "Where the presidency stands right now" just beneath it) colored figures red or green by whether the number read as favorable or unfavorable to the administration. A drop in border apprehensions was colored green and captioned "the administration's central claimed achievement" in two of the four; debt, job losses, and a falling approval number were colored red in the rest. All of this sat on pages whose own banner states "no scores... color marks standing only, not good or bad."

The Overview view's 23 headline tiles rendered in whichever order they had been appended to `headline.json` over time, which front-loaded costs (the first four tiles were cost, mixed, cost, cost) with both benefit-tagged tiles not appearing until well past the halfway point.

Of the page's three substantive sections ("favours him," "cuts against him," "nobody is right yet"), only "favours him" carried a colored border, a colored heading, and a working link to the sourced entries in the Ledger. The color used was STAND.ACTIVE (the "in force" legal-standing token) repurposed to mark this one section as the favorable one, with no equivalent path to verify the other two.

Two claim headings used narrative framing not matched elsewhere in the entry: "An organized effort submitted certificates from fake electors in seven contested states," and "The campaign pressured Vice President Pence to reject or delay certification of the electoral count." A third heading, "Legal consequences for the events themselves," was vague relative to its own sourced content (the Senate acquittal and Trump v. Anderson).

A completed roadmap item read "Accessibility: WCAG AA contrast, 44px targets, skip link, main landmark, drawer focus management."

Status read that tariffs "terminated February 24, 2026," and refund exposure was given as a single figure, "near $166 billion," sourced to "trade-law practitioner analyses."

151 distinct source-lean labels, including pure synonyms ("Academic" / "Peer-reviewed academic" / "Academic / peer-reviewed"), and British and American spellings mixed roughly evenly across 36 instances.

Status read "Partially implemented; multiple actions litigated," and the savings-claim effect cited press verification of documentation errors.

Stated that Section 232 tariffs "were struck down alongside the IEEPA tariffs in February 2026."

Effect claim "Federal deficits increase relative to prior law" was labeled Contributory.

Status detail said "roughly 450 miles were built."

Status read "In effect; under Supreme Court review."

Cited CBO’s 13-million projection without noting CBO’s later revision.

40 of 215 source URLs pointed at a publisher's homepage rather than the document the source's own label named. "Trump v. CASA, Inc. (2025)" resolved to supremecourt.gov; "Navigable Waters Protection Rule, Federal Register" resolved to federalregister.gov.

The entry's executive-order source was deep-linked to Executive Order 14151 (Ending Radical and Wasteful Government DEI Programs and Preferencing). The entry states four times (in its summary, its action, its status detail, and a Direct-attribution effect claim sourced to "Executive order text") that Executive Order 11246 was revoked. EO 14151 does not mention EO 11246, revocation, or affirmative action anywhere in its text.

The source labelled "American Hospital Association v. Azar, D.C. Circuit (2020)" linked to https://www.cadc.uscourts.gov/, the court's homepage. A reader clicking to verify that the rule survived challenge landed on a front page. It was left that way on the recorded finding that the D.C. Circuit publishes opinions under unguessable GUID paths and that the govinfo USCOURTS package IDs do not resolve.

Six source urls point at a publisher's site rather than at the document their label names, because each label names a plural or open-ended body of material, or in one case a document that is not public. The reasons were recorded in an internal document. On the site itself the links were indistinguishable from a citation that had simply gone stale.

The Flaaen & Pierce Federal Reserve source pointed at a URL ending "…2018-2019-tariffs.htm", which returns a Federal Reserve "Page not Found". The stored address was a truncation of the real one, the paper's slug continues "…-on-a-globally-connected-us-manufacturing-sector".

The PIIE Phase One tracker source pointed at /research/piie-charts/chinas-purchases-us-goods-under-phase-one-trade-agreement, which returns PIIE's "Page Not Found". PIIE reorganised its chart URLs under a year segment at some point after the citation was written.

The source labelled "FDA COVID-19 vaccine authorization record" pointed at an fda.gov page that returns "Page Not Found". FDA reorganised the section and did not redirect: their own current COVID-19 landing page still links to a "COVID-19 Vaccines" node that 404s as well, so there is no live successor to follow.

The entry stated that presidential-family crypto ventures had drawn ethics scrutiny, sourced to a single estimate whose source field read only "Government ethics organizations". No group named, no dollar figure anywhere in the entry despite one being disclosed and publicly reported.

The entry covered the certification delay, the impeachment and acquittal, and the dismissed federal case, but its effects never mentioned the fake-elector scheme or the pressure campaign on Vice President Pence, the documented conduct at the center of the House Select Committee's findings and the DOJ indictment.

The Disputed "Legality under domestic and international law" claim paired the administration's own position (a DOJ opinion it has declined to release, which does not itself assert legality) against a named Democratic legislator, Rep. Jim Himes, calling the operation "clearly illegal." No independent or right-leaning legal-scholar defense of the Article II / 2002 Iraq AUMF theory was cited, even though this archive cites exactly that kind of counterweight elsewhere on comparable questions, Justice Alito's dissent on the National Guard entry, and "minority scholarship... cited by the administration" on the birthright-citizenship entry.

Every entry drawer's "Cite this entry" block built its permalink from `SITE_ORIGIN` in `src/lib/citation.ts`, which still read `https://the-record-taupe.vercel.app` after `thetrackrecord.org` became the live, canonical domain. A reader citing any entry got a URL on an abandoned Vercel subdomain instead of the real address, directly beside the drawer's own sentence promising "this entry has a stable address and will not be silently rewritten," the one claim on the page specifically about citation durability.