Masthead & audit
Who publishes this, and a live audit of its own integrity.
Standing rule: no scores. Color marks whether something is in force, reversed, contested, concluded, or permanent, not good or bad.
New here? Read this first (30 seconds)
- Color is a legal fact, not an opinion. It shows (whether something is still in force, was reversed, or is contested in court), never whether it was good or bad.
- Every number links to its source and to how strongly it can be tied to the action (). Click any figure to see the full evidence.
- Nothing here is scored, graded, or ranked. When two credible sources disagree, this archive shows both numbers instead of picking one.
- Wins are in here too. See or filter the Ledger by .
- Nobody is asking you to trust a byline. There isn’t one — the archive is anonymous, on purpose, which is exactly why it has to be checkable without one. See and check one entry against its own source yourself.
Masthead
A tracker that will not say who is behind it has no standing to demand transparency from anyone else. The bracketed fields below are the ones only you can fill — they are shown as unfilled rather than quietly omitted.
Who publishes this
An independent archive, free to read and free to reuse. The maintainer is anonymous, deliberately.
No company, no newsroom, no institution, and no name, because the subject is a sitting political figure and naming a lone maintainer invites harassment rather than scrutiny. That is a real cost and it is not waved away: anonymity removes the usual way a reader checks a publisher, so the archive has to be checkable without one. That is what the rest of this page is for. Every claim links to a primary document, the full dataset downloads from the footer, and every error the archive has made is published in the corrections log rather than quietly fixed. Judge it on whether the sources say what it claims they say.
Who funds it
Nobody. No funding, no grants, no advertising, no sponsorship.
There is no money in this and therefore nobody with a stake in what it concludes. If that ever changes, every funder gets named here with amounts and dates, and any entry a funder has an interest in says so on the entry.
Who edits it
One anonymous person. There is no second editor.
This is the archive's biggest structural weakness and it is disclosed rather than buried. The CMS enforces a two-person sign-off (the account that last edited an entry cannot approve it), but that second person does not exist yet, so the check is unstaffed. Read accordingly, and use the correction form: reader submissions are currently the only outside review this archive gets.
Whether AI tools were used
[ NOT YET DISCLOSED ]
This page states the other uncomfortable operational facts plainly (no funding, one anonymous editor, no second reviewer) rather than leaving a reader to guess, and tooling should be held to the same standard. Whether and how AI tools were used to research, draft, or review any entry is not yet answered here. It will carry the specific, true answer once the maintainer confirms it, not a reassuring guess standing in for one.
How an entry gets published
Drafted → sourced → second-reviewer sign-off → published with a version number
The second-reviewer step is defined but not yet staffed. The audit below reports it as an open risk rather than pretending otherwise.
How to reach a human
The correction form on every page. There is no direct address.
The form is the whole contact route, and it is not a suggestion box. It is the mechanism the archive is accountable through. Submissions land in a review queue read by one person, so expect days rather than hours. Accepted corrections are published in the corrections log with the original wording kept intact; declined ones get a recorded response. Nothing is edited silently. You do not need to identify yourself to file one, though there is a field if you want the credit.
Licence
Open data, reuse with attribution: the data, not the code
The full dataset downloads as JSON and CSV from the footer below under CC-BY-4.0. Take it, republish it, rebuild the analysis, disagree with it. The site’s source code is not published, so this is not an open-source project in the software sense and you cannot inspect how the pages are assembled. That is a real limit and it is stated rather than glossed: what you can inspect is the part the conclusions rest on, because every figure links to a primary document and the whole dataset is in your hands. If the code and the data ever disagreed, the data is what you could check.
Verify one entry yourself
An anonymous editor asking you to trust which sources count as reliable and how entries get framed is not worth much on its own — so don’t take it on trust. Below is one entry, checked start to finish against its own primary source. It takes under ten minutes and no legal or policy background.
1. Open the entry
This is what it claims, in full: "Repeal attempts failed in the Senate on July 28, 2017 by a 49-51 vote. The individual mandate penalty was later zeroed out through the 2017 tax law."
2. Follow the primary source
The entry lists 2 sources under "Sources" in the drawer that opens. The first is the chamber's own record of the vote — not a news writeup of it.
Senate roll call vote on H.R. 1628, July 28 2017 ↗3. Compare the two
Read the tally on that page. Does it match the sentence quoted in step 1? If it does, this entry checks out against its own source — which is the same claim being made for the other 96 entries.
4. See this entry get caught being wrong
This part isn't hypothetical: this exact entry was corrected on 2026-01-30. Published text read "Cited CBO’s 13-million projection without noting CBO’s later revision." — corrected to "Both the original projection and CBO’s published methodological correction are shown together." Read why below, alongside every other correction this archive has made.
Live integrity audit
These checks run against the archive’s own data every time this page loads. They are published whether they pass or fail, because an audit you only publish when it flatters you is marketing.
Entries with fewer than two source documents0 / 97Pass
Every entry should rest on at least two independently checkable documents.
Effect claims resting on a single estimate0 / 203Pass
A claim with one estimate cannot show a spread. Where a genuine second estimate does not exist, the rule is to record the absence explicitly rather than leave the claim thin. The count above is computed from the data every time this page loads, not restated by hand.
Claims where advocacy sources appear alone0Pass
A Tier 3 source must never stand unaccompanied. Any count above zero is a policy violation.
Share of effect claims labelled Direct68%Informational
Direct should be the exception, not the default. It runs high here because most claims are vote counts and court holdings — a vote was 87-12, a court held X — which genuinely are direct. Every claim about an economic or social outcome is Contributory or weaker. If this figure ever climbs while outcome claims are added, the scale has stopped doing work.
Distinct source-lean labels in use137 / 520Informational
Every estimate carries a label describing its source’s orientation. Synonym drift makes those labels useless for filtering, so pure duplicates are collapsed; compound labels that carry real information — e.g. a nonpartisan scorekeeper read as center-left by critics — are kept verbatim rather than flattened.
Weak-attribution claims written in causal language0 / 203Pass
A claim labelled Contextual or Disputed must not be written as though causation were established. This runs on every page load rather than quarterly by hand, so a claim cannot drift into causal phrasing between audits.
Controls below the WCAG AA/AAA target-size minimum0 at 1280px desktop (CI-enforced); 0 at 390px mobile (verified by hand, not yet CI-enforced)(AA is 24px; AAA is 44px)Needs work
Re-measured live with getBoundingClientRect() against every button, link, select and input in every view, not assumed from the design file. The desktop pass runs on every build and fails the build on a regression. The mobile pass came back equally clean but currently only runs as a manual check, not an automated one — a control that only fails at 390px has nothing catching it today, which is why this stays a warning rather than a pass. Two documented, intentional exceptions: the compact search-results Clear button, and the Overview standing-bar segments, whose width is itself the data being shown.
Entries lacking a second-reviewer sign-off97 / 97Open risk
Computed from the review_queue table, not restated by hand: an entry only counts as signed off once its review status reaches "published," which requires a reviewer who is not the entry's own last editor (src/app/admin/entries/[id]/reviewActions.ts). No entry has reached that state yet, so this is the real current count, not a softened one — all of them. The bigger gap this number can't close by itself: recruiting an actual second reviewer is a staffing action, tracked as an open, unstaffed item in CLAUDE.md ("Needs a person, not a PR") — this row exists so the count moves once that happens, rather than being wired to the total regardless of what the workflow reports.
Entries carrying a provisional flag7 / 97Informational
Facts still moving. Flagged in the interface rather than hidden or quietly stated as settled.
Entries with a published revision history7 / 97Needs work
Every entry should carry its full prior-version history. The count above is computed from the data, so the two cannot drift apart. The remaining 90 cannot be backfilled honestly — a revision log records real edits, and inventing dates to fill the column would break the same rule this archive applies to everyone else. Histories accrue from here once entries are edited through a real workflow.
Status refresh queue — all 97 entries
Cadence is derived, not chosen by hand: an entry flagged provisional refreshes every 14 days, anything in live litigation every 30, and settled history annually. This runs live below; a script (scripts/compute-refresh-queue.ts) can persist the same due-dates to the database, but nothing schedules it yet, and no owner is assigned to any of these entries — that stays on the open-risk list until someone owns it.
Every 14 days7
+1 more
Every 30 days35
+29 more
Every 60 days28
+22 more
Every 120 days10
+4 more
Annually17
+11 more
Review record — as of 2026-08-16
This is not independent review
These passes were run by the author against the author’s own work. They catch mechanical and internal-consistency failures, which is real but limited value. They cannot catch a shared blind spot, and they are not a substitute for the second reviewer the audit above still lists as an open risk. Recorded in full because a review you only publish when it flatters you is marketing. This applies equally to every “fairness panel” or “fairness review” pass named in the corrections log below: those are the same self-run, author-against-own-work passes described here, not a second, independent reviewer.
A concrete, dated instance of that limit: until 2026-08-14, the "January 6 and the certification of the 2020 election" entry's effects never mentioned the fake-elector scheme or the pressure campaign on Vice President Pence — a real completeness gap in one of the entries this readership will scrutinize hardest, closed only because a self-run fairness pass happened to catch it before publication, not because a staffed second reviewer was positioned to. See that row in the corrections log below. A real second reviewer is the structural fix for gaps like this one; an ad hoc pass by the same person who wrote the entry is not a substitute, which is why this stays an open risk rather than a closed one.
Methodologist — self-run, not independent
Checked
Every claim labelled Contextual or Disputed, scanned for causal verbs — "caused", "led to", "resulted in", "drove". A weak attribution label paired with causal prose is the most common way a tracker smuggles in a conclusion.
Found
0 violations across 203 claims.
Fixed
Nothing to fix. Distribution published in the audit above so the ratio stays visible as entries are added.
Media-law self-check — self-run, not independent
Checked
All narrative fields — summary, action, status detail, contested — scanned for unqualified characterizations: lied, corrupt, criminal, fraudulent, illegally, deliberately misled.
Found
Six hits, all false positives. Every one was a legal term of art in correct use: "criminal custody", "criminal justice reform", "criminal culpability was never resolved", "unlawfully present", and a court’s own finding that a US Attorney was "unlawfully appointed".
Fixed
Nothing to fix. The archive states findings and attributes characterizations, which is both the editorial rule and the legal defense.
Fact-checker — self-run, not independent
Checked
The most volatile current-status claims, re-verified against primary and practitioner sources rather than against the archive’s own earlier text.
Found
Two real errors in the tariff entry: the termination date was wrong, and refund exposure stood as a single practitioner figure where the archive’s rules require a spread.
Fixed
Date corrected to February 22, 2026 with the CBP notice number; refund exposure replaced with a sourced spread including the dissent’s argument against full refunds. Both logged in Corrections.
Copy editor — self-run, not independent
Checked
Spelling consistency across the whole document, and the controlled vocabulary of source-lean labels.
Found
British and American spellings mixed almost evenly — 22 "programme" against 22 "program" — and 151 distinct lean labels where many were pure synonyms.
Fixed
36 spellings unified to American, preserving official titles that are British by name. Synonym labels collapsed to a controlled vocabulary, keeping every compound label that carries real information; the live label count is in the integrity audit below.
Hostile reader, from the right — self-run, not independent
Checked
Whether a supporter could reasonably call this a hit piece.
Found
The strongest defense is structural and already present: no scores, no rankings, colour marks legal standing rather than approval, and the record includes Warp Speed, the First Step Act, USMCA, the 92% fall in border apprehensions, and a net-neutrality entry where the critics’ predictions failed.
Fixed
No change. Aiming the site at this reader is what would destroy it — the Commonly misread view exists to make the two-directional evidence findable in one place.
Hostile reader, from the left — self-run, not independent
Checked
Whether the archive sanitizes by refusing to draw conclusions.
Found
A fair charge in one place: the press-access entry documents a pattern it declines to name, because no adjudication supports the connective conclusion.
Fixed
No change, and the entry says so explicitly rather than leaving the silence unexplained. Recording a pattern and asserting a motive are different acts, and only one of them is evidence.
Submit a correction
Every submission gets an answer, including the declined ones, with the reason on the record. A correction needs the specific claim, a document, and proposed wording — disagreement with framing is welcome but is handled differently.