The Standard Applies Here First
Corrections
This platform judges public officials against the record, so its own record has to be correctable in public. From launch forward, every substantive fix to a published figure, claim, or method is logged here, newest first, with what changed and how it was caught. Errors found by readers come in through the report form; errors found by our own scheduled hostile review get logged all the same. The ledger opens with one honest entry about everything that happened before you could see it.
Before launch, this platform was deliberately run through the harshest review it could arrange: five full passes by an outside AI reviewer, an independent pass by a second one, and the project's own automated hostile-review agent attacking a page at a time. Roughly forty substantive corrections came out of those cycles and were applied before any public reader saw the affected pages: a deficit figure updated to the enacted-law CBO estimate, an unsupported employment claim removed, causal language softened to match what studies actually show, a program cost restated from "tens of billions" to the honest $100-billion-plus arithmetic, evidence cards split so every claim links its exact source, and overclaims about household savings restated as design requirements to be demonstrated rather than results to be asserted. Findings that turned out to be wrong were checked and rejected rather than rubber-stamped, and one correction entry itself had to be corrected. The full, unabridged pre-launch log is preserved in the project's version history.
How it was caught: Recorded here as one entry so the public ledger below starts at launch. The point of the pre-launch gauntlet was simple: fix it before you see it, and keep the habit after you do.
Pre-launch archive: the itemized log (17 entries)
Seventh review pass, the pre-launch closer. The service plank stopped calling an incentive-backed program a mandate: "required term," "universal obligation," and "legal mandate" became "universal expectation backed by substantial incentives," with the equity risk stated (benefit-linked pressure lands hardest on those who need the benefits most) and answered with income-blind benefits, participation published by income band, and a class-skew tripwire. The Forces influence rubric was rerun against every remaining high card in one sweep: six lowered to medium for lacking a demonstrated effect (Heritage, API, AFP Action, PhRMA, the American Hospital Association, and NAR's housing-support card), and the five that kept high now state the concrete outcome that earns it on the card (a bill passed the House, a bill steered through Congress, a rule set aside in court, a concluded nationwide settlement, a court-ordered stay). One copy edit: "design is designed" deduplicated.
Outside review, seventh pass; the terminology point was conceded as philosophically correct, and the rubric sweep was completed across all eleven high cards rather than only the flagged ones.
Sixth review pass. The national-service brief briefly contained two incompatible enforcement models (a jury-duty-style civil process added in round five, against the brief's standing no-fine, benefits-only commitment); resolved in favor of the standing design, now stated as one coherent model: a legal universal expectation whose entire cost of refusal is the attached benefits, never a fine or criminal record. Five Forces high-influence ratings that lacked the demonstrated national effect the published rubric requires (Ossoff, Brennan Center, ACLU, America PAC/Musk, AFL-CIO) were lowered to medium with the basis stated on each card, matching the earlier Club for Growth and YIMBY re-ratings. Two design claims were restated as designs rather than results: fee-and-dividend is now "designed to be progressive in net terms," and the innovation pillar is "designed to be largely self-financing," with a new unscored Ledger row carrying that claim as intent. And this page itself was fixed: compressing the pre-launch log into one summary broke the About page's promise that every fix is logged publicly, so the itemized entries were restored beneath the summary as this archive.
Outside review, sixth pass; the enforcement contradiction was between two of our own passages, which is exactly the inconsistency an opponent would quote side by side.
Fifth outside-review pass, every claim verified before editing. The national-service brief called a four-million-participant program "tens of billions annually" when its own stipend and award levels imply at least ~$100 billion before benefits and administration; the arithmetic is now stated plainly in the brief and as a bound on the Ledger row. The $17.30-per-dollar service return now carries its two limits (combined social benefits, not federal savings; evaluated programs, not universal scale), and "pays for itself several times over" was replaced with the claim the evidence supports. The AI-displacement framing was corrected to occupational churn using the WEF and IMF figures already in the evidence base (92M displaced / 170M created; 60% exposure with roughly half complemented), which the prose had overstated. The word mandate is now defined: incentive-linked during build-out, jury-duty-style civil enforcement at maturity, never criminal. The Government brief still used the FY2024 $1.8 trillion tax-expenditure figure four times after the site standardized on $2.2T (FY2025); all corrected. The Hard Questions carbon-dividend answer made the same overclaim the healthcare answer once made ("most families receive more than they pay"); it now states the distributional promise as a requirement to be demonstrated in household-incidence tables. Ten Forces cards in the editorial core lacked the counterarguments a prior correction claimed all cards had (see the amended entry below); all ten added, and Club for Growth Action was lowered from high to medium influence because its basis showed financial capacity without demonstrated effect, which the published rubric requires for high. The Outcomes page no longer says an indirect measure moving means "the real thing is moving with it" (a proxy cannot guarantee that; signals are now read alongside direct measures and competing explanations), and a missed marker now triggers diagnosis and possible redesign rather than automatically "push harder," which had made the framework unfalsifiable. The Georgia Pilot received 13 sentence-level source links to the exact reports behind its claims. A template bug rendering "figure as of as of" on some Evidence rows was fixed, and the Forces last-updated stamp was corrected. Entry dates on this page are UTC, which is why a late-evening US fix can carry the next day's date.
Outside AI review, fifth pass; the service-cost arithmetic was checked by hand, the WEF/IMF framing was corrected against rows already in our own evidence base, and the Forces undercount was verified card by card before the prior entry was amended.
Eight fixes from the platform's own automated hostile-review agent, run against the Ledger and triaged with each claim re-verified before editing. The $190/ton social cost of carbon now states its status plainly: EPA's December 2023 peer-reviewed estimate, withdrawn from government use by the current administration in January 2025, carried as a central point in a contested range with the honest note that a lower official value shrinks the revenue and damage math. The $1.01 trillion caregiving figure is now labeled as AARP's revised 2026 valuation, nearly double the prior ~$600 billion edition because the counted caregiver population and hourly replacement value both rose (the agent suspected the figure; verification confirmed it is real, so the fix is framing, not the number). The tax-gap passage no longer calls $606 billion "the largest pot of no-new-taxes money": it now says the gap measures the problem's size, points to the scored ~$127B/decade enforcement yield the table actually books, and attributes the top-of-the-distribution concentration to the separate Treasury-affiliated research rather than to the IRS gap publication, which does not make that claim. The CBO single-payer savings sentence now names its load-bearing assumption (Medicare-adjacent provider payment rates). The IMF $7 trillion figure is relabeled fossil fuel underpricing, with the ~18% explicit-subsidy share shown separately. The $2.4T-vs-$600B gap paragraph now says the gap is a floor, not a ceiling, since older-window rows understate nominal costs and the healthcare shift sits on top. And the single-worker hard case graduated from diagnosis to commitment, here and in the healthcare brief: the default structure is a progressive schedule with an exempted first wage band, and no design advances that leaves below-median wage earners paying more than today.
Findings from the project's scheduled red-team agent (run on request); the two factual suspicions (AARP figure, SCC status) were independently verified before any edit, and one of them turned out to be wrong in the agent's favor.
Third-pass items from the same outside review. Hard Questions asserted that "for most families the tax is smaller than the premium," contradicting the healthcare brief's more careful formulation; it now states the household promise as a design requirement to be demonstrated in independently reviewed tables, with version zero published in The Ledger. The Evidence page claimed all sources were accessed June 26, 2026, which was impossible given sources added and re-pinned in July; the statement now distinguishes the core library review date from later additions recorded in this log. The tracker's mirrored chart implied its two bars were comparable magnitudes when one is an average and the other a maximum; the axis labels now name each side's rule (platform breadth vs. opposition threat) and the methodology states plainly that the bar lengths are not comparable. The Consensus page added direct survey links to every row, separated problem-salience from policy support in the leadership row, and labeled the one row (AI) whose survey link is still pending verification as directional context rather than a pinned claim. Separately, a second citation pass added 19 sentence-level footnotes across seven briefs, bringing the total to 60.
Each finding verified against the live page before the fix; polling links drawn from the verified evidence base, with two advocacy-adjacent sources flagged in their link text.
Four bundled or under-linked Georgia evidence cards were split and re-pinned to exact reports: the data-center exemption card now links the DOAA/Carl Vinson evaluation itself rather than the OPB homepage; the Ossoff card, which made three claims on one homepage link, is now three cards pinned to the Senate committee record, the Congress.gov bill page, and the GAO report; the poultry card is split so USDA figures link USDA and the contract-structure description links the (flagged) Farm Aid fact sheet; and the accountability-courts card is split into the state audit (recidivism, with its selection caveat in the claim itself) and the Carl Vinson savings study. On the Forces page: the Elon Musk and America PAC entries described the same $157M financing event and were counted as two actions; merged into one card so the tally does not double-count. YIMBY Action's influence was lowered from high to medium under the published rubric, since organizational reach is documented but a national policy effect is not. Every Forces card in the agent-added set received a stated counterargument at that point. Correction to this correction, caught by the next outside review: ten cards in the editorial core file were missed by that pass, so the claim as originally written here overstated the coverage; those ten received their counterarguments on July 21 and the entry is amended rather than silently rewritten.
Remaining items from the outside review's evidence and Forces sections; each replacement URL verified before the edit.
The tracker said CBO estimated "~11 million" people lose coverage under the OBBBA. That was CBO's estimate for the House-passed bill (10.9 million); its estimate for the enacted law is 10 million more uninsured by 2034, and the tracker now uses the enacted-law figure with the distinction stated. Separately, the Hard Questions "how will you pay for it" answer was restructured: it previously led with the large gross pots (unpaid taxes, tax preferences, improper payments) that the Ledger itself says are not recoverable pools. It now leads with the Ledger's scorekeeper-anchored revenue rows and names the gross pots as evidence of system leakage rather than as funding.
Second-pass items from the same outside review; the coverage figure was re-verified against CBO's enacted-law publication before the fix.
A second outside review prompted seven corrections, each re-verified against the primary source before editing. The Tracker's deficit figure for Public Law 119-21 was corrected from ~$2.8 trillion to CBO's enacted-law estimate of ~$3.4 trillion over 2025-2034 (about $4.1 trillion with debt service). The tax-expenditure total was standardized at ~$2.2 trillion (FY2025) across the Ledger and Hard Questions, matching the Evidence page, since the page had mixed the FY2024 ($1.8T) and FY2025 ($2.2T) figures without noting the years. The Georgia Pilot's claim that "60 percent of the state's agricultural jobs depend on" H-2A was removed as unsupported by its source, which cautions that certifications are seasonal slots (averaging ~5 months), not year-round jobs; the defensible figures (second-largest user, 43,000+ certified positions) were kept. Sen. Tim Scott's Forces card was moved from Pillar 3 to Pillar 4, where the ROAD to Housing Act belongs. Three causal-language softenings: the Georgia diversion section no longer says the model is "proven" (graduate-vs-nonparticipant comparisons are selection-prone; now labeled "promising"); the Georgia maternal section now states plainly that the "a ban costs lives" conclusion is the platform's inference, not a formal committee finding, while keeping the committee's actual finding about Thurman's preventable death; and the healthcare brief's "families pay less" line is reframed as a design requirement to be demonstrated in household tables, not a proven result, with the "nearly every wealthy country" single-payer claim corrected to reflect that peer countries use varied universal systems.
Outside AI-assisted review; each factual claim checked against CBO, JCT, USDA, and the cited committee records before the fix, and language tightened where a correlation had been stated as causation.
Seven factual corrections from an outside review, each verified before editing: the Larson redistricting announcement was misdated November 18 (source says September 18, 2025); the NAR settlement card wrongly attributed the earlier 2024 Sitzer/Burnett practice changes to the 2026 settlement, which NAR's own announcement says imposed no additional practice changes (card rewritten); API's energy agenda is for 2026, not 2025, and its seven-figure ad campaign now carries its own Axios citation; the BPC explainer date corrected to its stated February 10, 2026 update; the ABA card now describes the reproposal statement its source actually documents rather than appellate briefs; the AHA card now dates its site-neutral litigation to the 2018-2021 record; and the Heritage card's "template later enacted in H.R. 1" causal claim was softened to documented alignment. Also: Tim Scott reclassified to working-in-line per his documented action, America PAC raised to high influence per the published rubric, PAC cards gained non-FEC second sources for political purpose, and category counts now say "actions" with a unique-actor tally.
Outside AI-assisted review checked each card's source against its claims; every finding re-verified here before the fix.
Two Georgia rows were removed because they carried unresolved internal verification flags on the live page ("Add PSC fact sheet as primary source"; "NOT re-verified this pass"), violating the publish-only-verified rule; one was already superseded by the GAO institutional-investor row. The Right to Vote Act row was re-pinned from a senator's press release to the official Congress.gov bill record. New rule enforced: nothing enters the public table with an outstanding verification note.
An outside AI-assisted review spotted the internal notes rendering publicly; we agreed it was the most serious finding.
Causal language tightened to match what the data can support, with exact text logged. Old: "No confound yet advanced explains both halves away." New: "Largest identifiable driver, not sole cause: that is the claim, and it is enough." Old: "the reversal arrived on schedule when the credit, and little else, expired." New: attribution quantified via the National Academies (2M+ children lifted out of poverty by the 2021 credits) and Columbia's Poverty Center, with competing explanations named and the natural-experiment term explicitly limited ("not a randomized trial"). The section "Run the experiment, both directions" was retitled, and "experiments rather than anecdotes" became "evidence rather than anecdotes." PDFs and downloadable data regenerate automatically with each deploy, so all formats carry the fix.
Outside review found the prior correction had softened but not resolved the overclaim; second pass applied the recommended standard in full.
Upgraded to v1.1: the child-allowance row now links the JCT scoring document directly instead of a secondary explainer; the Social Security cap row is marked as dedicated trust-fund revenue excluded from general-fund comparisons; the paid-leave comparison uses the net figure after its own payroll tax; and the summary states that differing score windows make totals order-of-magnitude reading, never a precise sum (roughly $2.4T general-fund spending vs. $0.6T general-fund revenue at central values).
Outside review flagged noncomparable windows, gross/net mixing, and a secondary citation; all three accepted.
Four fixes from our own hostile-review process: the hunger figures were retold honestly (the 2021 low of 12.5% was only about a point below the 2019 baseline; the sharp move was the one-year jump to 17.3% after supports lapsed, with the later drift to 18.4% not statistically significant year to year); the food-security survey termination timeline was corrected (announced September 2025, before the final report published that December, not after); the essay now cites the contested Child Tax Credit employment-effects debate instead of claiming no confound exists; and Operation Warp Speed is credited by its actual mechanism, to both parties.
Caught by the weekly automated hostile review (issue #15); verified against USDA and reporting timelines.
Stance labels renamed to "With/Against/Mixed (key votes)" and the small denominator stated prominently: a senator's stance currently rests on one recorded vote and a representative's on at most two. The claim that "nothing here is opinion or estimate" was removed; the records are official, but the selection of key votes and the grading rules are editorial judgments, now stated as such.
Raised by an outside AI-assisted review; we agreed.
The methodology note now discloses that the two sides of the chart are aggregated differently (our progress is averaged; opposition is scored by its furthest-along measure) and that one enacted law can drive the opposition bar across multiple pillars.
Raised by an outside AI-assisted review; the asymmetry was a deliberate choice that had not been disclosed.
The air-pollution mortality figure was calibrated from a high-end 250,000 deaths per year to approximately 200,000 (MIT), with the range noted.
Internal verification pass against primary sources.
The lineage of universal voter registration proposals was corrected to President Carter's March 1977 election reform message (previously attributed to a different administration), pinned to the American Presidency Project archive.
Internal verification pass; the original claim could not be sourced.
See something else that needs a fix? Report an inaccuracy.