The export summed case counts across three aggregation levels at once — the national total, the circuit subtotals, and the districts — which inflated the national figure roughly threefold. The independent review flagged that the numbers didn't reconcile. Corrected to aggregate over the real districts only: the true total is 268,242 civil cases, and on the live page the 15 case categories sum exactly to that published total (the export additionally cross-checks it against the per-district totals).
What we know is wrong, limited, or pending
The record of what Atlas Commons has caught, what it limits, and what it has not yet built — the page to check before citing a figure.
A source-backed institution is only as trustworthy as the issues it admits. This is the running record of what Atlas Commons knows is wrong, limited, or unfinished about its own data — corrections caught by independent review and corrected, named limits shipped on the page, and pending gaps with a known fix. Every item links to the live surface it affects. It is the published record the Public Knowledge Charter commits the institution to (Article 8).
5 caught by independent review and corrected · 5 named limits shipped on the page · 2 pending. The honesty is the point: we publish, review, and correct in the open — an institution that hides its corrections cannot be verified.
Caught in review
An independent review flagged it and it was corrected — the live surface now shows the corrected state. We publish, review, and correct in the open; we do not claim the first version is always right.
The page showed an insurer's three-year aggregate premium next to a rebate that is computed on the current-year base — so the displayed premium read about three times the figure the rebate implied. Corrected to show the current-year premium (the same year as the rebate), which reconciles; the national premium figure was likewise corrected from a $1.63T three-year aggregate to the $594B annual figure.
A hero stat read “769 Medicare contracts rated,” but the 3.65-star average was computed over the 516 contracts that carry an overall rating, not all 769 — two adjacent big numbers implied one population. Relabeled “769 contracts (516 with an overall rating)”; a percentile that could read “higher than 100%” for a 5-star plan was also corrected to strictly-below.
Article 3 (evidence classes) was first badged Enforced, but the four-status warrant labeling is not yet uniform across every page, and two of the article's five named classes (personalized estimate, commercial option) are not instantiated at all. The review held the institution to its own evidence-class discipline; re-badged Partial.
The worked example warranted the $762 Silver-plan median as MODELED, which contradicted both the live Coverage page (which records the underlying premiums) and the standard's own rule that an aggregate over recorded figures is DERIVED. Corrected to “the DERIVED median over the RECORDED CMS Rate PUF premiums” — recorded rates, derived median.
Known limit
Shipped, with the limit named on the page — read it before citing the figure.
An MLR is a backward-looking measure of how an insurer spent premium across its whole book in a market and state — not a measure of any individual plan's quality, your denial odds, or whether a plan is right for you, and not advice.
A star rating summarizes a contract's measured quality across all its enrollees — not a guarantee of your experience, not a statement that a plan is sold where you live. Per-parent averages are unweighted, so a parent with a few small high-rated contracts can outrank a larger insurer most members actually experience.
Premiums are sticker prices before income-based tax credits; most marketplace enrollees pay far less, and net cost is never computed here. Scope is the federally-facilitated-marketplace states; state-based-exchange states are labeled as such, never shown as $0.
These are counts of cases filed (commenced), not outcomes — they do not measure who won, the merit of any claim, caseload per judge, or how a specific case would fare, and they are not legal advice. One statistical year.
The figure is the general statutory deadline, not your personal deadline — accrual, tolling, the discovery rule, repose, and government-notice rules are fact-specific. The surface routes hard to counsel and never computes a person's actual deadline.
Pending
A named gap with a known fix, not yet shipped.
This ledger is the published correction record. The open external-submission path Article 8 commits to — a public form, a tracked status, and a published reason when a correction is declined — is the next step on this surface, and is not yet built.
The atlases are published on workers.dev hosts; cross-atlas discovery links are deferred until the institution secures real domains and every atlas clears the quality bar. The canonical hosts will redirect to the institutional archive under the Charter's continuity article.
For agents. This ledger is machine-readable at /corrections.jsonld
(and as /corrections.md) — each issue with its surface, scope, status
(caught / limit / pending), and the live proof URL. Check it before treating any Atlas Commons figure as authoritative.
Agentic University · corrections ledger · the published record of Charter Article 8 · machine-readable at /corrections.jsonld · /corrections.md