The office that compiles the Code of Federal Regulations keeps a public log of its own editorial errors. We read all 3,410: it fixes them in weeks now, not the months it once took — and one title holds nearly a fifth.

The Office of the Federal Register compiles every federal agency's rules into the Code of Federal Regulations — and keeps a public log of the editorial errors it makes doing so. The eCFR Corrections Compilation records mistakes the office itself introduced while processing amendments into the code; errors made by the issuing agencies are excluded and are fixed through the ordinary rulemaking process. It is, in effect, the government's typo file on itself. We pulled the complete log through the eCFR's public API. It holds 3,410 corrections, each published on a date between June 1, 2005 and August 18, 2026, spread across 46 of the CFR's 50 titles. (2026 is a partial year — 48 corrections through August 18.) They are not spread evenly. Title 40, Protection of Environment, accounts for 640 of the 3,410 — 18.8%, nearly one in five, and more than double the next title, Federal Acquisition, at 283. The four most-corrected titles — Environment, Federal Acquisition, Transportation, and Wildlife and Fisheries — hold 39.5% between them; seven of the 46 titles hold half. Most are fixed quickly. Half of all corrections were published within 74 days of the amendment that carried the error, and 1,196 of them — about a third — within 30 days. And the fix time has fallen: the median gap ran about five months (158 days) for errors corrected in 2005 and about three weeks (23 days) for those corrected in 2024. The exceptions are old. Some 281 errors sat in the code more than a year before the office caught them, and 54 sat more than five years; the longest-lived traced to a 1976 amendment to the income-tax title's estimated-tax rule (26 CFR 1.6654-2), corrected in 2008 — 31.6 years later. But 47 of those 54 five-year-plus errors were themselves cleared between 2005 and 2010, when the electronic code was first assembled.

The following is analysis, not fact. Any single correction is a shrug — a subsection re-lettered, a table reset, a dead link restored ("Table corrected" and "Link added" each appear about 50 times in the log). Reading all 3,410 at once surfaces three things no single entry shows. First, concentration: one title, Protection of Environment, carries nearly a fifth of every processing error the office admits to. Second, a tail that has vanished — the decades-long errors are almost all artifacts of the mid-2000s, when the paper code was digitized and a backlog of old mistakes surfaced and was cleared at once; nothing corrected since 2010 comes close to them. Third, a speed-up that held: the median fix stepped down from months to weeks and stayed there. This is the rare government record that grades its own homework in public. It measures the compiler, not the compiled — how cleanly rules are transcribed into the code, not the substance of the rules themselves.

The count is not an error-rate ranking. Title 40 almost certainly leads because it is among the largest and most frequently amended titles — more amendments to process means more chances to mis-process one — and this log carries no denominator of total amendments per title, so it cannot show the office is careless with environmental rules, only that it corrects them most often ^sspeculative. The falling median is partly composition, not purely diligence. The 2005–2010 corrections swept in a backlog of long-buried errors inherited when the code went electronic, inflating the early-year gaps; strip that startup effect and the improvement narrows ^sspeculative. We report the observed decline, not a productivity verdict. And the scope is narrow by design. This is the Federal Register office's own editorial slips only. Substantive errors made by agencies — the ones that change what a rule means — are corrected through the normal amendment process and never appear here. Nothing in this log measures how error-prone federal regulations themselves are.