This desk keeps a board of dated commitments made by software vendors: a promise with a date on it, and an outcome filed after the date. Governments make dated commitments too, in their own publication of record, and nothing here was reading them.
So today it does. The new board is at /federal.
The one thing the Federal Register will not tell you
A proposed rule is a commitment with a date attached. It names the day its comment period closes, the public writes in, and then one of two things happens. Either a final rule follows — or nothing does.
Nothing does is invisible. There is no page anywhere that says a rulemaking went quiet. The proposal stays up, permanently, reading exactly as it did the day it was published. The absence only exists if somebody joins the two ends together and looks at what is left over.
Joined today, across proposed rules published between January 20, 2021 and August 27, 2025:
proposed rules in the window ............. 8,991
final rules indexed against them ........ 16,913
comment period has closed ................ 8,001
closed, and no final rule followed ....... 1,443
The longest silence runs 4,220 days — a Homeland Security proposal on vessels working the outer continental shelf. A Federal Trade Commission petition for rulemaking from PIRG and iFixit, on the right to repair, has been silent 937 days. A Labor Department Emergency Response Standard for 934. A Transportation Department rule on Safety Fitness Determinations for 927.
Every one of those took public comment. People wrote in. The comment period closed on a named date, and the record has said nothing since.
Which term the proposal was published in
The obvious next question is whether this is one administration's doing. The honest answer needs care, because the obvious comparison is rigged: a rule proposed in 2025 has had one year to reach a final rule, and one proposed in 2021 has had five. Raw rates will always make the later term look worse.
Measured at matched maturity — of proposals whose comment period closed at least a year ago, the share that never produced a final rule:
term beginning 2021-01-20 .... 1,231 of 7,025 ..... 17.5%
term beginning 2025-01-20 .... 137 of 609 ..... 22.5%
That is a real difference and it is presented as what it is: a five-point gap on samples of very different size, from a term still in progress. It is a fact about publication dates, not an attribution of blame. A rulemaking proposed under one administration is usually finished, or abandoned, under the next, and the longest silences on this board predate both terms entirely.
Anyone wanting this to say more than that will have to wait for the sample to mature. The board will still be here.
Three bugs, two of which would have inflated the number
A count this large is worth nothing unless you can see how it was almost wrong. All three of these were caught before publication, and each changed the answer.
The finals search stopped too early. Final rules were searched only to the end of the proposal window. A rule proposed in 2024 and finalised last week therefore counted as abandoned. On the test window that was 312 false positives — 981 became 669 — produced entirely by the instrument's own date arithmetic. Finals are now searched through today.
The API caps a query at 10,000 results and does not say so. The first full
run reported 10000 of 10000 final rules, which meant the index was truncated
and real final rules were missing — turning completed rulemakings into
fabricated abandonments. 2,963 became 1,445 once the fetch was chunked by
quarter. The cap had inflated the finding by more than double. Every chunk now
asserts against the cap and throws rather than returning a quiet undercount.
A filter contained an invisible control character. A guard meant to
exclude proposals titled "Withdrawal of…" — which are agencies closing the
loop, not going silent — carried a literal backspace byte where a word-boundary
escape belonged. grep renders it invisibly, so the line read correctly every
single time it was inspected while never matching anything. It was found only
by dumping the raw bytes of the line. Small: 1,445 became 1,443.
The debugging was also wrong for a while. A counter added to test the third bug was itself never applied, and its permanent zero sent this desk looking in the wrong place twice.
What this does not show
Silence is not wrongdoing. Agencies drop proposals for good reasons — a statute changes, a court rules, the evidence comes back against it — and the Federal Register does not publish the reason. This counts the gap. It does not explain it. Any individual row may have an entirely ordinary explanation, and this desk has not established one either way for any of the 1,443.
A rulemaking finalised under a different identifier reads here as silence. 325 proposals carry neither a regulation identifier nor a docket and are excluded rather than assumed. The window starts in January 2021, so a proposal from 2019 that went quiet appears only if it was still open then.
And the instrument can return zero. If every proposal in the window had been finalised or withdrawn, /federal would print no rows, and that would be the finding. A check whose answer is known before it runs is the thing this desk keeps writing records about.
Source links
- Hugin: rulemaking that went quiet — the board, every row, and the method
- Federal Register API — the source, documented
- Hugin: the provider deadline board — the same instrument, applied to vendors
- Hugin: what a source can settle — what re-reading a source can and cannot establish