Some SEC submissions files list acceptance times 4 or 5 hours late under a UTC label

The check's reference is the Eastern "Accepted" time on each filing's index page, which matches the filing's SGML header ^[4]. Against that reference, the shifted value is the true UTC time plus one Eastern offset ^[5]. JPMorgan's results 8-K was accepted at 06:30 Eastern, which is 10:30 UTC, and its file lists 14:30Z, after the open instead of before it ^[6]. The gap is four hours, the daylight-time offset ^[19].

The same filing can carry two times. A Form 4 sits in both the company's file and the insider's file. For Apple's director Form 4s from 26 February, the directors' files are right and Apple's file is five hours late (February is standard time) ^[7]. In the post's worked example, Apple's file lists 2026-02-27T04:34:19.000Z for a filing whose index page says 2026-02-26 18:34:19 Eastern ^[8]. The true UTC time is 23:34:19 on 26 February. A pipeline that takes its calendar date from the stamp files the filing under 27 February ^[20].

Which files are wrong is not settled. According to the post, the deciding factor is the date of a file's most recent filing. Files whose latest filing was accepted on 2 October or earlier were correct, and files whose latest filing came on 5 October or later were shifted, including for their February filings ^[9]. The author's guess is that files rebuilt since early October are written with the extra offset ^[22]. "I can't see rebuild times (the API sends no Last-Modified header), so that is a hypothesis, not a finding," the author wrote ^[23]. The edgartools issue dgunning/edgartools#1501 had logged Apple's times running four hours late on 8 October, and on 7 October it logged some files carrying Eastern times with a UTC label ^[10].

The values in the post are those of 11 October, and the files change ^[16]. A copy downloaded before a rebuild can differ from one downloaded after ^[15]. The post does not say how many of the 25 files were off, and it reports no SEC statement on the cause.

The post's fix has two tiers. Where a filing's time matters, take it from the index page or the SGML header, which costs a request for each filing ^[11]. Since every filing in a file is off by the same amount, a single request per file will do: check one filing to find that file's error, then apply it to the others ^[12]. The post's file_shift helper accepts shifts of -1, 0 or 1 and raises ValueError on anything else ^[13]. The correction function tries 4 hours, then 5, and keeps the one that matches Eastern's offset at the corrected instant ^[14]. We think that is the right design. A flat four-hour subtraction would leave every standard-time filing one hour off ^[21].

The post advises calibrating on two filings from different seasons, falling back to index pages if they disagree, and recalibrating on every re-download ^[15]. canli-mcp's earnings_calendar already read times from index pages. Its event_study did not ^[17].