How it works What it refuses Who it is for

Five readings of the public record.

All five written 4 August 2026 · Jayesh Sahasi

Analysis drawn from filings I fetched and checked myself: 102 public documents from eight commissions, each held with its source URL, its docket and its hash. Every quote names the file it came from and links to the commission's own copy.

So a reading you doubt can be checked against the record rather than against me. Where a count rests on a judgement I made, the piece says which judgement and where the line was drawn.

The pieces

  1. The Kollen correction moved 127 characters. The filing had already contradicted itself, 223 pages later.

    The 127 characters are not the correction, and the mistake they fix was readable on the day the testimony went in: the exhibit attached to it names a different plant.

    Kentucky PSC 2025-00113 · direct testimony of 9 September 2025 and its correction of 30 September

  2. Three of thirty-seven corrections were wording. The rest read perfectly well in place.

    Eleven of the thirty-seven fix a number that was simply wrong, three fix spelling or grammar, and nine could have been caught from the one document without opening a second.

    Sixteen correction filings · nine dockets · seven commissions

  3. Approved as modified: the language you have to file exists in no filing

    The term that governs the Ohio compliance tariff appears once in a 99-page order and never in the stipulation that order modifies, so the text a company must file exists only as the difference between two documents.

    Ohio 24-508-EL-ATA · Georgia 44280

  4. The changes tie. The baselines move.

    Across nine large-load reports every change the filer states reconciles with the period before it, while the fixed point the growth is measured from moves between Q3 and Q4 and nothing in the report says so.

    Georgia 55378 · NCUC E-100 Sub 208A

  5. Two settlements, thirteen days apart: the boilerplate matched 92%, the terms 26%

    Two stipulations filed thirteen days apart to settle the same case share 92 per cent of their procedural language and 26 per cent of their operative terms, and eleven of the forty distinct numbers appear in both.

    PUCO 24-508-EL-ATA · stipulations of 10 and 23 October 2024

Why these are here

Each was found with difflib and a text file, one pair of documents at a time. That is cheap, and anybody carrying a docket can do it on Monday morning; each piece ends by saying how.

Verbatim is what the same work becomes when forty dockets are open at once and every answer has to carry the line it came from — or refuse to be an answer.

See a proceeding move through it

Three versions of one docket, the changes between them, and a claim that refuses to assert itself when its citation does not check out.