2026-09-04 - (/loop) One question, written a hundred and forty-nine ways

719 questions / 275 premises / 40 shared   ->   713 / 135 / 27
suites 126   clippy/fmt/identity clean on both repos

Twenty-fifth cron tick, working the named axis: audit the 40 shared premises for the D537 split - is this a question for a console, or a statement about orbistoun?

The big one was invisible to the grouping

D538 warned that 275 was a floor, because grouping is word-for-word and a premise written two ways counts twice. The delegation question was written 149 ways: libScePosix's POSIX spellings each resolve to the vendor-named function beside them, and every entry asked whether the two behave alike with the target's name inside the question. A fifth of the ask list.

Two things in one sentence again. Resolves to X, and this describes that function is a fact about what orbistoun does, and for 122 of the 149 the question was the only place the delegate was named - so this had to be a move into edge_cases, not a deletion. The remainder, the two spellings are assumed to behave alike, needs no name: the entry is the subject. Split, they are one premise across 149 functions.

Fixed at the generator, not at the data

122 of them were generated - orbistoun-gen knowledge built the sentence with format! and interpolated the target, so editing the TOML alone would have let the next new delegation put it straight back. The wording is now orbistoun_hle::knowledge::DELEGATION_ASSUMPTION and the generator pushes that constant.

the_delegation_question_is_asked_in_one_wording gates the rest, made to fail by reverting one entry to the old form.

The heaviest misfiling, sharpened rather than deleted

24 entries, 101,511 calls: "The failure return is this project's placeholder rather than an errno." The placeholder is a settled choice - it avoids the high bit precisely so it can never be mistaken for a real value. No console can confirm it.

Replaced with the question one step along: which convention the platform's POSIX-named exports use on failure - POSIX's own, or the 0x8002_0000 | errno encoding D398 measured on their vendor-named twins. One provoked failure on hardware answers it. The count did not drop, which is the point: this axis is about the list saying true things, not about shrinking it.

Six that were facts pretending to be questions

strtod/strtof - "which is what the standard specifies", in entries that are published and cite ISO C 7.22. And four scePthreadMutexattr* entries carrying a width that D272 established by watching a whole-word write clobber a loop counter; libSceSystemService already files the same fact as an edge case.

What is not done

27 shared premises remain. The next defect down is a different one: cross-references that do not stand alone - "Same delivery caveat as posix_sigemptyset.", "As _open." - readable in the file, useless in a queue handed to a console.

Nothing here moves the wall. Third tick of one finding at three levels: D537 (records said nothing), D538 (the list could not see it repeated itself), D539 (the repetition, at its source).

Decision: D539.