412. The whole corpus was twelve days stale

2026-09-04 - directed

What was done

Re-ran the six titles whose records predated D563's answered measurement, honestly - learned.toml set aside - and at each record's own time limit, so the comparison is against like.

Five of the six had improved enormously, and nobody had looked.

Title	Recorded	Now	Reach
PPSA03416-app0	23 imports	188 (174 answered)	entered → flipped
PPSA25872-app0	14	129 (118 answered)	entered
PPSA04263-app0	4	70 (63 answered)	entered
PPSA21564-app0	13	57 (56 answered)	entered
obscene	178	193 (180 answered)	entered → flipped
PPSA28061-app0	47	25 - not recorded	entered

Three titles now reach flipped where this morning none did, and the ladder rung added in D558 turns out to describe a third of the corpus rather than one title.

Those records were dated 2026-08-23. This is D555's finding - the wall was real and the record was twelve days stale - repeated across the whole corpus rather than one entry. The work had happened; nothing had re-measured it.

PPSA28061 went backwards, and it is not today's doing

Twenty-five imports against a recorded forty-seven, deterministically - two runs identical. It was worth ruling out, and the trace does: it calls none of the functions implemented today. No setjmp, no setcancelstate, no sigprocmask, and it touches neither the event queues nor video-out at all.

Where it stops is libSceUlt - the user-level threading library, five functions, all unimplemented - and then the Agc calls. sceAgcDriverRegisterDefaultOwner and sceAgcCreateShader both answer the placeholder and the guest calls abort, after which it hands that placeholder to malloc as a size.

So the old entry is a claim from a build twelve days ago that the current one does not reproduce, and the cause is somewhere in those twelve days rather than in today.

The gap that leaves, which is the mirror of D563's

A compatibility record can only move up. beats records an improvement and refuses everything else, so a title that gets worse keeps its old entry and the record quietly overstates what the current build does. Nothing notices, because nothing is looking for it - the same shape as the standing blindness fixed this morning, pointing the other way.

PPSA28061's row now says 2026-08-23 next to six rows saying 2026-09-04, which is the only signal there is, and it is one a person has to spot. Recorded here rather than fixed, because a "best ever" record and a "what it does now" record are two different things and choosing between them is a design decision, not a bug fix.

Surprise

PPSA25872 reaches 129 imports at 3% standing - 11.5 million calls, and ninety-seven percent of them on placeholders. It is the clearest case yet of why standing still earns its place beside answered: the function count says the interface is mostly there, and the call share says the run is nearly all spin.