2026-08-26 - the substitute oracle, and mistaking my own comparison for a finding

Four loaders complete the full suite as of today, so consensus had four complete reports for the first time. It reported 142 agreed, 360 disagreed, and named PS5PCEM as the lone dissenter 328 times out of 358.

That is a damning-looking result about one project and it was an artefact of my comparison. The sweep builds a module per loader generation: PS5PCEM ran GEN=5 and the other three ran GEN=4. The reports say so in their own sysinfo line. Within a generation, same tool, same day:

cohort	agreed	disagreed
mixed generations, 4 loaders	142	360
same generation, 3 loaders	469	32

The outliers rebalance too - Kyty 15, shadPS4 8, fpPS4 5, where the mixed run had put 92% of them on one loader. Almost every disagreement was the comparison, and it pointed hard and specifically at a project that had done nothing wrong.

The real finding is what that leaves: the working oracle is three previous-generation emulators, and for the generation this project targets there is no cohort at all. Agreement is currently evidence about the previous platform. (D197)

The report corroborates itself

007-responsive and 035-libc reach the same functions by different routes. Over five string functions and three loaders they agree on every case - including fpPS4's strstr, which responds and only partially passes. Implemented and wrong, which is the distinction the two sections exist to draw and needs different work from an absence.

Kyty is silent and failing on all five: it implements essentially no libc string functions, which is what its fifteen outliers are. The patch did not cause that, it made it visible.