Through the ranked backlog

§11 relational checks: five to ten. Two opens of one path must give two descriptors, closing twice must not succeed twice, a clock must never go backwards, two live mutexes must not share a handle, a held default mutex must refuse a second lock.

descriptors-distinct has the worst failure mode in the section: an implementation returning the same descriptor twice passes every check in 040-file - each open succeeds, each read works - and a caller that closes one has silently closed the other's.

Eight pass under shadPS4, two skip. The skips are informative: the file relations need OBS_CAP_FILE, which is established by a check excluded for ending the process. Excluding a crashing check also withholds the capability it provides, which is correct and has reach beyond the checks named in the exclusion list.

§12 consensus, built - and its first run showed the design was wrong. 80 disagreements of 126, almost all "this platform has the function and that one does not". A skip is not an opinion; an implementation that lacks a function has said nothing about how it should behave. Counting skips separately took it to 61 real behavioural differences (D072).

The backlog said this needed a second emulator reporting. It does not: the host build is a real implementation of the POSIX and C library surface, so disagreement against it is disagreement with something known to work.

FreeBSD provenance, narrower than the task assumed. A mass relabel would have been wrong twice: sceKernelClose is not close, so POSIX settling close is an inference about the renaming; and a pattern rule would have swept up open-rejects-null, which POSIX explicitly leaves undefined. OBS_FROM_DERIVED says what is true, seven checks upgraded, 38 still assumed and most of them rightly (D074).

The graphics interface, censused. 87 names with libraries from public sources, every one verified against our own hash - which rejected five, two of them placeholders with the identifier embedded in the name (D075).

Then all 87 reported present under shadPS4, a previous-generation emulator with no implementation of the interface at all. The clearest evidence yet that presence measures a loader's stubbing policy: an entire console generation's graphics interface, reported present by a census working exactly as designed. The availability field carries current on those records, which is what a reader needs to discount them.

The drift gate earned itself twice, failing the build when checks were added and the documentation had not followed. That is the whole point of it.

State

778 records under shadPS4. 111 checks - 65 spec, 7 derived, 38 assumed, 1 documented, 0 hardware. Census 399 symbols across 17 libraries: 296 shared, 87 current-generation, 17 previous.