D192 - A project that reads the others is not another vote, and prosper says so in its own comments

prosper was added to the toolkit on the argument that a fourth mature implementation would promote several one-source constants to corroborated without needing hardware. Reviewing it before running it showed that argument was wrong, which is the whole reason to review before counting.

It cites Kyty and shadPS4 by name, throughout. A structure is described as "0x50 bytes, Kyty layout"; a setter "mirrors Kyty's"; an error value is sourced to "Kyty Errno.h". Where it agrees with Kyty, that is very often the same reading arriving twice, and treating it as two would be double-counting the first.

The rule this sets for the whole toolkit

Agreement between implementations is only evidence when the implementations are independent, and independence is a property to check rather than assume. Five loaders in a table look like five readings; some of them are one reading with four citations.

So corroboration is counted per claim, not per project, and a claim is worth what its provenance says:

what the source says about a claim	worth
annotated, citing FreeBSD or another published contract	strong - a citable source, the standing SPEC has
annotated, citing a live capture of a retail title	real - guest-observed, the platform's own software
bare constant, no annotation	no better than ours
explicitly derived from another project here	not independent; do not count

prosper makes this practical by grading itself - 40 CONFIDENCE: HIGH, 13 MED, 4 MED-HIGH, 3 LOW. That is the same instinct as this project's provenance ladder and orbistoun's known_by, and it is what makes the source usable rather than merely present.

What the review produced anyway, which is more than agreement would have

A named disagreement on mutex type 4. Both accept exactly {1,2,3,4} and refuse 0 - so the shape of D177 is corroborated and the POSIX numbering is ruled out twice. But Kyty maps 4 to NORMAL and prosper maps it to ADAPTIVE_NP, deliberately:

"Weight Kyty DOWN here: no PS4 title it runs exercises adaptive self-lock; FreeBSD libthr is the platform contract. CONFIDENCE: HIGH (FreeBSD source + the live wedge -> unwedge flip)."

015-sync/mutex-recursion already reports second-acquisition behaviour per type, so it distinguishes these: a type returning EDEADLK on self-lock is a different observation from one that succeeds. A specific contested value one hardware run settles is worth more than another agreement, and this project had no way to generate one before.

Independent confirmation that EINVAL is right. prosper returns 0x16 for an unrecognised type where Kyty calls EXIT. That is what POSIX specifies and what patches/kyty-probe-friendly.patch changes Kyty to do - reached without reference to that patch, so the patch is a correction rather than an accommodation.

The user model has a corroborated shape and an uncorroborated value. Capacity four with -1 for empty now holds across prosper, PS5PCEM and shadPS4. The initial user's value still has four implementations giving three answers, and prosper's 1 is unannotated. That split - shape settled, value open - is exactly what HANDOVER-ORBISTOUN.md asks orbistoun to build against.

Status: decided - the counting rule follows from what the source says about itself.