D195 - prosper is the sixth loader, the only headless one, and the first to have ground truth to check us against

boot_trace takes a game directory and runs it with no window at all. That matters beyond convenience: every other loader in the sweep opens a window on the user's desktop, and a run long enough to be minimised is a run whose screenshot fails - while the screenshot is the report for any loader whose text channel does not work. prosper has neither problem, so it can be run as often as wanted. scripts/prosper-run.sh.

obSCEne boots under it and the resume mechanism converges as designed, each blocker costing two rounds under D191's two-consecutive-failures rule:

round 1  118 records  last=015-sync/semaphore
round 2  165          last=018-relational/semaphore-counts
round 3  165          last=018-relational/semaphore-counts
round 4  180          last=018-relational/semaphore-state-is-per-object
round 5  180          last=018-relational/semaphore-state-is-per-object
round 6  181          last=018-relational/handle-fits-its-out-parameter

The result that could not be got anywhere else

hle_registry_dump prints prosper's own registry: 1,201 real implementations, 6 placeholder, and anything absent is unimplemented and logged as such at run time. The loader states its own answer.

007-responsive guesses that answer from behaviour alone - give a function two inputs whose results must differ, and call it a stub if both come back the same. Plausible, and until now never checked against anything that knew. Checked now, over every function the section tests:

obSCEne's verdict	prosper's registry	count	agreed
responds	real	45	45
silent	absent	9	9

54 of 54, no disagreements in either direction. No false stub, no missed one. That is the first independent evidence the heuristic measures what it claims to, and it is the kind of positive result principle 7 asks for: it proves the section works, where a negative check would only have proved it rejects bad input.

The SIGSEGV is prosper's, and the announcement said so first

Round 6 died with a fault report naming a host frame:

RUN ENDED: kind=2  SIGSEGV at addr=0x1af22510  rip=... (mapped/host)
  backtrace: eboot +0x442695 / eboot +0x46252d / mapped/host +0x1

rip is in host code with obSCEne frames only as callers, so obSCEne called into prosper's HLE and that faulted dereferencing its first argument. The five semaphore entries are all real in the registry, so this is a real implementation crashing, not a missing one. The check that provoked it holds a guard word beside its handle and never dereferences a handle at all.

obSCEne's contribution was the dangling try with no res, which named the call before it was made. That is announce-before-attempting working on a loader it had never met, first run.

What it does not establish

Nothing about hardware. prosper is a fourth reader agreeing with us about module layout and a sixth loader running our code; it is not a console. The responsive result is evidence about the heuristic, measured against one implementation's idea of its own completeness.

Status: derived - 54/54 measured against a registry the loader publishes about itself.