D044 - Every check records where its expectation came from

Status: decided.

A conformance suite is worth what its expectations are worth, and they are not all worth the same. "strlen returns the number of characters" is settled by a standard anyone can read. "closing an invalid handle returns non-zero" is a belief this program holds that nobody has confirmed. Both produce a FAIL, and the report presented them identically - so an emulator author reading one could not tell whether it was their bug or ours.

obs_check now carries obs_provenance, appended to the res record:

	means
spec	ISO C or POSIX. Settled by a document.
documented	vendor documentation describes this behaviour specifically
hardware	observed on a real console and recorded
assumed	this program's own reasoning; sensible, unconfirmed

Nothing carries hardware yet, and the report says so out loud. The current split is 51 assumed, 30 spec. pretty prints that summary and, while no check is hardware-backed, adds the caveat that a failure marked [assumed] may be this suite's belief rather than a bug.

That absence is the point. It makes the value of a hardware run concrete: such a run does not merely check the suite, it upgrades it, moving checks out of assumed one at a time. Only after that is a green run something an emulator can be held to.

Done now because it gets more expensive with every check added, and the suite is at its smallest it will ever be. assumed is zero so the conservative answer is the default

• a new check that says nothing about its provenance is correctly described as an assumption.

The initial assignment is by section: ISO C, maths and pthreads are spec, everything else assumed. That is deliberately coarse and deliberately pessimistic. Refining it is what hardware is for.