D357 - An experiment declares how its result is read

Status: decided Date: 2026-09-26

An experiment names both the axes to run and a Reading that decides the question from recorded numbers, or Undecided with a reason. A run records the memory map it actually built in Conditions::memory_map. An answered question becomes an inert proposal against the entry that asked it, with known_by = "measured", and is not proposed again once settled.

Why: fault and reach cannot say whether a guest accepted a map; the offset it queries next can, and that is arithmetic. A run that could not decide must not be recorded as a negative. Recomputing the map from its configuration is a second copy of what is measured and wrong when a shape falls back. Re-proposing a settled answer fills patches/ with noise. Questions about correctness still need the conformance probe or a person.

Rejected:

• Reading every experiment through the generic diagnostics: answers a different question.
• Recomputing the presented map: disagrees when a shape falls back.