SPAR's core idea is simple:
reliability and admissibility are not the same thing.
Reliability asks whether a system produces stable, repeatable outputs. Admissibility asks whether those outputs deserve the meanings attached to them.
In practical terms, SPAR treats admissibility as claim-worthiness:
- does this result justify the interpretation attached to it
- does the reported maturity state still match the implementation state
- does the system have enough basis to make the outward-facing claim it is making
A system can stay green while still becoming less honest about what it is doing.
Examples:
- a stub can return a stable value and still support a false claim
- a heuristic path can be reported as if it were exact
- a registry can lag behind an implementation change
- a score can remain smooth while its justification is weak
Regression catches continuity problems. SPAR is designed to catch claim drift.
SPAR does not promise truth.
It prevents unjustified confidence.
That is a more defensible engineering target than universal truth adjudication. It lets teams force a warning, downgrade, or reclassification when output, implementation state, and interpretation stop matching.
Physics and mathematical model validation are the first hard proof case because they make the difference between:
- stable output
- justified claim
- declared maturity
sharp enough to review rigorously.
That same structure then generalizes into other systems where interpretation cannot be allowed to ride for free on top of green output.