SPAR is built first for mathematical and physics-grade model review, but the review pattern is broader.
Use SPAR when:
- output can remain numerically stable while justification weakens
- analytical anchors matter
- maturity state changes what a result is allowed to claim
- approximation, boundedness, or partial closure must remain visible
- runtime memory state and invariant continuity should tighten review, not stay outside it
In the current physics adapter, this contextual tightening already appears as:
B5— MICA runtime stateC10— MICA invariant continuity
Use SPAR when:
- reproducibility is necessary but not sufficient
- review must distinguish exact, approximate, heuristic, and bounded results
- implementation changes can outpace governance language
This is also the natural bridge to the next adapter direction:
- generic scientific-model review for PDEs
- dynamical and control models
- inverse and calibration models
- constrained optimization systems
- scientific ML surrogates
Use SPAR when a model can predict well but still overstate what part of the underlying theory it actually instantiates.
Use SPAR when output status, maturity labels, and implementation reality can drift apart across releases or operating environments.
Use SPAR when tests pass but the claim of completeness, closure, or capability still needs a second review surface.
Use SPAR when a green dashboard is not enough and downstream users need to know whether a result is exact, partial, heuristic, or environment-conditional.