A label keyed to a proxy variable contradicts the model as soon as the parameters leave the regime it was calibrated in

Fact

Classifying state by a stand-in variable is cheap and usually right, because in the default configuration the proxy tracks the thing. Change the parameters and the two come apart, while the label keeps reporting with full confidence.

A cosmology view labelled its eras by scale factor alone. For the measured universe that is correct: the present is the dark-energy era. Set the matter density high enough to recollapse and the same code still displayed "Dark energy era. Expansion is accelerating." beside its own readout of a positive deceleration parameter and a fate of "Big Crunch".

Why it matters

The contradiction is between two things on the same screen, both computed correctly, so no test of either one catches it. It is also the output most likely to be trusted: a label is prose, and prose reads as authored rather than derived.

It only appears once someone moves a control away from the default, which is exactly what an explorable model is for and exactly what is least exercised during development.

How to apply

• Separate the boundaries that are genuinely regime-independent from the ones that are not. Recombination happens at 3000 K whatever the expansion history; whether today accelerates is a fact about the parameters.
• Derive the parameter-dependent labels from the same quantity the readout shows, not from a correlate of it.
• Test the contradiction directly: assert that the narration for a recollapsing universe does not contain the word "accelerating".
• When a proxy really is the only thing available, say what it assumes in the label rather than asserting the conclusion.