Reverse engineering
through glue

Start from the parts Research protocol

Record each partial explanation’s variables, symmetries, validity scale, measurement inputs and assumptions. Then provide a dictionary identifying shared quantities. Different notation or scales do not by themselves imply a contradiction.

Four possible outcomes Research protocol

1. Necessary: what every joint solution must preserve.
2. Free within limits: parameters, degeneracies and distinct realisations passing the same checks.
3. Impossible: partial constraints admitting no joint solution.
4. Missing structure: hidden variables, couplings, memory or an incorrect dictionary as candidate explanations of a mismatch. A mismatch does not establish which addition is correct.

A small consistency test Conditionally derived

Take binary variables A, B and C. Three partial models require A=B, B=C and A≠C, each with certainty. Each pair model is possible in isolation, even with identical uniform single-variable marginals. Together they are impossible: the first two imply A=C. Agreement on overlapping marginals alone is therefore insufficient.

The next question is which assumption fails: the equalities, their certainty, the variable identifications or their domain of application. A hidden variable cannot resolve this exact contradiction while all requirements remain in force.

Reverse engineering without building in the answer Research protocol

Freeze the inputs, derive the joint constraints and keep the target outcome separate. Compare a full model with its effective description. Report agreement, information loss and counterexamples. The resulting retrodiction validates the bounded application, not automatically the philosophical whole.