Module compare 

Candidate comparison over a reproduced case (spec 004, 3.4.4 to 3.4.7).

A candidate plan runs against the case’s snapshot, evaluation time and scope. A retained raw output is reused only for a candidate primary request whose identity equals the one it was produced under, and only when every retained output with that identity agrees. Historical runtime failures are never reused to predict a candidate’s failure, and fallback execution is not simulated: a request without an equivalent successful output makes the case incomparable.

Structs

Compared

A candidate’s judgment beside the reproduced historical one. Agreement is the task adapter’s to define; whole-judgment equality is not it.

Enums

CandidateIncomparable

Why a candidate cannot be compared on this case.

Functions

compare

Run candidate over the reproduced case, reusing retained outputs by exact request identity only.