Module explain 

Resolving a model’s stored sampling configuration, with provenance.

The live pipeline answers “what did this request end up sampling with?”. This answers the question the explain surfaces ask instead — “what would this model sample with, and which rung supplied each value?” — with no request in hand, so the top rung is empty by construction.

Why it is shared rather than written twice

gglib model explain and the GUI’s sampling panel had a copy each, and they had to agree: the same ladder, followed by the same stage-5b effort gate applied to the resolution rather than to a request. Two copies of a rule can only ever drift into two accounts of one resolution, and the whole value of an explain surface is that it describes the resolution that actually runs. So the rule lives here once, and the callers keep only what genuinely differs between them — one prints, the other builds a DTO.

It sits beside effort_gate because suppress_stored_effort does, and applying that gate offline is half of what this function is.

Functions

explain_stored

Resolve stored sampling for one model, and report where each value came from and whether the template gate silently dropped an effort level.