E.23:1 - Problem frame

When the entry phrase is “loop engineering”, “agent loop”, “harness loop”, or “improve this with an agent”, treat the phrase as a recognition cue, not as an FPF kind. First recover the object version under improvement and the evaluation that can be rerun. If those cannot be named, this is not yet an E.23 use; name the live claim and use its subject pattern for it. Common exits are work, transformation-flow structure, evolutionary retention and publication, source use, refresh, gate-decision publication, and DPF framework authoring.

Use E.23 when an object version will be improved through repeated passes under a declared object-under-improvement evaluation. The object can be a pattern, DRR, FPF corpus object, engineering quality object, naming candidate, OEE and NQD candidate, archive or front member, selected set, parity report, refresh report, or declared transformation result, if an exact evaluation supplies values and stop meanings for that object kind.

Not this pattern when one direct quality evaluation is enough. Use E.22 to frame one evaluation and then run the named object-under-improvement evaluation. Use A.19.ECS first if the needed evaluation characteristic space does not exist.

Use E.23.CAE first when the apparent object to improve remains ambiguous because a holder’s capability may be outside its claimed envelope, unavailable through the current configuration, not selected as applicable, inaccessible or inactive, context-dependently unexpressed, unadapted, unenacted, or actually changed. That separate probe returns observations, a qualified disposition, surviving rivals, and candidate routes; it does not choose the object under improvement.

Use E.23.CDI instead when a separate applicable steering or choice result has made capability development current for one admitted holder System and named Work family, and the change must be checked in representative Work. That is a separate capability-development Method; E.23 points to its description without copying its actions. Use a population assessment when the question is a distribution across member capabilities, C.36 when generation, transmission, recognition, selection, retention, or loss across a cultural population is current, and C.32.MWA first only when the target-practice architecture itself must be recovered or compared.

First useful move: name the object version under improvement, the exact evaluation that will re-evaluate it, the improvement aim, protected trade-offs, cost and risk account, and local stop condition. Here move is Plain instruction wording: it names no Move kind, method, plan, performed Work, or actual Transformation.

What goes wrong if missed: teams close discharge rows instead of improving quality, retry blindly, optimize visible values while damaging protected qualities, stop merely because every value is 5, or let a review recommendation become decision, work, evidence, selected-set result declaration, actual publication, parity, or refresh by stealth.

What this buys in practice: each pass has a declared object version, an intended evaluation-result change, a rerunnable evaluation, protected trade-offs, and a stop or switch condition. Effort can then change substantive quality and stop when no non-dominated change is worth its cost, instead of merely producing more review state.

Primary EntityOfConcern in plain terms: the repeated quality-improvement method for one object version under one declared evaluation.