Jump to text
On this page
Copy and source
Published source confirmed at last check

Source changed 2026-11-18 21:11:39 UTC · snapshot created 2026-11-18 21:12:35 UTC · last check 2026-11-18 22:10:10 UTC

A.19.CPM:0 - At a glance (didactic, informative)

CPM is the CHR comparison kernel: it compares two admitted profiles under an explicit, admissibility‑gated comparator and returns a set‑valued comparison outcome.

Purpose. CPM answers: “Given two admitted profiles and an explicit comparator, what relation holds under the declared admissibility frame?” It does not answer: “Which one should we pick?” (selection) nor “What is the score?” (scoring).

Use this when. Two admitted profiles must be compared under a declared rule. Recover their value meanings, select the admitted comparator and test its use conditions. Return only justified relation tokens when comparison proceeds; retain the eligibility judgment and withhold a comparison result when it must abstain. If this answers the receiving question, stop without adding selection or authorization.

What this buys. A completed admitted comparison gives the practitioner one set-valued outcome that downstream selection can consume. The Compare application keeps the profile pair, comparator, claim scope and selected context slices, optional A.19 predicate, reference plane, evaluation window, policies, and output binding recoverable. Partial order, incomparability, missing evidence, and scale limits remain explicit instead of becoming a hidden scalar winner.

First output. When comparison returns a result, read the by-value set bound to ComparisonResultSlot: only the relation or poset tokens. Read comparator, comparison scope, predicate when used, plane, window, eligibility value, and evidence use from the operation application and their neighboring relations; they are not fields hidden inside the output.

Manager quick checklist (before you trust a comparison):

  • Comparator is explicit: do we have a ComparatorSpecRef, and is it admitted by CG‑Spec.ComparatorSet?

  • Admissibility is declared: do we cite CG-Spec and any required SCP, and restrict a degraded result to a lawful outcome explicitly permitted by the failure policy?

  • Evidence is not faked: are missing or unknown inputs treated as degrade|abstain under the effective MinimalEvidence policy (never as pass)?

  • Partiality is preserved: are we willing to accept incomparability and ties as first‑class outcomes (set‑valued result), rather than forcing a winner?

  • Suite stage: compare (pipeline order lives in A.19.CHR:4.5, not in the mechanisms[] enumeration).

  • Input (conceptual): left profile, right profile, CN-Spec, CG-Spec, an explicit ComparatorSpec, one U.ClaimScope with selected A.2.6 U.ContextSlice members, an optional A.19 CharacteristicSpacePredicate when the comparison depends on one, effective reference plane, explicit evaluation window, and optional MinimalEvidence override.

  • Output (conceptual): the by-value ComparisonResultSlot set of relation or poset tokens.

  • Planned use: an A.15.2 baseline selects ComparatorSpecRef editions and policy ids. A.15.3 and SlotFillingsPlanItem apply only to independently declared receiving positions under A.19.CHR:4.7.2. The Compare application carries its argument and result bindings under §4.1. Any dated comparison Work and A.10 evidence-provenance account retain their independent grounds.

  • Reproducible comparisons: for parity and benchmark style runs that require a stable run package plus report record (editions, windows, parity pins), use G.9 (Parity and Benchmark Harness). CPM stays kernel-only.

  • Core safety commitments: admissibility gate via CG-Spec.ComparatorSet + CG-Spec.SCP + CSLC; tri-state admissibility (pass|degrade|abstain); unknown never coerces to “pass” or to a fabricated outcome; no silent scalarization or totalization.

  • Where method details live: in editions of ComparatorSpec and their SoTA wiring (Part G packs and extensions), not inside CPM’s kernel semantics.

  • Quick rule of thumb: use USCM when you need scores from a declared scoring method and SelectorMechanism when you need a selected candidate set. Obtaining measured values follows the C.16 measurement method; normalization follows UNM. CPM compares admitted profiles and returns relation tokens.