WM-11: Preserve policy identity through prior, inference and choice

First authored design, 2026-09-15. This elaborates exactly WM-11. It is a reviewable design, not implementation, independent acceptance or machine completion. The existing build loop continues under claude-2; these documents do not dispatch new jobs or override recorded decisions.

All WM cascades · Checklist item · Machine-readable cascade

The problem and existing evidence

WM-11 must keep distinct interpreted policies distinct through every probability calculation and into enactment. A normalized prior does not demonstrate an observed-data posterior. The current progressive-construction rule also constrains which alternatives can be accepted as extensions, independently of the probability normalization machinery.

Governing checklist snapshot

WM-11 — Candidate prior/posterior and choice preserve full policy identity. PARTIAL: full cascade+precedence identity, declared prior/choice rule, applicable per-policy F and complete detail coverage, no mission-only identity collapse, and independent selected-to-enacted correspondence. Prior-only mathematics is not the observed-data posterior. Proposed fuel accounting in selection is pending, distinct from variational F; runtime charging does not establish it. [S,T,W16,W23] Built evidence: Mathematical admissible priors and some production-pinned policy/posterior witnesses exist.

The preceding evidence is the checklist’s dated account. It has not been re-audited as a current deployment census for this design.

Pattern applications

WM-11-C1 — Define the actual candidate family

Pattern: wm-pattern-design/same-policy-through-comparison.

Retain full cascade, precedence, interpretation/model revisions, target and parent/delta where applicable. Two policies for one mission remain two policies. Record generated and excluded candidates and the reason for each exclusion; a missing model is not evidence that a candidate is impossible.

WM-11-C2 — Declare which probability law applies

Pattern: aif/declare-the-conditioning.

Name whether this decision uses a base prior, habit prior, precision prior or observed-data posterior. The pinned semantics distinguishes these formulas and labels combined habit/precision forms as extensions. Retain sources of E and any precision, with domain and positivity requirements. No configurable scalar is automatically per-policy variational F.

WM-11-C3 — Construct observed-data inference where required

Pattern: wm-prediction/name-the-uncertainty-to-learn.

For a posterior claim, produce the actual per-policy variational quantity from the admitted observations and model, with its inference/optimization assumptions. Prior-only mathematics does not close this producer. Keep state inference, parameter learning and policy inference distinct; do not substitute a channel error norm merely because it is also named free energy.

WM-11-C4 — Normalize with declared numerical semantics

Pattern: wm-prediction/probability-laws-with-runtime-meaning.

Validate the complete family and score/log-weight domains, including empty support and invalid evidence cases. A numerically stable implementation must correspond to the chosen law, not clip or default weights until it returns a distribution. Retain per-policy G, applicable F, E, precision and normalized mass with exact identities.

WM-11-C5 — Separate extension permission from choice

Pattern: wm-pattern-design/partial-design-without-imaginary-completion.

At authoring grain apply WM-10’s legal strictly-improving extension gate before accepting a proposal. Any subsequent choice among improving proposals follows an explicit rule. At enactment grain identify the applicable selection law separately. A probabilistic prior theorem cannot authorize worse extensions under the iff-decreases requirement.

WM-11-C6 — Bind selection to actual work

Pattern: wm-connected-learning/same-interpretation-predicted-and-enacted.

Retain the selected candidate and all interpretation/guard/history identities at enactment, including any authorized tie or stochastic draw information. A receipt repeating a mission ID loses the choice. Confirm actual operation correspondence, not only a copied selected vector.

Composition and revision

These are design input/constraint relationships; they do not assert automatic runtime firing. Each contribution constrains the next actual artifact.

Build interfaces

Candidate registry → full identity map. Prior/inference producers → typed E/G/F/precision with provenance. Normalizer → lawful distribution and numeric receipt. Construction gate/choice mechanism → selected full policy and reason. Enactment binding → independently checkable correspondence. Unsupported posterior inference is a named missing producer.

Each envelope must retain its schema, content identity, source/model revisions, task/occurrence and information boundary. Typed missing inputs remain missing; do not replace them with success-shaped defaults. Implementers may reuse existing compatible components, but must identify the real producer and consumer before crediting integration.

Worked situation

Two handoff designs can target the same mission but differ in whether evidence is checked before transmission. A map keyed only by mission merges them before softmax and makes the apparent selection meaningless. Preserve them through every term and show which one the receiver actually obtains.

This situation is illustrative, not a measured production result or an adopted probability table.

Build order and acceptance

1. Resolve the first two applications against the real work and exact authority. Preserve the unresolved decisions below; bell design questions before choosing dependent behavior.
2. Implement the middle producer/checker interfaces with explicit source and numeric contracts. Retain a meaningful failing example before repair and independent review after it.
3. Connect the final consumer and collect the particular evidence described above. Unit tests, source acceptance and ordinary use receive separate credit.

Distinguishing controls: Test same mission/different precedence, absent per-policy F under a posterior claim, mismatched model revisions, empty family, invalid weights and lawful ties. With equal other terms, check the ordering implied by the selected law; do not extend prior ordering to a posterior when F differs. Reject probabilistic acceptance of a non-improving extension.

Do not rerun the entire machine merely to validate a local edit. Select checks for the changed invariant; retain commits, commands and results. A future qualifying run still requires every required node and connection.

Dependencies and open decisions

WM-01 · WM-07 · WM-09 · WM-10 · WM-12 · WM-14

Open: the applicable decision law at each grain, actual F inference, tie/search policy and any proposed fuel extension. Fuel is not variational F. Do not silently choose coefficients or infer posterior completion from the existing finite prior proofs.

The dependency links express interface constraints, not a topological build order. A producer and consumer can jointly constrain an interface without requiring that all other implementations be finished first. Use a bounded packet once its relevant semantics are settled.

Sources and review boundary

The companion JSON pins the governing checklist, September rulings, current cascade-policy semantics and every applied pattern. Where the semantics source distinguishes a ruling, extrapolation or open proposal, that distinction controls. No generic finite witness is promoted to a machine instance. Existing patterns are reused within the application stated above; their other examples, thresholds and historical implementation claims are not automatically adopted.

Existing AIF reference snapshots

Retained source text, not new patterns or adopted historical claims.

@flexiarg aif/declare-the-conditioning
@title Declare the Conditioning (Score the Fit, Not the Form)
@keywords conditioning, situation, context, fit, EFE, curried, referent
@audience aif implementers, pattern authors, futon developers
@tone technical
@style pattern
@references [aif/currency-before-merge aif/valuation-reads-the-paperwork]

@how aif/structured-observation-vector
@see-also aif/term-to-channel-traceability
@why no matching problem dossier node (@holds-at absent; not among the twenty dossier nodes R1-R17, R20, TRACE). Negative claim per the committed check futon2 holes labs library-loop runs l3_no_dossier_check.py, receipt runs L3-no-dossier-check.edn (enumerates every aif pattern's @holds-at against the dossier node set extracted from PROBLEMS-*.md headings). Rationale not invented (L3 rationale-backfill pilot, zai-1, 2026-09-05)
@how own mechanism: Every valuation surface must declare what slice of the situation it consumed; G(candidate) is a lambda — only G(candidate | situation) is a value. (source: this pattern's ! conclusion line; L3 rationale-backfill pilot, zai-1, 2026-09-05)
! conclusion: Every valuation surface must declare what slice of the situation it consumed; G(candidate) is a lambda — only G(candidate | situation) is a value.

  + context: A stack grows multiple valuation surfaces (blend terms, wholeness, ΔG legs, field diagrams), each labelled with the same vocabulary ("EFE", "value", "score").

  + IF:
    You want valuations to be comparable, auditable, and honestly labelled.

  + HOWEVER:
    Different surfaces silently consume different slices of the world — one eats the live observation vector, another a 160-char prose scrap, a third nothing at all — so identical labels name incomparable objects, and cross-surface comparisons are apples-to-oranges without anyone having lied.

  + THEN:
    Give every valuation surface a three-axis declaration: SENSE (which formal object it claims), SELECTION-SEMANTICS (extremum / amortized sample / bound — over what menu), and CONDITIONING (what situation-slice it consumed). Render the declaration where the number is rendered. In pattern terms: Context → (Problem → Solution) — application is modus ponens; the world supplies the antecedent; score the fit, never the form alone.

  + BECAUSE:
    Canonical expected free energy is G(π; Q(s), C) — conditioned on beliefs and preferences by definition; a policy has no value in vacuo, and Alexander says the same of form and context. Surfaces that skip the declaration re-invent the mislabeling one level up.

    + evidence: M-evaluate-policies §9.6–9.7 + exhibit "conditions on" column (futon2, 2026-07-03): the atomic-action lane conditioned on the live observation vector while the cascade lane ate a title+status scrap and two native components consumed no situation at all — one arena, incomparable conditioning. E5 measured the conditioning channel: situation-enriched ψ rebuilt 11/13 cascades. The fold contract already typed circumstance as an argument; the plumbing had starved it.

  + NEXT-STEPS:
    - Add the sense × selection-semantics × conditioning triple to each EFE-labelled surface's declaration (extends the efe-referent-declaration shape).
    - Route all lanes toward ONE situation source (the structured observation vector) where feasible.