First authored design, 2026-09-15. This elaborates exactly WM-06. 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
A declared preference distribution and a prediction can both normalize while describing different outcomes. WM-06 must connect the specified C to the actual Q consumed by the scorer. The terminal machineC seed is bounded existing evidence; it does not settle the broader family, a state-to-disposition bridge or production consumption. This design keeps those responsibilities separate without making any of them optional.
WM-06 — C is the specified preference distribution on Q's outcome domain. PARTIAL: actual matched Q/C scorer consumption and evidence. Owner bell invoke-1789481249314-21037-9af06802 limits one interpretation-design packet to terminal machineC; it does not narrow the whole build. The proposed state-to-disposition bridge remains subject to the model/C authority; a constant adapter is not evidence of state-dependent prediction. Preserve zero-preference refusal and exclusion of preferences over evidence content. No replacement preference numbers are implied. The full C-family obligation is separately retained below. [T,C,S] Built evidence: Ruled terminal machineC, twelve tagged masses, seven zeros, normalized seed and supplied-kernel risk exist. Channel preferences demonstrably affect controller scores.
The preceding evidence is the checklist’s dated account. It has not been re-audited as a current deployment census for this design.
Pattern: wm-purpose-and-account/recipient-benefit-shapes-the-model.
Identify whose continuation the design serves and what successful or unsuccessful work means at this grain. Preserve the distinction between the existence of a communicable witness and preferring particular evidence content. Output a domain rationale, not invented preference weights. A convenient telemetry channel cannot replace a participant-relevant outcome.
Pattern: wm-pattern-design/preference-family-needs-composition.
Select the member authorized for this grain and observation time, with provenance and composition inputs. Terminal machineC is one available member; an interpretation packet limiting itself to it did not waive general C. Output a member request and explicit unresolved dependencies on WM-13. Do not substitute the terminal member at every step because it is implemented.
Pattern: wm-prediction/outcome-meaning-across-models.
Bind both distributions to the same tagged outcome identities, model context and comparison schedule. If a bridge maps state or observations to dispositions, state its interpretation and validation evidence. A constant adapter that ignores its input does not establish state-dependent prediction. Output a checked Q/C envelope or a typed domain/bridge refusal.
Pattern: wm-prediction/probability-laws-with-runtime-meaning.
Require normalized nonnegative preferences under the adopted numeric contract. Preserve the terminal seed’s seven named zeros; never smooth them to obtain a finite risk. Under the current risk-admissibility contract positive Q at zero C refuses finite scoring; zero-Q terms contribute zero. Output the support check and exact reason for any inadmissibility.
Pattern: wm-qualifying-run/certificate-and-world-must-meet.
Bind the source member, any fold and bridge, model revisions and independent review to the actual envelope. A proof with arbitrary C proves a conditional law, not that this C was constructed by the machine. Keep author tests, reviewed source and production use as separate evidence.
Pattern: wm-connected-learning/observation-to-next-choice.
Pass that same envelope to the canonical G consumer and retain which member and outcome mapping it read. A controlled test may vary an authorized preference input to show sensitivity; this does not authorize changing the live seed. Production evidence must show actual consumption rather than a logged unused C.
These are design input/constraint relationships; they do not assert automatic runtime firing. Each contribution constrains the next actual artifact.
| From | To | Relationship | Why |
|---|---|---|---|
| WM-06-C1 | WM-06-C2 | prerequisite | Participant concerns constrain which preference member is relevant. |
| WM-06-C2 | WM-06-C3 | prerequisite | The selected member fixes the outcome meaning to match. |
| WM-06-C3 | WM-06-C4 | prerequisite | Risk admission must check the actual paired domains. |
| WM-06-C2 | WM-06-C5 | prerequisite | Evidence must identify the resolved member and fold. |
| WM-06-C4 | WM-06-C5 | prerequisite | Admission evidence includes the zero-support decision. |
| WM-06-C5 | WM-06-C6 | prerequisite | The consumer must use the admitted envelope. |
| WM-06-C6 | WM-06-C1 | revision-feedback | Consumer-domain mismatch requires reconsidering the outcome rationale. |
Member resolver: grain/time plus authority → member and source pins. Q/C adapter: actual prediction plus member → matching-domain envelope or refusal. Risk boundary: envelope plus numeric contract → canonical contribution and retained admission. Consumer receipt: envelope identity plus call-site identity → demonstrated use. WM-05 supplies Q; WM-13 supplies the family; neither can be fabricated inside the adapter.
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.
For a disputed handoff, an emitted record can exist while the successor cannot resume. A model mapping both situations to the same “successful” disposition loses the very difference C should value. Revise the outcome interpretation through its authority; do not compensate by assigning arbitrary weights to evidence text. A refusal exposes missing compatibility and does not complete WM-06.
This situation is illustrative, not a measured production result or an adopted probability table.
Distinguishing controls: Reject swapped outcome order, missing vertex tags, stale member revisions, unvalidated bridge output, positive Q on zero C and silent preference renormalization. Include valid zero-Q/zero-C handling. Test a constructed compatible envelope independently of the machine producer, then exercise the real producer/consumer connection. Preserve the distinction in receipts.
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.
WM-01 · WM-04 · WM-05 · WM-10 · WM-12 · WM-13
Open: the applicable processual member at each scored step, validation of the observation/disposition bridge, and exact numeric correspondence at risk boundaries. The terminal seed is not reopened merely because these connections are incomplete. Mission WANT-fact/softness proposals require their own decisions.
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.
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.