First authored design, 2026-09-15. This elaborates exactly WM-14. 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
WM-14 makes the agent’s pattern work part of the actual loop: interpreting retrieved arguments, creating missing patterns when warranted and revising retrieval when it is unhelpful. A token compiler, ambient notification or a generated library entry is not this behavior. The new process must remain accountable to the task and to independent admission.
WM-14 — In-loop agent interpretation, missing-pattern authoring and search feedback. OPEN: actual agent interpretation in the loop, bounded ARGUE/genesis for missing patterns and a feedback test/path to reconsider retrieval when picks do not help, with source-bound evidence and independent review. A token compiler or a list of unhelpful picks cannot substitute. Built evidence: Source interpretation/retrieval/receipt machinery exists. Joe's September 15 session instructions are reported by claude-20; the supplied transcript line numbers no longer resolve in the current shortened file, so immutable original-event retention is still requested.
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-pattern-design/pattern-before-operation.
Start from the selected problem’s context and competing concerns. Read retrieved patterns as arguments and say which response could help. Keep an unhelpful result explicit; neither a mission tag nor a familiar title establishes fit. Output a source-bound interpretation attempt for WM-08/03.
Pattern: wm-purpose-and-account/recipient-benefit-shapes-the-model.
Define what a useful result must enable in this situation: for example identifying missing handoff evidence or a justified next design change. Retain the assessment and its evidence. A lack of immediate score improvement is not automatically proof that retrieval was useless; distinguish interpretation failure, missing model inputs and genuine absence of an applicable response.
Pattern: wm-connected-learning/proposal-earns-later-use.
When the library lacks the needed response, create a draft pattern with context, IF, HOWEVER, THEN, BECAUSE, sources and an explicit evidence hole. Preserve alternatives and failure modes. Do not rename an operation or invent attestation. Scope the ARGUE/genesis work and retain its parent problem and reason for creation.
Pattern: wm-connected-learning/interpretation-carries-its-argument.
Bind source revision, task-specific effect/guard meaning, interpreter and independent review where required. Generated pattern and executable interpretation are separate artifacts. WM-03’s semantics govern any controlled rule; WM-08’s independent expectations/designations remain independent. Publication alone is not admission.
Pattern: wm-pattern-design/partial-design-without-imaginary-completion.
When picks do not help, retain the attempted strategy and why it failed, then change a declared query, source scope or search procedure. Keep prior results and bounded search coverage. Feed legal interpreted proposals into WM-09/10; no model-backed score is fabricated just to rank a newly authored pattern.
Pattern: wm-connected-learning/observation-to-next-choice.
Record the admitted pattern or revised search result reaching an actual subsequent design decision and later outcome. Separate creation, retrieval, admission and use counters. An unchanged or unhelpful result remains a finding that can revise the search, not evidence of learning merely because another file appeared.
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-14-C1 | WM-14-C2 | prerequisite | Usefulness must concern the actual interpreted problem. |
| WM-14-C2 | WM-14-C3 | prerequisite | A documented gap can justify authoring a proposal. |
| WM-14-C1 | WM-14-C4 | prerequisite | Existing patterns still require source-bound interpretation. |
| WM-14-C3 | WM-14-C4 | prerequisite | New arguments require admission before executable use. |
| WM-14-C2 | WM-14-C5 | prerequisite | Unhelpful retrieval supplies the reason to revise search. |
| WM-14-C4 | WM-14-C5 | prerequisite | Construction needs admitted interpretations. |
| WM-14-C5 | WM-14-C6 | prerequisite | Later use tests whether the revised strategy helped. |
| WM-14-C6 | WM-14-C1 | revision-feedback | Later failed use can reveal that the original problem was misunderstood. |
Problem/retrieval envelope → interpretation attempt. Usefulness assessor → supported gap classification. ARGUE producer → draft pattern and provenance. Admission/reviewer → scoped accepted interpretation. Search-policy reviser → versioned strategy and bounded results. Loop integration → actual later-use evidence. Do not add an automatic library-wide admission side effect.
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.
The handoff search returns patterns for formatting a report, but none addresses preserving a disputed claim and its limits. The agent should explain the gap, broaden or change search, and if necessary draft the missing argument. Only after justified interpretation and admission should it affect the design; the receiver’s later use is separate evidence.
This situation is illustrative, not a measured production result or an adopted probability table.
Distinguishing controls: Exercise helpful retrieval, misleading title, no useful result, missing interpretation, a justified new draft and rejection of an invented operation-as-pattern. Verify a real strategy change after documented unhelpfulness, with prior attempts preserved. Reject self-issued review authority and generated-pattern counts presented as successful-use evidence.
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-03 · WM-08 · WM-09 · WM-10 · WM-11
Open: operational usefulness criteria, bounded ARGUE/search authority and the actual integration path. The checklist’s original transcript line references were reported stale; later supplied evidence IDs are discovery leads until resolved and pinned. This new design request authorizes authoring cascades, not silent changes to the runtime library or historical authority.
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.