Select/call and Cordis integration workshop

This workshop tests how Cordis and its use in DeepSeek Harness change the scope, state model, and downstream use of the select/call decomposition lemma. It is an edit-mode cp-skill-write-multistage run. The library target remains untouched until a reconciled candidate passes audit and acceptance review.

Run identity

  • Immutable run key: kb/notes/any-symbolic-program-with-llm-calls-is-a-select-call-program.md
  • Current intended target: kb/notes/any-symbolic-program-with-llm-calls-is-a-select-call-program.md
  • Mode: edit
  • Collection: kb/notes/
  • Type: kb/types/note.md
  • Acceptance review: required — the note is a load-bearing formal claim used to transfer results across the computational-model cluster

Governing question

Under what exact conditions does the select/call decomposition remain valid when the harness is dynamically composed, executes non-LLM effects, or lets an LLM produce control decisions? The revision should preserve the mechanical lemma where it holds while preventing readers from treating an LLM-call trace normal form as complete operational semantics for an effectful harness.

Inputs

Local evidence and theory:

Pinned primary implementation sources (DeepSeek Harness revision 47f943859bef60e4160492346772ded9b24f765a, observed 2026-08-14):

DeepSeek Harness is a developer preview. The pin fixes the workshop's implementation evidence; claims about later revisions require fresh inspection.

Workflow state

  • [x] Incumbent preserved in original.md
  • [x] brief.md
  • [ ] reconstruction.md
  • [ ] claim-skeleton.md
  • [ ] draft.md
  • [ ] audit.md
  • [ ] candidate.md
  • [ ] acceptance.md — required
  • [ ] Promotion and deterministic validation

Unresolved decisions and blockers

  • No current blocker to source reconstruction.
  • The current run owns one target only. A general effectful select/dispatch normal-form note is a possible synthesis outcome, not an authorized second target in this run.
  • Whether dynamic composition state belongs inside K or should be named separately as a configuration Γ is an open theoretical choice. Reconstruction and the claim skeleton must compare both formulations before drafting.
  • DeepSeek Harness evidence is revision-scoped to 47f943859bef60e4160492346772ded9b24f765a. The Cordis paper must independently warrant any general mechanism claim.

What closes this workshop

The workshop closes when it produces and promotes a validated revision that:

  1. states whether the lemma is a complete execution normal form or only an LLM-call trace normal form;
  2. gives exact treatment to external effects, dynamic code/configuration, and LLM-produced control decisions;
  3. preserves downstream theorem transfer only under explicit preconditions;
  4. uses Cordis as a mechanism-level challenge or witness without generalizing from DeepSeek Harness adoption or preview status;
  5. resolves every audit finding and passes a fresh acceptance review.

Because the user requested a workshop, retain this directory and its active-workshop entry until promotion is separately requested or completed.


Complete file listing (generated at build time)