Skip to content

TAS-184: Clean-room weight causality and interpreter ownership gate #307

@AtlantisPleb

Description

@AtlantisPleb

Summary

  • prove that admitted interpreter behavior comes from Transformer forward-pass
    computation and turn that into one ownership verdict

Roadmap Position

  • Phase: Phase 7. Close single-run posture, state discipline, and weight ownership
  • Sequence: article-gap closure after TAS-156A
  • Owned-route boundary: any canonical article route referenced here means the psionic-transformer-based route established by TAS-156A / TAS-160, not a reintroduced mixed implementation spread across model or runtime crates.
  • Claim posture: necessary for article-equivalent closure, but not sufficient by itself to widen public capability claims

Description

  • build the interpreter-in-weights conformance matrix over the declared article
    breadth envelope
  • audit the canonical route for:
    • forward-pass sufficiency
    • route purity
    • hidden host-side execution substitution
    • no external oracle dependence
    • preprocessing shortcuts that collapse the real compute
    • route drift
    • no runtime-owned control flow
    • helper-module or import mediation that changes where the work lives
    • cache or retrieval layers that silently own the decisive step
    • no cache-owned control flow
    • cache-ownership exclusion
    • artifact-lineage-to-behavior closure
    • weight perturbation sensitivity
    • causal scrubbing and intervention behavior:
      • zeroed or randomized weight subsets
      • clamped attention heads
      • removed layers
    • computation-locality characterization:
      • localized
      • modular
      • or diffuse
  • publish one computation mapping report that states:
    • where the program representation lives
    • where state lives
    • how control flow is realized
    • whether those mappings remain stable across runs
  • widen direct no-tool proof from demos to the declared generic suite
  • combine all of that into one interpreter-ownership gate

Why This Matters

  • correct outputs alone do not prove article-equivalent ownership in the
    weights

Primary Surfaces

  • psionic-transformer (canonical owned article-route base for this issue)
  • psionic-models
  • psionic-runtime
  • psionic-eval
  • psionic-research
  • psionic-provider

Validation

  • conformance matrix
  • clean-room causality audit
  • generic direct-proof suite
  • weight perturbation report
  • causal scrubbing / intervention report
  • computation mapping report
  • locality / modularity characterization report
  • ownership gate checker

Claim Discipline

  • Keep machine-legible boundaries explicit; partial progress must not be presented as full article equivalence.
  • Preserve typed refusal, fallback, blocked, and suppressed posture where closure is still incomplete.
  • Keep public language bounded to the declared article envelope rather than implying generic arbitrary-program closure from a partial green result.

Done When

  • one interpreter-ownership gate exists and is green

Metadata

Metadata

Assignees

No one assigned

    Labels

    tassadarPsionic Tassadar roadmap work

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions