Skip to content

ABI Layer 4: prove the FFI result-code seam is injective + faithful#75

Merged
hyperpolymath merged 5 commits into
mainfrom
claude/new-session-znxgm7
Jun 28, 2026
Merged

ABI Layer 4: prove the FFI result-code seam is injective + faithful#75
hyperpolymath merged 5 commits into
mainfrom
claude/new-session-znxgm7

Conversation

@hyperpolymath

Copy link
Copy Markdown
Owner

Summary

Layer 4 (seal the ABI↔FFI seam): proves the FFI result-code encoding (resultToInt) is a sound, lossless wire encoding — decoder intToResult with resultRoundTrip, from which resultToIntInjective is derived. Distinct ABI outcomes never collide on the C wire. Complements the structural abi-ffi-gate.py.

New module *.ABI.FfiSeam (imports Types). Positive + non-vacuity controls.

Testing

Idris2 0.7.0 --build → exit 0, zero warnings. Adversarial rejection confirmed. build/ removed. No believe_me/postulate/sorry.

🤖 Generated with Claude Code

https://claude.ai/code/session_01A6PSzJWpRxtzGDjUCEh7Mx


Generated by Claude Code

claude and others added 4 commits June 27, 2026 19:31
Add Iseriser.ABI.Semantics proving the headline domain property: a
generated -iser scaffold is Conformant exactly when all five required
components (Manifest, Idris2 ABI, Zig FFI, Codegen, Rust CLI) are
present. Conformant is built from genuine Data.List.Elem membership
obligations with no catch-all constructor, so a scaffold missing any
component has no witness.

Includes a sound+complete decConformant : (s) -> Dec (Conformant s),
a soundness fact certifyConformantSound, a positive control
(completeIsConformant via explicit Elem positions), and a negative
control (ffiMissingNotConformant : Not (Conformant ffiMissing)).
Non-vacuity confirmed: a deliberately-false Has Ffi witness for the
FFI-missing scaffold is rejected by idris2.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A6PSzJWpRxtzGDjUCEh7Mx
Add Iseriser.ABI.Invariants over the existing Layer-2 Semantics model
(Component/Scaffold/Has/Conformant). Two new, deeper, distinct theorems:

  1. Generation soundness (correct-by-construction): genScaffold provably
     emits (s ** Conformant s) for any LanguageModel, with a corollary
     tying it back to the Layer-2 certifier (conformantCertifies).
  2. Upward-closure / monotonicity: conformantStable proves Conformance is
     preserved under extension, via a genuine Elem-weakening lemma
     (elemAppendRight) — an algebraic closure law, not the Layer-2 decision.

Includes a sound+complete Dec (decExtendConformant), a positive control
(extendedGeneratedConformant) and a non-vacuity negative control
(extendedBrokenNotConformant : Not ...). Builds clean with zero warnings;
adversarial false proof rejected. No believe_me/postulate/assert_total.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A6PSzJWpRxtzGDjUCEh7Mx
Add Iseriser.ABI.FfiSeam proving the resultToInt encoding is sound:
- intToResult decoder + resultRoundTrip (lossless/faithful encoding)
- resultToIntInjective derived from round-trip (distinct outcomes never
  collide on the wire)
- positive controls (concrete decodes by Refl) and a machine-checked
  non-vacuity control (resultToInt Ok /= resultToInt Error)

Genuine total proof: no believe_me/postulate/assert_total/etc. Registered
in iseriser-abi.ipkg; package builds clean with zero warnings.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A6PSzJWpRxtzGDjUCEh7Mx
@hyperpolymath hyperpolymath marked this pull request as ready for review June 28, 2026 06:44
Signed-off-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com>
@hyperpolymath hyperpolymath merged commit 2e9d19f into main Jun 28, 2026
4 checks passed
@hyperpolymath hyperpolymath deleted the claude/new-session-znxgm7 branch June 28, 2026 06:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants