Skip to content

Retire the model parts the shipped client never runs; v0.5 refinement pass - #141

Merged
afflom merged 9 commits into
mainfrom
claude/project-thread-y1kbnt
Nov 12, 2026
Merged

afflom merged 9 commits into
mainfrom
claude/project-thread-y1kbnt

Conversation

@afflom

@afflom afflom commented Nov 11, 2026 •

Copy link
Copy Markdown
Member

Requested by Alex · project thread

Ready to merge. It contains main at cd43ca8 (#149) and goes before the joint export, so every later CI round builds 89 modules instead of 342.

What

The import graph rooted at the shipped client (Uorc.Client), its executed search and step-machine controls, the SHA-256 contract, Uorc.Page and Uorc.Neuromorphic reaches 89 of the 342 modules under src/. The other 253 modules (446,090 lines) model a v0.1 product that the client, library and page do not implement:

  • a format3/machine1 archive bound to UORC-SHA256-STREAM-MAX-v1 (Integrity, 35 modules)
  • a runtime machine with heap and admission (Core, 104)
  • a workbench application with resume, compare and basis-building requests (Product, 32)
  • owned and streamed copies of the search and wire layers (Search 39, Wire 43)

The formal legs built and verified all of them on every change.

This PR removes those modules and the following:

  • rows UC-SHA-02 and UC-INT-02..05, with their tests
  • the integrity binding, stream, transcript, canonical and consumer harnesses
  • the product mutation scripts
  • validation/{product,integrity,integration}
  • the diagnostic trees under tests/
  • docs/implementation

docs/verification/pages-client.md and frame-domain-audit.md stay, because they are current evidence. product-machine-runtime.mjs becomes executed-controls.mjs, which keeps only the control runner. The integrity obligation's closure becomes Uorc.Integrity.*. UC-SHA-01, UC-INT-01 and every client row keep their tests.

Refinement pass

  • The charter's status and scope, and the UC-OPT-01 statement, now describe the shipped client. That covers per-frame portfolio coding chosen by outcome (bytes, then decode steps), named model states, the x86 filter, reference frames, the raw floor and optional search under a step budget. README, UORC-VERIFICATION and CONFORMANCE are regenerated.
  • The unused template example examples/TaskScorer is removed.
  • The language-1.1 source-format preflight is removed, along with its fixtures and its 3 runs per offline-validation workflow. lexlean.toml selects language 1.2, where that guard gives no verdict, so it checked nothing.
  • Both release gates were checked on main b29ae74. An x86 trial is excluded when the filter rewrites nothing (UC-PERF-07). Byte ties go to fewer decode steps (UC-PERF-08).

Performance check before retiring

None of these parts carries a size or speed gain. Measurements are from the release build at efd3d17 on 4 cores.

  • STREAM-MAX halves the SHA-256 calls. SHA-256 runs at about 1 GB/s and the coder at about 1 MB/s, so the whole seal is about 0.1% of the time.
  • Search. With --steps at 1e6 and 1e7, search found no smaller archive on 8 inputs. It took 20 to 233 s per file.
  • Runtime machine. It runs only program archives, and the client never writes one.

Verification

  • cargo xtask check-model is clean with 36 ids. xtask validate passes every check up to audit-bootstrap, which needs the SDK image.
  • lexlean build --all and verify --all (LexLean 0.4.0, Lean 4.32.1) verified 89 modules in 838 s at width 4.
  • The node tests and cargo clippy --workspace --all-targets are clean.
  • CI was green on the previous head.

Generated by Claude Code

The import graph of the shipped client (Uorc.Client), its executed
controls, the SHA-256 contract, the page and the neuromorphic standard
reaches 89 of the 342 modules under src/. The other 253 (446,090 lines)
modelled a v0.1 product the client, library and page do not implement:
a format3/machine1 archive bound to UORC-SHA256-STREAM-MAX-v1, a runtime
machine with heap and admission, and a workbench application with
resume, compare and basis-building requests. They were built and
verified by the formal legs on every change.

Removed with them: rows UC-SHA-02 and UC-INT-02..05 and their tests,
the integrity binding/stream/transcript/canonical/consumer harnesses,
the product mutation scripts, validation/product, validation/integrity,
validation/integration, tests/ diagnostics, docs/verification and
docs/implementation. UC-SHA-01, UC-INT-01 and every client row keep
their tests; the executed search and step-machine controls stay.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
@afflom afflom self-assigned this Nov 11, 2026
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
…lient.md, the page's evidence

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
… and Cursor ship in the client

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
…op the unused template example

The charter's status and scope still described the v0.1 source
(runtime semantics, resumable search, file-oriented interfaces). They
now name what ships: the CLI, the library and the WebAssembly page,
per-frame portfolio coding chosen by outcome, named model states,
reference frames and the raw floor. UC-OPT-01 names the candidate set
the client checks today and the bytes-then-steps order. The
examples/TaskScorer scaffold was never built or read by any gate.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
…e 1.2, where it gives no verdict

The guard predicted LLP2003 for native notation under language 1.1 (issue #86). lexlean.toml selects language 1.2, for which the script delegates every input unchanged, so its three offline-validation runs and fixture suite checked nothing. lexlean fmt/check remain the source-format authority.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
@afflom afflom changed the title Retire the model parts the shipped client never runs Retire the model parts the shipped client never runs; v0.5 refinement pass Nov 12, 2026
@afflom
afflom marked this pull request as ready for review November 12, 2026 15:16
@afflom
afflom merged commit 870a46e into main Nov 12, 2026
17 checks passed
afflom pushed a commit that referenced this pull request Nov 12, 2026
The hints listed 340 Uorc modules; 87 exist. lexlean verify reads hints only for modules it builds, so the stale entries changed nothing but the file.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QdiL6gUZQx4FxLkEH
afflom pushed a commit that referenced this pull request Nov 12, 2026
lexlean reads hints only for modules it builds, so results are unchanged.
Carries the change of #156.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3LszhHCxmceQ2V8xxx
afflom added a commit that referenced this pull request Nov 12, 2026
Bumps the workspace and internal dependency versions to 0.5.0 and drops the verify hints of the 253 Lean modules retired in #141.
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