Skip to content

Project CMUX workload receipts into adaptive verification observations - #1100

Merged
teamleaderleo merged 8 commits into
mainfrom
issue-547-cmux-workload-observation-adapter
Sep 22, 2026
Merged

teamleaderleo merged 8 commits into
mainfrom
issue-547-cmux-workload-observation-adapter

Conversation

@teamleaderleo

@teamleaderleo teamleaderleo commented Sep 22, 2026 •

Copy link
Copy Markdown
Owner

Follow-up to #547 after manaflow-ai/cmux#13411 landed the canonical cmux-workload-result/v1 corpus.

What this adds

  • a pure cmux_workload_verification_adapter;
  • input: exact CMUX semantic-result bytes plus the already-correlated glaeda-cmux-workload-observation/v1;
  • rechecks source/profile/validator/result-digest/terminal-state binding before projection;
  • maps declared CMUX stages into Make SmolRunner an adaptive verification compiler: learn expensive work, compile execution plans, and reuse exact derived artifacts #547's bounded VerificationObservation vocabulary;
  • carries source tree, toolchain digest, observed SDK, architecture/resource class, exact runtime-input identity, aggregate output bytes, benchmark reuse class, semantic result, and named stage timings;
  • emits typed JSON plus human output from the same batch.

Conservative validity boundary

The current CMUX v1 result publishes source/toolchain/SDK/architecture and semantic product identity, but it does not publish build configuration or compiler flags as independent validity parents. The adapter does not infer those from CMUX commands, paths, or profile names. Repeated compile receipts can therefore rediscover reuse_exact_compiled_product, but the candidate remains observed / advisory until CMUX explicitly exports the missing validity inputs.

cmux.macos.app-host-test-shard@1 exact product input is projected as reuse with the exact runtime-input digest and rebuilt_before_test=false.

Repository-specific validation remains an ignored-but-recorded stage because Glaeda has no authority to move/reinterpret CMUX's semantic validator.

#13365 boundary

Selective-layer transport still lacks real consumer-specific transfer timing/byte evidence in the v1 semantic result, so this PR makes no selective-layer performance claim and does not change CMUX workflows.

Tests cover stage mapping, exact-product test reuse, digest/identity mismatch refusal, passed-result cleanup/artifact consistency, JSON/human rendering, and the key #547 behavior: three compile receipts generate a reuse candidate that stays advisory because configuration/flags are absent.

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.

1 participant