Integrate proven public code repair and cooperative compute foundations

[erikleblansch]

Integrated in main (80e79327)

Merged normally at 2026-10-04 19:01:07 UTC after all exact f064e990 source checks succeeded: Quality 37225082109 and CodeQL 37225079526. The manual-only KVM check was skipped as expected; no new live trial was dispatched.

Verified main commit 80e7932718e79fb4c853b943bd6d5f11e50bda81, tree 95ba20214fe5cc4d601315a6df3a8fee71ae4c9d, with exact parents 4cda03c622bdcee4bebddde3aa6f235df384dafa and f064e99083263020d6168489a312d346e3526050. This integrates the bounded, publicly consented Node-owner/remote-model/file-edit/original-test milestone. It does not claim native editor UI/planner success, confidential private remote execution, general coding quality, or the separate newer diagnostic work. Earlier failed runs and their original outcomes remain below.

Integration source-check follow-up (f064e990)

The original quality run 37223838526 on f38fcbd7 remains failed at strict Clippy. Its five findings were confined to test spellings/layout; this commit changes no production behavior, assertions, budgets or acceptance gates. Original job-log SHA-256: b0acddfd22b400a18e6a9fb5a77aba464e6d10b5c3c6a7c96e376aeb7b809912.

Targeted checks: 2 coordinator tests, 4 diagnostic tests, 1 explicit Qwen4B-profile test and 2 agent code-proposal tests pass; exact-file rustfmt and diff checks pass. The test module was moved unchanged after the unchanged production function. New source CI is pending; no additional live model or KVM trial was started. The immutable public execution proof and its limitations below remain unchanged.

Development milestone and source binding

Merge the proven public-code source 2a1b4ad347b7d9f12a6a4c2beee40ff8706bd477 (including the histories behind #197, #188 and #178) into current main 4cda03c622bdcee4bebddde3aa6f235df384dafa. This PR preserves both parents; it does not merge the independent generation-progress candidate or Code PR #5.

The original Code run37221727043, attempt1 / job111493248899, passed every original public-proposal gate. Exact workflow source: 54af18cc5d30a880e3be298192af150daf01b805; immutable Node owner driver: f27576ebd7e7ded2f1319186f34df87f48e970d7.

A real discovered Qwen3-0.6B peer produced the unmodified 31-byte / 12-token EOS replacement. Initial tests failed; explicit owner edit and test approvals then passed the same three original tests and independent verification. Original source/model/dataset/receipt binding, separated discovery/task privacy captures, selected-executor traffic, byte-preserving control observation and complete guest cleanup passed. Zero owned objects remain; guest before/after hashes match.

This proves a bounded public Node-owner-helper -> real peer -> approved file edit -> original-test chain. It does not prove native editor UI/OpenCode autonomous planning, general coding quality, full immune review or confidential private-peer execution. Native/private candidates retain their separate incomplete status. Earlier failed runs37213737334,37216555196,37218917021 and37220221345 remain failed, with original evidence retained.

Explicit conflict resolutions

• Shared workflow and VM scenario lists preserve main's Cloud upload and Signal fragments plus the cooperative/native scenarios. Existing budgets, closed export inventories and argument scopes are unchanged.
• The benchmark diagnostic dispatcher also retains main's cloud-private-upload prefix, with an added real-shell contract case; selection and retry behavior are unchanged.
• KVM preview/cleanup/scenario registration retains both histories.
• Status and Unbound documentation preserve main's reader organization, Bank research, DNS/package evidence and feature history; the latest genuine public pass is clearly bounded.
• Product compute/broker/IPC/worker/content/helper bytes are identical to proven2a1b. Differences in crates/ against that source are main's test/doc-path/comment updates only.
• The old single-clause timeout-test regex could not parse main's grouped Signal expression. The replacement checks the exact closed expression, preserving native240 / aggregation180 / both Signal150 / default120 minutes.

Narrow verification

• Public proposal18, actual local socket observer13, cooperative VM contract5 and bundle3 tests: PASS.
• Cloud upload report16 / wiring5 / readiness8; Signal fragments4 / VM wiring4: PASS.
• Benchmark15 plus the new targeted Cloud-upload diagnostic contract: PASS.
• Existing inert topology static contract, changed shell syntax and focused ShellCheck: PASS. These checks do not execute a model or VM.
• Original run37221727043 smoke report revalidated by the unchanged checker on this integration: PASS. Five exact-byte scenario/observer/cleanup/capture-source comparisons against2a1b: PASS.
• Changed documentation local links and diff checks: PASS. No additional local Rust build, model execution, host networking mutation or live workflow dispatch.

Normal main-target source checks should now evaluate this combined revision. Replaying the original proof does not turn it into a fresh live test of the merge.

Original evidence digests

• ZIP (36 files): 8a17e92c2bc4148093651a020cf0a2daaf46b2e14fcad74a4e4343024de02d11
• Original job log: 0dbc4650b89ba5d6bb8a3474f886ff4016feb1a0042d88b52c9e5391776fee26
• Original smoke report: 514cdae4d76d06aa8b0441d1937e80dce1b92769e8f0fe0d819fc0af1e965235
• Matching guest network state: e36d3544b51972d62c5f0393048a69643dbcb6d318126ad4f15016ea795abf59