Reality probe: independent before/after witness for Continue #12576

[hippoley]

This is an independent CounterProof Reality Probe for:

continuedev/continue#12576

@christianalmonte112 — you bumped that PR on 2026-09-16 because the fix is still waiting for maintainer review. I tried to answer one narrow question that may help the review: does the regression test added by the PR actually distinguish the fix from the pre-change code?

Independent replay

Exact commits:

• HEAD: 0f2e62e54b6f5bc2bc749e22e15b0dd5342bde2a
• BASE: a3076b444af3fb0d9fd3805c10f75177d8f70137

Changed test:

• core/config/profile/doLoadConfig.vitest.ts

Result on a clean GitHub runner:

• HEAD + changed test: 5/5 PASS
• BASE + the same changed test: 4/5 PASS, 1 FAIL
• CounterProof status: WITNESSED
• result protocol: json-v1
• evidence digest: sha256:23b69b350eb966d65faf9616af31eb6c961a765e4db2672083cde7ca336b6866

Runner:
https://github.com/hippoley/CounterProof/actions/runs/35972260870

So the submitted Vitest file is not merely green on the PR branch: it contains behavior that fails against the exact pre-change base.

Scope

This does not mean “the PR should merge.”

It proves the submitted migration regression distinguishes HEAD from BASE. It does not replace maintainer judgment about API/design direction, backwards compatibility, or other config shapes beyond the test coverage.

The earlier Cubic finding about arrays appears already addressed in current HEAD; this receipt is not a substitute for that review history.

Something this PR taught CounterProof

This external PR broke CounterProof twice before producing a receipt:

1. *.vitest.ts was not recognized as a changed test.
2. the temporary BASE worktree reused only root node_modules, while Continue resolves Vitest from core/node_modules.

Both gaps are now fixed in CounterProof main because this PR exposed them:

• fix: discover Vitest-named regression files #19
• fix: reuse package-local dependencies in monorepo base replay #21

Question

Would an independent before/after receipt like this be useful when you are trying to get a stalled PR reviewed, or is it mostly noise compared with the project’s normal CI?

Even “useful / not useful + why” is enough — I am using real contributor/reviewer reactions to decide what CounterProof should become.

Readable receipt:
https://github.com/hippoley/CounterProof/blob/main/reality/continue-12576/WITNESS.md

Assertion-level reproduction

A second clean run reproduced the same semantic split and exposed the exact failing BASE assertion:

• HEAD: 5/5 pass
• BASE + same changed test: 4/5 pass, 1 fail
• failing test: should transform a JSON-style tabAutocompleteModel object into a roles-based model entry
• BASE failure: the passed YAML content still contains tabAutocompleteModel, exactly the behavior #12576 is intended to migrate away

Second runner:
https://github.com/hippoley/CounterProof/actions/runs/35975618130