feat: import bounded external test evidence

[YuzeJ21]

Summary

• add a bounded, local, bytes-only JUnit XML parser that never executes target-repository code, follows artifact references, or fetches remote resources
• require explicit human criterion mappings and bind every validated import to repository, PR, exact 40-character head, ordered confirmed-criteria digest, and criteria-source provenance
• persist imported test results as a versioned Pydantic envelope separate from E1/E2 candidates, E3/E4 runtime evidence, GitHub CI observations, human resolutions, final acceptance, and deterministic gate truth
• cap each review at 20 retained imports and each warning or limitation at 1,000 characters; a new analysis cannot inherit imports from an earlier criteria revision
• expose one shared core workflow through CLI and Streamlit, with atomic saved-review mutation, reopen validation, safe exports, and comparison-aware provenance
• cover the installed-wheel import, inspect, save, reopen, and safe-export round trip in loopback-only Chromium
• retain the exact 1 MiB widget/parser boundary while allowing multipart transport overhead; require Streamlit 1.53+, the first compatible per-widget implementation

Evidence boundary

Imported JUnit results are externally supplied, non-gating context. ScopeProof did not execute the tests or target-repository code. The artifact digest identifies imported bytes only. Importer identity is asserted, not authenticated. Explicit criterion mapping is organizational context, not proof. An import never changes a finding or decision, satisfies E3/E4, creates or preserves Ready, records final acceptance, or carries a decision to another head.

This pull request is engineering evidence only. It does not establish correctness, customer validation, accessibility conformance, authenticated identity, broad platform support, demand, adoption, or Stage 3 activity.

Verification

• Ruff: passed
• complete suite: 2,495 passed, 2 intentional skips
• combined coverage: 95.11%
• repository contracts: 125 passed
• final re-analysis lifecycle regression: passed red-to-green; rejected state remained unmodified
• deterministic acceptance benchmark: 12 cases, zero mismatches, zero must-have False Ready outcomes, zero false blockers
• comparison benchmark: 2 cases, zero mismatches
• two byte-identical wheels: SHA-256 f6d888b5b033a045d2d8069b43cafafc3179436729d393e41a792e98f9bb051b
• artifact inventory: 104 expected entries, 1,425,047 uncompressed bytes, 306,631-byte wheel; .coverage 2, local state, worktrees, caches, and raw JUnit XML absent
• clean final-wheel Python 3.12.0 install: dependencies compatible, source/installed version 0.2.4.dev0, both CLIs correct, and both installed benchmarks returned zero mismatches; final-head hosted Python 3.11 and 3.13 compatibility lanes and the Windows package/storage/CLI lane are green
• installed-wheel Chromium: final repaired-head run passed 5/5; the preceding synchronization repair also passed three consecutive fresh 5/5 runs, with loopback-only networking and no captured external requests, console errors, or page errors
• exact installed loopback workbench health returned ok; the server then stopped and the endpoint was unreachable
• final diff and commit audit: clean

Review

Automated review surfaced four P2 findings after initially green heads: a criteria-revision reset could preserve stale imports, individual limitation strings were unbounded, accumulated imports were unbounded, and re-analysis attachment could preload imports outside the append-only lifecycle. All four were reproduced with failing regressions and repaired in the shared core. No unresolved Critical or Important findings remain; all hosted checks are terminal green except the two intentional skips.

Product stage truth

• Stage 1: closed_not_pursued_by_owner
• 0/5 qualifying reviews
• 0/3 independent practitioners
• 0/3 public repositories
• 0/3 independently observed under-ten-minute completions
• 0/2 reuse-intent signals
• Stage 2: owner_led_productization_active
• Stage 3 has not begun

Unsupported environments

Local installed-browser evidence is Chromium on macOS/POSIX. Hosted compatibility lanes provide Python and package/storage evidence, not real desktop workflow evidence. No real Windows or Linux desktop workflow, real screen-reader workflow, or accessibility-conformance claim is included.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1c6f4b75ab

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dc4c34e082

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1a527ee777

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 40ad63eb08

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".