🧪 Identify core refusals and retained test observations

[taras]

Why

Selected core and Workflow tests can accept an unrelated error as the refusal they describe. Structural readiness and retained-result failures also obscure which observation failed. This delivers the assigned core/Workflow and navigation slice of #914.

What changes

Before: a matcher replacement, generic Files error or unrelated hostile-provider fixture error can satisfy broad failure checks; Workspace observation maps every read error to absence.

After: assertions identify the original inner assertion, provider-unavailable classification and named hostile-answer refusals. Workspace observation accepts only documented ENOENT as absence and preserves other failures by identity. SR7 signals first delivery explicitly; SA4 identifies occurrence and retained result. Storage-view cases use WAC27–WAC29, preserving crash-recovery WAC24–WAC26. FL37/FX4 and Git source/emitted-copy evidence have accurate navigation.

What must stay true

• EI7 retains installed-read success and provider restoration; hostile variants retain zero admissions and no replacement-body execution.
• Both SR7 integration and transport boundaries remain; SR8/SR10 stay unchanged from PR 🧪 Make lifecycle and publication tests observe complete outcomes #915.
• Both Syntax reads and records remain. Storage commit/rollback/capability lifetime and real crash/restart/history reconstruction remain separate evidence.
• No production, public API, dependency, runtime exclusion or alias changes.

How to verify it

• Explicit affected Deno files: 27 tests / 275 steps passed. Independent File/fixed-dispatcher, real Workspace crash and Git consumers: 10 tests / 87 steps passed. Final diagnostic/helper and lint-correction focused reruns passed.
• Seven controlled faults failed as intended: matcher replacing the inner assertion; generic Files error; SR7 integrated run-ahead; omitted transport demand gate; unexpected retained Syntax result; unrelated hostile fixture exception; swallowed Workspace read failure. Temporary faults were restored byte-identically, with first outputs and hashes retained.
• All eight affected portable files passed under Node and Bun. Node required authorized local IPC access for the existing tsx runner; first sandbox failures were preserved. Workflow SQLite cases retain their existing Deno-only eligibility.
• deno task lint, deno task check, pnpm exec tsc --noEmit -p tsconfig.node.json and deno task check:jsr passed; JSR ended with Success Dry run complete.

Exact local commands, dispositions, controls and outputs: /private/tmp/xmd-test-improvements-20261014/core-artifacts/handoff.md and adjacent artifacts. Architect PASS covers signed head 99862522eed42c44c32b050a1789be91244be6f6.

Scope

This slice was developed on #915, exact reviewed base 39317e4e97281fb867eaf320583302b9b5801a4d; original main was 62d1086703ec72ca6d3b6e90a596ee716c571237. #915 merged before publication. Current main b037f2243733f34ea53fe4c3df5718119eb7b5cc contains that commit and has its identical tree, so this PR targets main with exactly the ten-file core/Workflow slice. It does not close #914: packaging, Web and script diagnostics have separate owners.

Risks and limitations

Focused evidence establishes the selected assertions and preserves independent consumer boundaries; full runtime corpora remain CI-owned. The dependency merged before publication, so this PR targets main. Local evidence does not establish required remote checks, which remain the merge gate. No CI verdict was used or monitored during this work.

Scope confirmation

• Every changed file supports the purpose described above.
• Unrelated cleanup and formatting changes are excluded.
• Mechanical identifier and navigation changes are identified.
• The description matches the final diff and test results.

[github-actions]

PR #916: 🧪 Identify core refusals and retained test observations

10 files, +129 / -50

Scope

✅ PR scope looks good.

Structural

✅ No structural bloat detected.

Slop

✅ Slop indicators look low.

Static Analysis

✅ Oxlint found no issues.

Correctness

No extraneous code patterns detected.