Fix failure-isolation event attempt_index

[chris-colinsky]

Summary

Fix FailureIsolatedEvent.attempt_index to report the wrapped node's final / exhausting attempt when FailureIsolationMiddleware wraps RetryMiddleware, per spec's ruling on the attempt-index reconciliation. Previously it reported the post-reset baseline (0), because RetryMiddleware resets the attempt-index ContextVar in its per-iteration finally as the exhausted exception unwinds, before the outer isolation middleware catches it.

This is the last v0.14.0 build piece tied to the failure-isolation conformance work; it un-defers fixture 061.

How

An isolation-scoped _terminal_attempt_index ContextVar carries the final attempt past retry's reset:

• FailureIsolationMiddleware.__call__ establishes a fresh scope (None on entry, reset in a finally).
• RetryMiddleware, on give-up (exhaustion or a non-transient error), records the final attempt into that scope before re-raising.
• _emit_event reads the recorded terminal index, falling back to the live attempt index when no retry exhausted (so the no-retry fixtures stay at 0).

The blast radius is limited to isolation's attempt sourcing: retry's control flow, the live per-attempt scoping, and the nested-retry token stack are all unchanged.

Spec

The contract was already in the spec (§6.3 lineage correlation + fixture 061); the prior impl comment claiming the baseline was "spec-confirmed" was unsubstantiated and is removed. No spec change or proposal was needed.

Validation

• Fixture 061 now passes; the three-piece unit test asserts attempt_index == 2 (final of attempts 0/1/2); the no-retry fixtures (058-063) confirm the fallback to 0.
• Full unit + conformance suite: 1252 passed, 0 failed; ruff + pyright clean; conformance-manifest guard OK.
• No CHANGELOG entry: this corrects unreleased 0050 behavior, and the existing FailureIsolationMiddleware entry already describes the event carrying the wrapped node's lineage.

[Copilot]

Pull request overview

Fixes FailureIsolatedEvent.attempt_index so that when FailureIsolationMiddleware wraps RetryMiddleware, the event reports the wrapped node’s final/exhausting retry attempt index (per proposal 0050 §6.3), rather than the post-reset baseline.

Changes:

• Add an isolation-scoped “terminal attempt index” ContextVar in correlation primitives and plumb it through failure isolation event emission.
• Record the terminal attempt in RetryMiddleware on give-up so outer failure isolation can read it after retry resets attempt_index.
• Un-defer and extend tests to assert the corrected attempt_index behavior (including conformance fixture coverage).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tests/unit/test_failure_isolation_middleware.py	Adds an observer capture and assertion that the failure-isolation event reports the exhausting attempt index.
tests/conformance/test_pipeline_utilities.py	Removes the deferral for the failure-isolation three-piece composition fixture.
src/openarmature/observability/correlation.py	Introduces terminal-attempt ContextVar helpers and exports them for use by retry/isolation middleware.
src/openarmature/graph/middleware/retry.py	Records the terminal attempt index on give-up before re-raising.
src/openarmature/graph/middleware/failure_isolation.py	Establishes/reset terminal-attempt scope and uses it when emitting FailureIsolatedEvent.
conformance.toml	Updates proposal 0050 “partial” commentary to reflect passing failure-isolation fixtures and the new attempt-index behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.