fix(observability): preserve skipped and failed replay states

[GuddXzy]

Scope

Scope boundary: Make read-only replay transcripts distinguish successful, conditionally skipped, and failed pipeline steps after decision logs have been written and loaded. Persist a small outcome field before privacy filtering removes exception text, using additive decision-log schema version 17. Historical logs without enough information to distinguish a failure from a skip display UNKNOWN.

Non-goals: Change pipeline execution, expose exception text in persisted logs, or add a WebUI surface.

Branch

Base branch: main

Target exception: N/A

Issue

Linked issue: Fixes #1503

If None, reason: N/A

Release Note

Release note: Replay now displays SKIPPED for conditionally skipped pipeline steps and preserves FAIL for exceptions even when their text is removed from logs. Historical ambiguous records display UNKNOWN.

Tests

Ruff: ruff check src tests — passed.

Type checking: mypy src/opensquilla --show-error-codes — passed for 1,605 source files.

Pytest (Python 3.12, explicit local source path):

• Targeted replay, decision-log contract, and privacy preflight — 35 passed.
• python -m pytest -vv --tb=short tests/test_observability — 258 passed.

Regression tests: added. Coverage includes a real pipeline with a gated production step, successful and failing steps, empty exception text, private primary/debug JSONL persistence, hydration, and the replay CLI. Historical records with missing, null, empty, explicit, and unknown status values are covered.

The default tests remain offline, deterministic, credential-free, and safe for forks. The local full repository suite was not run; repository CI provides the required acceptance checks. No WebUI or packaging source changes are included.

Maintainer Live Check

Maintainer live check: no

Surface: N/A

Maintainer-only note: No credentialed provider call is needed for decision-log persistence and read-only replay validation.

Safety

Exception text remains filtered from primary and debug logs. The outcome uses an existing safe metadata field; privacy rules are unchanged. Historical ambiguous records are not inferred to be successful or skipped. The added dataclass field has a default and older readers ignore it. The change is platform-neutral and does not change filesystem, subprocess, or platform-specific behavior.

No secrets, local-only artifacts, private prompts/transcripts, channel identifiers, AI session artifacts, non-public fixtures, or tests/_private/ contents are included.

Third-Party Origin

Third-party origin: none

Details if non-none: N/A

Documentation Changes

Updated docs/diagnostics-and-replay.md to explain replay outcomes and historical UNKNOWN records.

• Links point to existing repository files or stable external pages.
• Code fences and Markdown tables render correctly on GitHub.
• Examples avoid real secrets, local private paths, and private transcripts.

[GuddXzy]

Thanks for the focused fix and regression tests. I checked the current main branch, and skipped steps are still rendered as FAIL(None), so #1503 remains unresolved.

CI, PR body lint, and PR target branch currently show action_required, with no CI jobs executed. Could a maintainer check whether these workflows need approval and review this PR? The branch is also behind main; please advise whether an update is needed before validation.

[github-actions]

Merge queue CI: success — run 35167683077, attempt 1.
Tested queue candidate: f78d48286393dae95d5e760641e110e3f20aabaf. This result belongs to that combined tree, not necessarily the PR's current head.

The run summary lists reused suites and executed checks. A green PR check alone does not mean queue validation has passed.