MVP2 backlog batch: caplog-isolation fix + 6 cross-model-reviewed spec/plan pairs

[SoundMindsAI]

What this PR does

Drives the next 7 highest-priority MVP2 idea-stage items forward in one batch (one branch, per the single-branch-per-session convention). Item #1 was a bug (fixed for real); items #2–#7 are taken from idea.md → feature_spec.md + implementation_plan.md via idea-preflight → spec-gen → impl-plan-gen, each with the mandatory GPT-5.5 cross-model review. No feature is implemented here — the six are planning artifacts.

Code changes (2 commits)

fix(logging) — e4707cd0 — closes bug_backend_suite_nondeterministic_caplog_isolation (P1)

The backend suite was nondeterministically red: capture_logs()-based unit tests asserted on [] in the full suite but passed in isolation. The idea's pytest-randomly theory was wrong (that plugin isn't installed). Real cause: configure_logging() handed structlog a brand-new processors list instance on every call; loggers cache against the instance (cache_logger_on_first_use=True), so a second configure_logging() (e.g. an integration test's FastAPI lifespan) left earlier-bound loggers pointing at a stale list while capture_logs() mutated the new one → blind capture. Reproduced deterministically (8 failed → 75 passed). Fix: mutate the list in place. Also fixed a second polluter — a hand-rolled capture fixture in test_position_bias_prior.py doing configure(processors=[cap]) + reset_defaults(). Regression test added; bug_fix.md written.

fix(tests) — 2c287a48 — closes bug_contract_allowlists_outdated_after_mvp2_features

Three hand-maintained contract allowlists never got 'solr' (+ 5 new MVP2 endpoints) after the Solr adapter shipped, leaving the suite deterministically red on a clean tree. Updated test_resolve_engine_type_wire, test_health_contract subsystems, and EXPECTED_ENDPOINTS. Folded in (operator decision) so this batch's CI can go green. Contract layer: 325 passed.

Planning artifacts (1 commit, 09585893)

Feature	FR / stories	Notes
feat_apply_path_normalizer_declaration	8 / 7	Design-ahead — gated on feat_query_normalization_tuning Phase 1 merge + operator-friction evidence; plan refuses /impl-execute until both clear
feat_overnight_studies_summary_card	6 / 7	Phase 2 of shipped feat_overnight_autopilot; ready
feat_query_normalizer_typed_pipeline	9 / 8	Design-ahead — Phase 2 of feat_query_normalization_tuning; Story 0 hard precondition gate
infra_generated_artifact_freshness_gate	9 / 6	CI freshness gate; report-only until operator adds required check
chore_arq_pool_aclose_deprecation	4 / 2	preflight found a second call site (workers/all.py:225) + corrected the arq version claim
chore_cluster_detail_rung_badge	8 / 8	surface UbiRungBadge on cluster detail

Each idea.md was preflighted (stale claims corrected, dependency status recorded); each folder carries a pipeline_status.md through the Plan stage. Dashboards regenerated.

Not in scope

No production feature code. The two design-ahead specs (#2, #4) are explicitly not green-lit for /impl-execute until their Phase 1 dependency merges.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request resolves the non-deterministic test isolation issue (bug_backend_suite_nondeterministic_caplog_isolation) by ensuring that configure_logging mutates the existing structlog processors list in place rather than replacing it with a new instance. It also updates several contract tests to include Solr and other new endpoints, adds regression tests for the logging fix, and introduces several design-ahead specifications and plans for upcoming features and chores. A critical review comment points out that the rewritten capture_logs fixture in test_position_bias_prior.py will raise an AttributeError if run in isolation when structlog defaults to an immutable tuple of processors; a check should be added to handle this case gracefully.

[gemini-code-assist]

If test_position_bias_prior.py is run in isolation and configure_logging() has not been called yet, structlog.get_config()["processors"] will return a tuple (the default in structlog). Since tuple is immutable and does not have .copy(), .clear(), or .extend() methods, this fixture will raise an AttributeError and fail the test.

To prevent this, check if configured is an instance of list before attempting to mutate it in place. If it is a tuple (or other non-list type), fall back to configuring it directly and restoring the original value on teardown.

Suggested change

	configured = structlog.get_config()["processors"]
	saved = configured.copy()
	configured.clear()
	configured.append(cap)
	structlog.configure(processors=configured)
	try:
	yield cap
	finally:
	configured.clear()
	configured.extend(saved)
	structlog.configure(processors=configured)
	configured = structlog.get_config()["processors"]
	if isinstance(configured, list):
	saved = list(configured)
	configured.clear()
	configured.append(cap)
	structlog.configure(processors=configured)
	try:
	yield cap
	finally:
	configured.clear()
	configured.extend(saved)
	structlog.configure(processors=configured)
	else:
	structlog.configure(processors=[cap])
	try:
	yield cap
	finally:
	structlog.configure(processors=configured)

[SoundMindsAI]

Gemini Code Assist adjudication

#	Severity	File	Finding	Verdict
1	High	test_position_bias_prior.py:55	In-place mutation of get_config()["processors"] would raise AttributeError if the default is an immutable tuple	Accepted (as future-proofing) — premise is not reachable on the pinned structlog 25.5.0: verified structlog.get_config()["processors"] is a mutable list at default, and the test passes standalone (15/15). Adopted the suggested isinstance(list) guard anyway (commit 685cdebd) — cheap, harmless, and mirrors the same guard already in configure_logging().

No other review findings. CI: 11/13 checks green, 3168 backend tests pass. The 2 red checks (backend → only test_demo_seeding_ubi_full; smoke) are a pre-existing Solr-CI-readiness gap unrelated to this PR — captured as infra_solr_ci_readiness (P1). Merging over per maintainer decision.