test(golden): graduate hard-query xfails to hard assertions

[silversurfer562]

The hard-difficulty golden queries have carried xfail(strict=False) since the marks were added to document the keyword-retriever gap, gated on an embeddings retriever (task 2.5). Two facts make the marks obsolete:

• All three hard queries pass on main today (they showed as XPASS) via alias expansion + path-keyed summaries — the marks only mask future regressions.
• The embeddings retriever they were waiting on is permanently deferred (docs/specs/archive/alias-expansion-sweep/README.md), so the "until embeddings" condition can never arrive.

Scope: removes _entry_marks so all difficulties assert hard, and records the graduation in the module docstring. The paraphrased info-only xfail set and the aggregate R@3 watermark are deliberately untouched — their threshold-gating decision remains deferred per M13.3.

Verification: golden suite 41 passed / 2 xfailed (known paraphrased misses gqp-015b, gqp-031a) / 78 xpassed; full suite green locally on top of #217.

Supersedes the stale May branch fix/golden-drop-stale-xfail (deleted in today's housekeeping sweep).

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
attune-rag	Ready	Preview	Aug 18, 2026 11:35pm

[github-actions]

Perf delta — within baseline

Blocking on regression in: keyword_retriever_retrieve.cpu, rag_pipeline_run.cpu. Other metrics are advisory and don't block merge.

Metric	Baseline mean (s)	Current mean (s)	Δ	Threshold (s)	Status
directory_corpus_load.cpu	0.000060	0.000052	-13.3%	0.000089	ok
directory_corpus_load.wall	0.000060	0.000052	-13.3%	0.000089	ok
keyword_retriever_retrieve.cpu	0.005573	0.003761	-32.5%	0.007221	ok
keyword_retriever_retrieve.wall	0.005573	0.003761	-32.5%	0.007219	ok
rag_pipeline_run.cpu	0.000639	0.000559	-12.5%	0.000919	ok
rag_pipeline_run.wall	0.000638	0.000559	-12.4%	0.000918	ok

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!