Skip to content

v0.2.0 — first SemVer-binding cut

Choose a tag to compare

View all tags
@silversurfer562 silversurfer562 released this 25 May 03:52
· 36 commits to main since this release
v0.2.0
81a54a5
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

attune-rag 0.2.0 — first SemVer-binding cut

Cut via the freeze-override mechanism (W4.2 hard gate). Phase 4's review/gate deliverables landed ~3 weeks ahead of the nominal calendar; rather than wait the remaining cadence-soak weeks for calendar-elapse alone, the user-facing additions shipped via the per-PR override mechanism.

From 0.2.0 forward, docs/POLICY.md §2 binds — symbols PUBLIC in 0.2.x stay PUBLIC through every 0.2.z. Classifier stays at 3 - Alpha (the Production/Stable flip is a Phase 5 deliverable).

Highlights

Added — new public surface

  • attune-rag-measure console script + attune_rag.measure_corpus public module (#136) — quality harness for benchmarking your own RAG corpus. Public measure(...) Python API + MeasureResult frozen dataclass with full per-query/per-difficulty breakdowns, CI-suitable --watermark-r3 flag on the CLI.
  • attune_rag.corpus.load_aliases_from_file() + DirectoryCorpus(extra_aliases_file=...) kwarg (#130) — load path-keyed extra aliases from a JSON file. Schema mirrors attune-rag's own aliases_override.json.

Changed — perf methodology + verdicts

  • V2 multi-run perf methodology (5 invocations × 20 runs = 100 measurements per metric) — replaces the V1 single-invocation baseline. Threshold formula is now mean + σ × inter_run_stdev, separating intra-run from inter-run noise.
  • D5 reranker-evaluation closed — verdict rerank-default-off. Ratifies the existing RagPipeline.reranker=None default with N=5 live measurement against the bundled corpus + golden query sets.
  • Perf gate promoted advisory → blocking on CPU-time axis of KeywordRetriever.retrieve + RagPipeline.run (Phase 4 W3.1).

Quality baselines (now binding)

  • precision_at_1 ≥ 0.95
  • recall_at_3 = 1.00
  • mean_faithfulness ≥ 0.9686 (σ ≈ 0.005)

Same discipline powers attune-author's polish/fact-check pipeline — attune-rag's faithfulness primitives are the contract the attune-* family's content-quality story is built on.

Override rationale + per-gate evidence

Full Phase-4 exit summary with the override rationale, per-gate status, and what-was-traded matrix:
docs/specs/downstream-validation/exit-summary.md

Install

pip install attune-rag                     # core
pip install 'attune-rag[attune-help]'      # + bundled help corpus
pip install 'attune-rag[claude]'           # + Claude adapter
pip install 'attune-rag[gemini]'           # + Gemini adapter

Full changelog

Complete entry list (including all ### Changed items not summarized above) at CHANGELOG.md.

Follow-up: 7-day no-hotfix watch starts on this publish. Phase 5 specs (reranker-evaluation, user-corpus-onboarding, perf-baseline-multi-run) activate after the watch closes clean.

Assets 2
Loading