37 lines
2.2 KiB
Markdown
37 lines
2.2 KiB
Markdown
# Changelog
|
|
|
|
## 0.2.0
|
|
|
|
### Reader-facing output
|
|
|
|
- Split reader-facing Markdown from internal `provenance.md` and `evidence-map.json`.
|
|
- Default technical-blog citations to `hidden` so source IDs, local paths, access dates, and prompt scaffolding do not appear in the article.
|
|
- Added lint rules for evidence-process narration, internal markers, repository paths, and date boilerplate.
|
|
- Added a curated Korean `application-core` golden example with no unsupported SLF4J rationale.
|
|
|
|
### Evidence retrieval
|
|
|
|
- Added local repository collection for `wiki/projects`, `wiki/concepts`, `raw/branch-notes`, `raw/official-docs`, and `raw/company-tech-blogs`.
|
|
- Added source hierarchy, status, heading, line range, claim IDs, decision IDs, and retrieval priority.
|
|
- Added rationale-oriented ranking so constraint, reason, alternative, cost, and guardrail evidence outranks name-only matches.
|
|
|
|
### Writing and review contracts
|
|
|
|
- Replaced the technical-blog sequence with problem scene → constraints → options → decision rationale → mechanism → verification → trade-offs → conclusion.
|
|
- Added the corpus-derived `woowahan_tech_blog_ko` profile; it is explicitly not represented as an official company house style.
|
|
- Added a dedicated decision reviewer and review dimensions for decision rationale, source usefulness, and reader-facing prose.
|
|
- Added a separate editor reviewer for opening strength, paragraph focus, transitions, repetition, terminology, and canned LLM phrasing.
|
|
- Added revision instructions that remove unsupported intent rather than inventing a plausible reason.
|
|
- Separated semantic information order from sentence form using an eight-article Woowahan Tech Blog sample, and added `STYLE001` for repeated abstract ordinal paragraph openings.
|
|
|
|
### Verification
|
|
|
|
- Expanded the suite to 44 tests.
|
|
- Added regression checks for the exact leakage and missing-rationale failure classes.
|
|
- Added local-corpus, golden-example, provenance, manifest, wheel-build, and clean-install smoke tests.
|
|
- Added prompt and lint regressions for sentence-form guidance while preserving genuine ordered procedures.
|
|
|
|
## 0.1.0
|
|
|
|
- Initial contract-first pipeline with deterministic document structures, multi-provider adapters, lint, reviews, revision, quality gate, and artifact manifest.
|