Skip to content

v0.8.0 — Qualitative narrative extraction (get_narrative)

Choose a tag to compare

View all tags
@ajtgjmdjp ajtgjmdjp released this 17 Jul 12:21
· 11 commits to main since this release
v0.8.0
863cea1

Added

Qualitative narrative extraction (get_narrative) — read 有価証券報告書 text sections as plain text, from Python or any MCP client:

  • Sections: 事業等のリスク (business_risks), MD&A (mdna), 経営方針 (business_policy), 事業の内容 (description_of_business), コーポレート・ガバナンスの概要 (corporate_governance), 研究開発活動 (research_and_development)
  • EdinetClient.get_narrative() returns NarrativeSection with provenance (context_ref, doc_id, filing_date)
  • Paged MCP tool (max_chars/offset/next_offset) so long sections never blow the LLM context
  • Semantic XBRL context selection (filing-date instant → current-period duration → others; dimensionless preferred) with deterministic tie-breaks
  • HTML→text conversion preserving headings, paragraphs, lists, and tab-joined table rows; style/script dropped; no double-unescaping
  • Pre-parse 50 MB instance size limit; bounded per-(doc_id, section) cache; source_truncated flag — hardened via two-round external code review (gpt-5.3-codex)
  • Annual reports (有価証券報告書) only in this release

See CHANGELOG.md for details.

Assets 2
Loading