Skip to content

Releases: golearnalanguage/DBabel

DBabel v1.5.0

Choose a tag to compare

@golearnalanguage golearnalanguage released this 24 Sep 16:53
  • Added DBabel Review Workbench for explicit human-in-the-loop bilingual review.
  • Added .dbreview Review Sessions, review schemas, evidence/issues/decision
    contracts, and local/portable review surfaces.
  • Added DOCX native export with Export Gate, original-file hashing, anchor checks,
    non-destructive output, and round-trip verification.
  • Bound Portable Review decision exports to their originating Review Session and
    made malformed or cross-session imports fail closed.
  • Prevented bilingual DOCX extraction from silently dropping unmatched paragraphs
    when source/target paragraph counts differ.
  • Added an Agent-to-Workbench handoff contract so suggestions and findings never
    become fabricated human approval.
  • Added explicit review-only versus native-export mode feedback.
  • Added explicit auditable DOCX alignment maps for ALIGNED, SPLIT, MERGED,
    UNALIGNED, and AMBIGUOUS relationships, with complete paragraph coverage
    checks and fail-closed native write-back for non-ALIGNED units.
  • Strengthened DOCX anchors with Word w14:paraId identity and fail-closed
    behavior when a previously anchored paragraph ID disappears.
  • Added DOCX structure guards for field codes and tracked revisions while
    regression-testing paragraph extraction inside tables, hyperlinks, and
    content controls.
  • Made desktop bulk review updates transactional, preserved reviewer notes,
    prevented hidden selections from leaking across filters/pages, and retained
    safe local bulk behavior in Portable Review.
  • Removed or explicitly disabled remaining inactive Workbench controls and
    completed Portable Review copy/theme compatibility hardening.

DBabel v1.4.0 — Accuracy Core and Progressive Runtime

Choose a tag to compare

@golearnalanguage golearnalanguage released this 24 Sep 02:04

DBabel v1.4.0 introduces the Accuracy Core and a progressive, format-aware runtime for database terminology review and technical-localization QA.

Highlights

  • Progressive Resource Router: agents load only the references and worked examples required by the current task state.
  • Accuracy Core: deterministic bilingual QA for protected literals, placeholders, URLs, paths, filenames, CLI options, environment variables, versions, numbers, and units.
  • Project glossary governance: JSON/CSV validation with scope-aware enforcement of PROJECT_APPROVED entries only.
  • Format-aware preflight: content-based format detection, capability probing, backend selection, and fail-closed extension/content conflict handling.
  • Ingest validation: parser selection is separated from actual document coverage.
  • Controlled repair: detected issues do not authorize edits; repair remains evidence-, authorization-, and round-trip-QA-gated.
  • Cross-contract validation across schemas, routers, examples, workflow configuration, glossary templates, reports, and package metadata.
  • CI coverage on Python 3.9 and 3.11 across Ubuntu and macOS.

Compatibility

DBabel Core remains Python 3.9+.
Audit reports produced by DBabel 1.3.0 remain accepted by the 1.4 report contract.

Notes

DBabel remains data-light and does not bundle vendor terminology corpora, proprietary translation memories, or mandatory third-party document parsers.

The v1.4.0 tag identifies the immutable release commit. Later README improvements remain post-release documentation changes on main.

DBabel v1.3.0 — Review workflow and agent setup

Choose a tag to compare

@golearnalanguage golearnalanguage released this 23 Sep 05:32

DBabel 1.3.0 improves installation, evidence tracking, and the workflow from terminology review to a verified corrected copy.

Changes

  • Added complete installation and invocation examples for Codex, Claude Code, and other agents, plus a Chinese usage guide.
  • Connected findings to source records and added coverage, QA results, repair outcomes, and completion status to structured reports.
  • Moved final reporting after repair verification and added feedback paths for failed checks.
  • Added report validation, regression tests, and macOS/Linux CI.
  • Removed personal repository setup and publishing instructions from the package.

Download

Download DBabel_v1.3.0.zip and extract the dbabel-database-terminology-audit folder. Keep SKILL.md and its supporting directories together. See the included README for installation and invocation.

DBabel_v1.3.0.sha256 contains the archive's SHA-256 checksum.

Validation

The extracted release archive passed Skill metadata validation, package checksum verification, and all 36 regression tests.

Report compatibility

Version 1.2 reports need the additional evidence, coverage, QA, and repair fields to validate against the 1.3 report schema. See the migration notes.

DBabel v1.2.0 — Initial Public Release

Choose a tag to compare

@golearnalanguage golearnalanguage released this 23 Sep 04:16

DBabel v1.2.0 — Public Skill baseline

Positioning

DBabel is an evidence-driven Agent Skill and methodology for database terminology review and localization workflows. It ships the method, not a vendor terminology database.

Included

  • normative SKILL.md;
  • end-to-end Agent state machine;
  • terminology classification and protection rules;
  • vendor/product/version/context scoping;
  • authoritative-source search and evidence policy;
  • cross-vendor concept discipline;
  • structured-document handling policy;
  • terminology-aware MT/translation workflow;
  • QA and release gates;
  • project-local glossary governance rules;
  • technical-claim routing boundary;
  • JSON schemas and YAML machine-readable configuration;
  • synthetic examples and regression cases;
  • GitHub issue/PR templates;
  • security, third-party, disclaimer, publishing, and licensing documentation.

Licensing

DBabel v1.2.0 is released for noncommercial use under PolyForm Noncommercial License 1.0.0.

Official terms: https://polyformproject.org/licenses/noncommercial/1.0.0

Not included

  • vendor terminology datasets;
  • scraped glossaries;
  • copyrighted manuals;
  • licensed standards text;
  • proprietary translation memories;
  • bundled MT/LLM services;
  • a production CAT application.