[pie575]

This follow-up contains the remaining integration and comparison changes across eight KB articles. The original eleven issue-specific commits remain linked below for history; subsequent editorial corrections and fixes now included in #653 are reflected in the current diff.

Stack: #648 → #653 → this PR. The base is #653’s branch, kb/gt-vs-leaderboard-tools-fixes, now at 4193e2c5. The original review commits remain in history; some fixes are now included in the parent. Review the current diff for the remaining follow-up changes. No new article slugs.

Editorial correction: The Python SDK defect discussion has been removed from #653 and this branch. Its proposed fixes are tracked in generaltranslation/gt-python#36 and generaltranslation/gt-python#37. The obsolete cache-description commit is dropped; the review index below reflects the rebased branch.

Review each change:

1. Catalog plural completeness — Rails, Nest, and i18n-tasks: distinguish generation from target-language plural completeness.
2. Rails locale adapter — Keep stable GT catalogs outside Rails’ load path; import existing translations and export validated locale-rooted files.
3. Mobility integration — Reject partial results, capture English explicitly, and guard against stale source edits, reviewer edits, and replayed jobs.
4. Weblate comparison — revised argument — Restore GT’s development and release advantages, with Faster / Cheaper to run / Higher quality / Lower risk benefits. Keep the gettext workflow as adoption guidance, accurate competitor capabilities, and the existing POT configuration.
5. Paraglide benchmark — Name the actual comparator, message counts, and transfer-size measurement; fix the framework sentence.
6. Locize timing — Replace the unmeasured seconds claim with documented prerequisites and delivery conditions.
7. Intlayer fill defaults — Distinguish output configuration from actually running translation with a configured provider.
8. Symfony migration — Include locale arguments and require message/domain/review parity before removing XLIFF originals.
9. Equivalent Polish examples — Use complete target catalogs on both sides, generic JSX plural source branches, and explicitly labeled rendering evidence.
10. Paraglide coexistence — Show a compatible i18next-plugin catalog path that retains the compiler; keep native variant metadata out of generic JSON translation.
11. In-house build versus buy — Compare complete workflows, correct bundle and plural claims, and acknowledge automation with existing libraries.

Validation: The earlier integration pass passed all content validators, 201 validator tests, a 595-page build, browser checks, and the rendering/integration checks documented in the review record. For the latest audit, KB, internal-link, reference-link, unsafe-HTML, callout, and structure checks pass. Four extracted catalog configurations pass gt 2.26.5 dry runs; these check source discovery, not hosted generation.

Evidence limits: The live GT API rejected the validation request with HTTP 401; no test project was created. Polish rendering uses supplied translations, and the JSX branch-expansion test mocks the AI boundary. Hosted generation quality and preservation of reviewed edits across a subsequent generation run remain unverified. Mobility checks use SQLite sequential edits, not PostgreSQL concurrency; Symfony migration commands were checked against documentation rather than run in a full Symfony app. These limits are not claims of end-to-end publication readiness.

KB editorial audit — October 19: Inherited the five-article defect audit from #653 and removed the remaining failure narratives in the follow-up. The i18n-tasks example now calls the Rails adapter actually provided in this branch. The original eleven review commits remain in history; this audit supersedes their defect-related wording. KB, internal-link, reference-link, and unsafe-HTML validation pass. Engineering defect evidence stays outside published KB prose; the NestJS scanner reproduction is tracked in generaltranslation/gt#2474. This audit does not merge or publish the articles.

Positioning review — October 19: Carried the revised #653 copy into this branch: clearer GT benefits, a scoped competitor benchmark, and factual corrections without internal rebuttals. The separately proposed integration arguments remain here. All content validators pass; fenced code examples are unchanged from the previous head. Some original commit links now describe changes already included in #653.

Weblate review — October 19: Restored the GT recommendation in the summary, benefits, and conclusion. Moved the same-gettext comparison into the coexistence section and removed the fallback illustration and repeated defensive qualifications. The POT configuration is unchanged. All content validators pass; the production build passed before the final quickstart-link correction, which passed the link and MDX checks. No new hosted translation run was performed.

[cursor]

Stale comment

Risk: medium. Not approved: this updates 11 public knowledge-base comparison articles with substantial workflow and claim changes, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}

[cursor]

Stale comment

Risk: medium. Not approved: ten public knowledge-base comparison articles have substantial workflow and claim edits, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}

[cursor]

Stale comment

Risk: medium. Not approved: ten public knowledge-base comparison articles have substantial workflow and claim edits, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}

[cursor]

Stale comment

Risk: medium. Not approved: eight public knowledge-base comparison articles have substantial workflow and claim edits, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}

[cursor]

Stale comment

Risk: medium. Not approved: eight public knowledge-base comparison articles have substantial workflow and claim edits, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}

[cursor]

Risk: medium. Not approved: eight public knowledge-base comparison articles have substantial workflow and claim edits, which is above the low-risk approval threshold. Human review is needed, and no reviewers were assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router}