Rolino fixed changes

Blog corrections rebuild source references from the completed text during review, while preserving facts about different products that use similar wording.

Blog comparison checks preserve similar statements about different products instead of removing important differences as repetition.

Blog corrections now reference exact article paragraphs, preventing copied-quotation errors from blocking otherwise valid drafts.

Blog regeneration no longer fails when the writer suggests a source ID as a link. Links are selected later from the verified site catalog.

Blog corrections remove exact repeated sentences while preserving source references and code. Briefs now require valid source references before drafting, and review checks for editor instructions left in article text. Review findings are restricted to verified source IDs.

Blog editing removes repeated explanations across sections and treats article length as guidance, so short source material does not require filler.

Blog link preparation can recover when suggested link text is absent from the article. The recovered link still passes the final review before saving.

Blog drafts use only claim and source IDs from their approved brief. Corrections select exact saved passages and tolerate harmless JSON line-break errors. Verified source quotes and minor review notes no longer cause unnecessary generation failures.

Image retries now show a clear queued status. Image service billing failures keep the article available and show a temporary service notice.

Blog link preparation can propose exact matches to relevant page titles when the model returns none; final review must approve them before saving.

Blog preparation keeps reviewed descriptions and excerpts unchanged during link selection, and keeps hypothetical examples separate from factual claims.

Failed Blog generation no longer shows a publication-schedule warning when no approved schedule exists.

Blog release tests use the same model request settings as the worker and include 18 article comparisons and 20 reviewer trials.

Corrected release checks for supported agent permissions, restarted media uploads, and time-dependent Analytics data.

Blog preparation now inserts selected catalog links into the article body, checks the selected format before writing, and blocks repeated prose after its single correction.

Blog previews now render lists, emphasis, headings, and links with the same Markdown reader as the editor.

Blog draft corrections now change only the affected passages. Link checks run after link selection and avoid unrelated catalog pages. Optional action buttons without a verified destination are omitted. Minor review notes no longer block an otherwise complete draft; major issues still stop generation. Failed quality checks show clear Needs attention messages and review actions.

Fixed Gemini rejecting Blog editing requests. Rolino still checks article size and source requirements before saving.

The Blog pipeline test now limits writer reasoning and allows more time for the draft response, while keeping each request within the worker deadline.

Blog test drafts can repair evidence quotations during editing. Quotation mismatches use the existing single correction before final validation. Final drafts still require exact quotations and valid sources. Failed provider calls retain more diagnostic details.

Blog draft retries and preparation now allow enough time to save the generation context and reserve credits together on slower database connections.

Blog preparation tests now pause after each saved AI stage and reject expired results before saving an article or charging an article credit. Late provider responses retain their cost records.

Idea headings and search results now show the content type once and preserve meaningful text before colons. Saved titles, editing, exports, and published headlines are unchanged.

Agents can check TikTok and LinkedIn publishing options when the response includes the connected account identity.

Hosted agents now check publishing options for the selected account when a brand has multiple connections.

The agent guide now requires exact publishing-account selection and explains hosted MCP updates versus local package updates.

Documented the ChatGPT renewal setup, including the required base permission, to prevent repeated sign-in prompts when a short-lived connection cannot renew.

MCP sign-in challenges now request renewable access. OAuth token exchanges record safe status details to help diagnose repeated connection prompts without logging credentials.

Restored scheduling, publishing, and media upload tools for connected agents on the production MCP server. The limited review profile no longer hides these tools from customer connections; existing permissions and confirmation requirements still apply.

Manual Blog performance keeps query-based article URLs separate from other pages. Publication dates use the Blog site time zone, and new-article path defaults no longer change connected publishing routes.