[magnus919]

Research could finish gathering and checking sources, then fail while preparing its reports because the renderer included conclusions that had not passed verification.

Reports now leave out unverifiable extra conclusions and say how many were omitted. If an unverifiable conclusion is needed to answer the main question, the API explains why it cannot publish an answer instead of returning a generic failure. Unknown source freshness remains unknown, and requests for a historical answer retain their existing safeguards.

The renderer and publication gate share the same eligibility rules. Completed search and model operations are not repeated. The renderer and its audit have a new version because their output has changed.

Validation: regression tests reproduced both original failures before the fix. The focused publication, construction, rendering, and recovery tests pass (114 tests). Added an HTTP acceptance case covering undated current sources, the error response, and recovery without extra provider calls. Required CI will run the complete isolated suite.

[factory-droid]

Droid finished @magnus919's task —— View job

---

The change aligns report rendering with publication eligibility and provides specific terminal explanations when required evidence checks fail. I found no independently verified issue among the submitted candidates.

0 inline comments posted