Changelog

The public surfaces arrived: documentation written page by page against the real product, and a blog for the build story.

• Docs cover every page as it actually ships, named as the sidebar names them.
• The blog runs on the same site, ready for the series.

When new content contradicts your map, both sides now stay live and the disagreement itself comes back with your search results.

• Search responses carry a conflicts list: both atoms, both dates, both directions of the link. Nothing is silently merged and nothing is deleted.
• Computed entirely in the database, zero AI tokens, nothing decrypted on the way.
• Same day: re-forwarded emails and re-uploaded documents are now fingerprinted and skipped before they burn any processing. The same content never costs you twice.
• The extraction queue also got adaptive: when the backlog grows, the pipeline speeds itself up, and fails open if the depth check ever errors.

Every fact on your map now knows when it became true, when it stopped, and when it was last re-confirmed.

• When a newer fact supersedes an older one, the old fact closes its validity window and stays on the record with its dates. Nothing is overwritten.
• Research nobody has re-checked in ninety days is flagged stale instead of served as current.
• Unknown dates stay honestly unknown; nothing is fabricated to fill a column.

Tier limits are now enforced at the database layer, not by application politeness.

• Metering rules live beside the data they meter, checked on every write.
• Failures are loud and attributable, never silent losses.