Three reliability fixes, all of them cases where Alfred had the answer and lost it on the way to you.

## Telescope columns no longer fail on work that succeeded

A Telescope cell could read "We could not extract this column" even when the extraction had gone perfectly and every claim, quote and page reference was ready. The answer was there; the step that hands it over rejected it and marked the cell failed. Re-running it produced the same result, which made the column look genuinely impossible rather than merely stuck.

Those columns now come through with their claims intact. Nothing about how strictly we check the evidence has loosened: a column that truly has nothing to report still says so honestly, and a column we cannot verify still refuses to invent one. If you have an analysis with a failed column, re-run it and it should fill in.

## Offline drafts recover on their own

If your browser closed its connection to the offline store partway through a session, which can happen when a second tab opens or the browser reclaims space in the background, the workspace kept accepting your edits without actually saving them offline. The pending-changes indicator went on saying work was queued while nothing was being written.

The workspace now notices and reconnects on its own. Your writing keeps being saved, and what you had already written was never at risk.

## Slow sources fail instead of hanging

When Zotero, Mendeley or a paper host went unresponsive, a request could sit waiting long enough to time the whole page out, giving you an error with nothing in it. Those calls now give up on their own and tell you which source did not answer, so you can retry that one thing rather than the whole action.