Jev field notes: checking the path from saved post to evaluation

What this record covers

These notes summarize ReDeck’s recorded engineering audit from 27 September 2026. They concern evaluation delivery after a History recovery fix, not model accuracy or a new benchmark run. Only aggregate counts are reported; account identifiers and private post text are omitted.

The initial gap

The audit recorded 69 Timeline History entries saved since activation, but only 7 evaluation events and 7 completed TypeSafe requests. The other 62 entries contained text. The precise lost-notification step was not established.

The fix stored an inbox entry alongside each newly saved History post and reconciled persisted work, rather than depending only on transient notifications.

The observed result

After reloading the updated extension and ReDeck, the recorded live check reached 105 post-activation History entries. All 105 had an evaluation event, with no missing events and an empty inbox at that checkpoint.

The final recorded statuses were 91 below-threshold, 5 errors and 9 pending actions. These counts sum to 105. They show that every saved entry had a recorded outcome at that checkpoint; they do not show that 105 X actions succeeded.

What we cannot conclude

The audit did not label every post as a true or false match. It cannot establish precision, recall, a spam-detection success rate or performance for every account. Pending actions and errors still need separate inspection.

Source: the maintainer’s Jev History recovery audit dated 27 September 2026. The record was reviewed for this article; the live experiment was not rerun on the publication date.

A useful check for your own workflow

Preview a narrow rule, then review a sample of saved posts alongside their evaluation and action statuses. Separately record false matches, missed matches, request errors and action failures.

A model score is not a completed action. This distinction matters for bookmarks, list membership and account actions alike.