Midnight Shift: TripIt Pivot and Operator Hygiene

The TripIt Developer API is gone. After confirming the OAuth 1.0a connection was dead, we abandoned that path and rewrote tripit-exporter in TypeScript to consume TripIt's private iCal feed instead. Using node-ical, we parse the .ics output into structured travel objects. This is a pragmatic pivot: we trade a richer API for a reliable, read-only feed that doesn't require active developer approval.

Operator Backlog Sweep

The stuck-lane-sweep surfaced several distinct hygiene issues that needed immediate attention.

• Plaud-enrich dead-letters: Five rows in dead_letter state failed with all_selected_recordings_failed. These are not obsolete; they are recent exhaustions. We deferred revisiting them for seven days to watch for a poison-batch pattern rather than clearing them prematurely.

• KG Facts Drift: A storm of 322,576 mirror failures flooded the alerts channel. The root cause was a status vocabulary mismatch where archived was passed through without mapping to the allowed expired state. We are implementing a mapper fix to remap archived to expired and resolving the backlog.

• Dual-Delivery Gaps: Matrix only receives a thin subset of passive traffic. The gap is in the enqueue path: postHumanReadable posts to Discord but never calls the Matrix mirror. We are wiring a fail-soft mirror into the human-readable post path.

Noise Reduction

We are also tightening the noise floor in our operator channels.

In #signal-review, graph-curation cards were dumping policy blobs with zero evidence. We are rewriting the card builder to lead with a one-line claim and evidence snippets, and we are gating posts to skip candidates with zero evidence items entirely. Similarly, #runs was spamming worktree conflict embeds. We are adding a hygiene filter to ignore CHANGES.md overlaps and implementing a rate limit so the same conflict set doesn't reappear every few minutes.

Review Card UX

Finally, #review was exposing raw MCP tool invocations as call-to-actions. We are migrating these to a button-card pattern where operators can Approve or Reject via Discord buttons, removing the need to parse agent syntax from the channel body.