Decisions queue

Whenever a category rule wants to take an action you haven’t yet trusted, InboxMate holds it instead of firing silently. It shows up in Entscheidungen — one global queue for every mailbox, so nothing gets buried in a single conversation. This is where supervised automation actually happens: the AI proposes, you decide. The Entscheidungen page has two tabs: Warteschlange — the live queue this page describes — and KI-Reife, the overview of how much autonomy each rule has earned. See KI-Reife for the second one.

​

What lands here

Any action a matched category wants to run that isn’t yet trusted enough to auto-fire: forwarding an email, sending or auto-replying, filing a ticket, archiving, marking read, moving to a folder, or a hard delete. Each is a suggestion with a preview — a forward shows its recipient and note, a ticket shows its subject/priority/assignee, a reply shows the drafted text. Actions from the same inbound email are grouped into one chain card — the email’s subject, sender, category, and confidence appear once, followed by one row per pending action in execution order.

​

Deciding on an action

Each row offers three choices: For a chain with several steps, use Alle bestätigen to approve the whole chain at once, or Einzeln to expand it and decide row by row. Steps that depend on an earlier one show a locked chip instead of buttons — läuft danach automatisch if that step is set to auto, or danach freigeben if it’s a further decision you’ll make once the step above is approved.

​

Edit before you approve

For forwards, sent emails, and auto-replies, the row shows a snippet of the drafted text with Nachricht ansehen & bearbeiten. That opens a split view — the sender’s original message on one side, your editable draft on the other — so you can rewrite the note or reply body without losing context. Confirming there (Freigeben & senden) approves with your edited text; it’s the same approval as clicking Ja on the row.

​

Resolved history

Below the open queue, Erledigte Entscheidungen lists what you’ve already approved or ignored, each with Details ansehen for the full record: why the AI classified it that way, what happened, and who approved or ignored it and when. Nothing here is a silent auto-fire — only decisions an operator actually made show up, so the history reads as a decision log, not a raw activity dump.

​

How this feeds learning

An Ignorieren with a reason, or an Anders correction (a different recipient, assignee, or priority than the AI proposed), is remembered — this is the light learning loop covered in Learning from corrections. It doesn’t retrain a model; it keeps a short memory of recent corrections that gets fed back into future classifications for that categorizer.

​

How this connects to automation

Every Ja (and every Anders) on the same rule counts toward that rule’s trust. Approve it enough times and InboxMate offers to promote it to run automatically, no more suggestions. You can watch that progress for every rule — and promote deliberately — on the KI-Reife tab; the full graduation path, including which actions can graduate and which never can, is covered in KI-Reife.