Evidence traceability in AI-generated follow-ups

Users need to know what source message, capture, or thread led to the suggestion, who appears responsible, what action is expected, and why it is being surfaced now.

Vortyx uses review-first workflows to keep AI-generated suggestions inspectable before they become tracked work.

• Source message, email, thread, Slack message, voice capture, or text capture.
• The detected owner.
• The expected action.
• A timing cue, if one exists.
• The confidence or review requirement.
• The reason for resurfacing.
• The current state, such as open, waiting, blocked, needs setup, overdue, or handled.

This does not require exposing every internal model detail. It requires enough evidence for the user to decide whether the suggestion is useful.

Source evidence answers the question: where did this suggestion come from?

• Email thread: Client onboarding checklist.
• Slack thread in #launch.
• Voice capture from today.

The user should be able to connect the suggestion to the original context. This is especially important for follow-ups and waiting items because the suggestion may appear days after the original conversation.

Ownership answers who appears to owe the next action. Examples include: you owe sending the revised deck, waiting on Finance approval, or owner unclear and review before tracking.

Timing answers when this matters. Timing can come from explicit dates, relative phrases, recurring language, or conversation state. Vortyx should preserve timing uncertainty when the cue is vague.

Traceability also includes why the item is visible now. It may be overdue, waiting on someone else, blocked, recurring and due again, ready for a draft, or reaching a scheduled follow-up window.

This helps users decide whether to act now, snooze, dismiss, or mark handled.

Evidence traceability matters because detection is imperfect. Vortyx should allow users to confirm a suggestion, edit the title, owner, due date, or wording, dismiss a false positive, snooze a valid item, mark it handled, and improve or discard a draft.

This keeps AI suggestions useful without making them autonomous.

Vortyx should avoid opaque suggestions such as 'Follow up soon' without source, owner, or timing context. It should also avoid presenting uncertain detections as final truth.

A better pattern is: possible follow-up detected from this email, review before tracking. Another safe pattern is: ownership is unclear, confirm who should handle this.

Vortyx is currently evaluating these signals with a closed set of users. Quantitative metrics will be published only after there is sufficient repeatable evidence, including a labeled evaluation set, annotation rules, sample size, precision, recall, false-positive rate, and model/version context.