Stability

A journey that passes once has not been shown to pass reliably.

Reports record, per journey:

• Each attempt and its status
• The criteria passed out of criteria total for that attempt
• The aggregate journey score
• The configured pass threshold
• Which criteria were unstable, and on which attempts they passed and failed

An unstable criterion is more actionable than a failed one: it tells you the behaviour exists in the agent but is not reliably reached. The published card-block acceptance report passes its threshold while carrying exactly one unstable criterion: on two attempts the agent blocked the selected card, and on the third it asked for a rigid confirmation and never acted.

A criterion that fails on every attempt is a different signal — stable, and simply broken. The trajectory report shows one of those: the same confirmation-wording defect on all three attempts.