Severity & thresholds

Every finding a Drone emits carries a severity. Severity is how SwarmADE separates “this must be fixed before the task lands” from “good to know, but not a blocker”. Your Hive then applies thresholds to turn those severities into a decision: block the task, or let it land with the finding recorded as advisory.

The severity levels

Drones grade each finding on a four-level scale, from least to most serious:

• info — informational. Context or a minor suggestion; never a blocker by default.
• warning — worth attention, but the task can usually still land.
• error — a real problem that should be fixed before landing.
• critical — a serious issue, such as a leaked secret or a destructive migration, that almost always blocks.

How thresholds turn severity into a decision

A Hive defines a threshold that marks where blocking begins. Findings at or above the threshold block the task from landing; findings below it are advisory and recorded for your review without holding the task up.

For example, with a threshold of error, a critical migration finding blocks the task, while a warning style nit lands alongside an advisory note. Thresholds are set per Hive, so a security-sensitive Hive can block on warning while a docs Hive only blocks on critical.

What happens when a task fails review

When a Drone produces a blocking finding, the task fails review and returns to the Queen. The Queen does not simply discard the work — it routes the task back into the loop with the Drone’s feedback attached, so a Worker can fix the specific problems and resubmit. If a task keeps failing, the Queen escalates rather than retrying forever.

This is the heart of the review loop: Drones grade, thresholds decide, and the Queen turns a failed review into a concrete next action.

Where to go next

• Replanning + retries — how the Queen handles failed reviews and escalation.