Alerts
Last updated on July 22, 2026
Alerts watch every Run you upload and tell you when a line degrades, without you defining a single threshold first. Detection is on from the moment your first runs land, and you tune it afterwards only if you want to.
Alert types
The six things TofuPilot watches, and what each one means.
Automatic detection
How severity is computed from each procedure's own baseline.
Alert rules
Scope, custom thresholds, severity, and email routing.
Alert lifecycle
Episodes, escalation, auto-resolve, and manual resolve.
Where alerts live
The Alerts page sits under Monitor in the sidebar.
| View | Path | Shows |
|---|---|---|
| Alerts | /{org}/~/alerts | Every alert, grouped into Active and Resolved. |
| Alerts (scoped) | /{org}/{procedure}/alerts | The same list, limited to one Procedure. |
| Alert Rules | /{org}/~/alerts/configure | Your rules, plus recently deleted ones. |
| Alert detail | /{org}/~/alerts/{id} | Summary, what fired it, chart, and activity. |
The Alerts sidebar item carries a badge with the number of currently active alerts, colored by the most severe one open: red for critical, orange for warning, yellow for info. It counts what your rules make visible, refreshes about once a minute, and shows 99+ beyond a hundred.
The badge is a shared view of the organization, not a personal queue. Everyone with alert access sees the same count, and it drops when an alert is resolved — not when you read it.
Trends and events
Every alert type belongs to one of two families, and the family determines how severity and resolution behave.
| Family | Severity comes from | Resolution |
|---|---|---|
| Trends | The magnitude of the metric, recomputed every minute. | Automatic, once the metric recovers. |
| Events | Fixed at fire time. | Stays active until you resolve it (reference samples excepted). |
Yield drop and measurement drift are trends. Golden sample failed, failing sample passed, unit retest threshold, and run failed are events.
The default rule
A rule named Default Alert Rule already exists in every organization. It monitors every standard alert type with automatic detection, across all procedures, stations, and parts, and it follows the catalog as new types ship.
- Alerts page: on. Every severity this rule fires appears there.
- Email: off. You opt in explicitly, so a new organization is never flooded.
- Recipients, once you turn email on: the organization owners.
You can change its severity floor, its scope, and its email setting. Its name, type list, and thresholds are fixed, and it cannot be deleted. Turning it off is the organization-wide off switch for automatic monitoring.
Run failed is opt-in and is deliberately not in the default rule. Add it with a custom rule only on lines where a single failure is genuinely an exception. See Alert types.
Who can see alerts
Alert access follows your role.
| Role | Alerts | Rules |
|---|---|---|
| Owner, Admin, Member | View and resolve | Create, edit, delete, restore |
| Viewer | View only | View only |
| Operator | No access. The Alerts item is hidden from the sidebar. | No access |
| Station (API key) | No access | No access |
What to do first
- Let runs accumulate. Automatic detection needs history before it grades anything, and it says so rather than guessing. See Warm-up.
- Open Alert Rules and set Email on the default rule to the severity you actually want interrupting you.
- Add custom rules only where the automatic behavior is wrong for a specific line.
How is this guide?
MCPPreview
Connect Claude, Cursor, and any MCP client to your TofuPilot test data with OAuth. Read, and optionally write, runs, units, parts, batches, procedures, and stations from your AI agent.
Alert Types
Reference for every TofuPilot alert type: yield drop, measurement drift, golden and failing sample checks, unit retest threshold, and run failed.