Tune detection thresholds
Summary
Configure → Detection controls how the platform interprets what your sensors report: when it calls something stale, anomalous, risky, or worth your attention. Nothing on these pages changes what is collected; that is decided under Capture & devices and under Organization.
If the product feels too noisy or too quiet, this is the area that governs it.
Before you start
You'll need the Manage Users or Manage Tenant permission to open Configure; Chain detection and KPI thresholds need Manage Tenant, and Chain detection is part of the Ultimate plan. Risk scoring's write actions require the Professional plan or above: on Starter, the model's defaults apply and the controls are read-only.
Change one setting at a time. Everything here alters how existing data is interpreted, so changing several at once makes the effect impossible to attribute.
Adjust monitoring thresholds
Go to Configure → Detection → Monitoring thresholds. Five settings, each with a permitted range:
| Setting | Range | Governs |
| User coverage staleness (Hours) | 1-168 | When a user's coverage is treated as stale |
| Sensor stale threshold (Minutes) | 5-1440 | How long a quiet sensor waits before being flagged as stale |
| Min baseline events | 1-100 | How much activity is needed before a baseline is trusted |
| Browser DLP volume threshold (KB) | 1-10240 | The data volume at which DLP volume detection triggers |
If sensors are being marked stale too aggressively, raise the Sensor stale threshold before investigating devices. A device that checks in less often than this threshold is flagged even though it is healthy.
If anomaly detection is producing nothing, a baseline may not exist yet. Baseline window and Min baseline events must both be satisfied before there is anything to compare against: a long window with a high minimum takes real time and activity to reach, and the product doesn't announce when it gets there. This is the usual explanation in the first weeks of a deployment.
If DLP volume alerts are too frequent, raise the DLP volume threshold (KB). This governs volume detection only, not content-based DLP.
Tune paste monitoring
Go to Configure → Detection → Paste monitoring. Paste detection looks for paste events in a window either side of an AI interaction:
- Paste window before AI interaction (minutes), 1-30: how long before an AI request to look for paste-into events.
- Paste window after AI interaction (minutes), 1-30: how long after an AI request to look for paste-from events on destination sites.
Monitored destination domains lists the sites where paste-from events are watched, one per line.
If pastes are being missed, widen the windows: a paste outside the window isn't correlated to the AI interaction. If a specific site is missing, confirm it's in the monitored domains list; only listed domains are watched.
Reset to defaults replaces your list entirely, so copy anything custom first.
Set exposure-chain auto-promotion
Go to Configure → Detection → Chain detection. A new exposure chain doesn't go to your Inbox straight away.
Under Auto-promote to inbox, set how long a chain may wait before becoming an Inbox action, per severity: Critical, High, Medium, Low. Then set Auto-close after inactivity (hours) and the Default suppression window (hours).
Blank means never. Leaving a severity empty is a setting, not an absence of one. By default only critical chains auto-promote, and immediately; high, medium and low are blank.
Watch the two different ranges on this page: promote timers accept 0-168 hours; auto-close and suppression accept 1-720.
If you see "auto_promote_hours.{severity} must be between 0 and 168 (or blank for never)", the value is out of range: enter 0-168, or clear the field to disable promotion at that severity.
If your Inbox is filling up, check these timers before suspecting detection: a full Inbox usually means a non-critical severity was given a timer. If nothing ever reaches the Inbox, the severity you're waiting on is blank.
Use do-not-disturb
Do-not-disturb (pause auto-promote) stops high, medium and low severity chains promoting.
Critical chains still promote. Do-not-disturb does not apply to them. If you're planning a maintenance window around it, expect critical items to keep arriving.
Change KPI alert bands
Go to Configure → Detection → KPI thresholds to override the alert bands on the Overview and Dashboard KPI cards:
- Coverage card → yellow below and → red below, as percentages (0-100)
- DLP card → red at this many critical violations (0-10000)
- Unapproved-platforms card → yellow at this many (0-10000)
These change when a card turns color, not the figure on it. If a number looks wrong, this isn't the page that fixes it.
If you see "Couldn't load the KPI thresholds", reload before changing anything, so you don't save over values you never saw. The same applies to "Couldn't load the chain detection settings".
Adjust risk scoring
Go to Configure → Detection → Risk scoring to tune the weighted dimensions, scenario scores and tier thresholds behind each agent's risk score.
This re-scores existing agents, so tiers can shift across Users, Platforms and Data flows at once. Change one weight, save, and review the effect before changing another. Note the previous values before you start: the product keeps no history of what they were.
If the controls are read-only, your plan doesn't include risk-model configuration (it needs Professional or above) or your role can view but not change it. On Starter, the default model applies.
Contacting support
Say which setting you changed, what it was before, and what you expected. Because these settings change interpretation rather than collection, the previous value is usually the fastest route to explaining a change in behavior. If data is missing rather than misinterpreted, this is the wrong area: see An endpoint sensor isn't reporting.