Risk and alert relevance
Not every detected difference deserves a ping in Slack. This page ties severity, breaking flags, scopes, and alert preferences—the same levers you adjust in the dependency or subscription UI—into a coherent policy model.
Signal ladder
Think of Intello’s output as a ladder:
safe— informational drift (documentation tweaks, additive optional fields) you may batch into digests.risky— likely requires engineering review soon (behavioral ambiguity, tightened validation) but may not immediately break clients.breaking— high probability clients must change (removed operations, required fields added, incompatible types).
Your alert policies should never treat all three identically unless you enjoy alert fatigue.
Alert preferences (per dependency or subscription)
Alert preferences are rules attached to a monitored object. Important controls:
| Control | Effect |
|---|---|
scope | api (whole dependency), domain, or endpoint narrow targeting. |
scope_value | Domain name or HTTP operation identifier when scoped. |
min_severity | Minimum safe / risky / breaking threshold. |
notify_on_breaking | Extra gate for breaking-tagged events. |
channels | Allow-list of channel names (slack, teams, email, pagerduty). Empty → all configured org channels eligible. |
enabled | Soft-disable a rule without deleting it. |
Channel allow-lists solve common routing needs:
- Breaking → PagerDuty and Slack
#integrations - Risky → email digest only
- Safe → in-app feed only (no chat)
API routes:
- Dependencies:
/api/dependencies/:id/alert-preferences - Provider subscriptions:
/api/provider-subscriptions/:id/alert-preferences
Details: Alert preferences.
Billing and broadcast cutoffs
Some plans limit how far back you can read provider-broadcast changes (catalog fan-out). On those plans, older catalog changes may not appear in the Changes UI even though they existed historically—see Plans and usage.
BYO dependency changes are not subject to the same broadcast cutoff model.
Details: Plans and usage.
Operational playbooks
- Paging: route only
breaking+ affected domain in{auth, payments}. - Daily review: Slack thread with
risky+safebatched, no PagerDuty. - Schema blast: wait for grouped cause rows instead of acking each endpoint individually.