November 17, 2026: Rules on alerts, ratings and change history
This update includes rules on the alert event type, alert ratings in the alert history, the metric chart in Slack notifications and the change history on configuration screens.
Rules on alerts
The _alert event type now accepts a default rule, run on every alert event with the actions of event rules. The new segment_<field> fields separate each monitored dimension, and the Monitor data enrichment publishes the monitor’s name. In accounts with Cases enabled by Glass Data, the Open case action turns the alert into a case.
For example, the condition $event == "opened" and $segment_merchant_id != null, with the Add to list action, records the merchants with an open alert. See Rules on alerts.
Ratings in the alert history
Each rating of an alert, Helpful or Not helpful, now creates a Rated event in _alert, with the reason and the category, without who rated or the comment. Ratings can be measured by metrics and queried in Explore, to find the monitors recipients consider helpful. The current total is the sum of the votes field.
For example, a metric on _alert that sums votes, with the Monitor and Rating dimensions, shows the net ratings of each monitor. See Ratings in the history.
Metric chart on Slack
The metric chart now accompanies alert notifications on Slack, added right after the text and updated when the alert is normalized. A workspace authorized before this update keeps receiving only the text until it is authorized again, under Reconnect.
See Reconnecting the integration.
Change history
With the new change history, the pages of event types, fields, metrics and monitors show who changed the resource, when, and what changed. For API keys, the history is on the key’s edit page, restricted to the Administration profile. The history includes changes made before this update.
For example, a metric’s history shows who removed an aggregation. See Change history.
Cell activity
On the active version or the version in observation (shadow) of an analysis, the cells table now shows each cell’s activity over the last 30 days: triggers, decisions, evaluations with no data and records. The cell’s name opens the new cell page, with its settings, its activity and its history across versions.
See Cell activity.
Other improvements
- Lists: an item’s description now accepts up to 255 characters, and a list’s description up to 1,000. The new limit also applies to
POST /lists/{list_id}/items, as a compatible change. See Lists. - Rules: a condition’s expression is now validated as soon as the cursor leaves the box, and errors appear below the expression before saving. See Authoring errors.
- Agents: the
add_list_itemsandremove_list_itemstools add and remove up to 1,000 values from a list per call, with the same validation as the panel.describe_listnow reports, row by row, whether the rule queries the list or adds items to it. Ask, for example: “In the loja-br account, add these 20 CPFs to theclientes-quarentenalist, expiring in 7 days.” See Tools. - Agents: tool responses now contain text only, with all the data, without the
structuredContentblock. A client that processes responses in code must read the text. See Operations. - Dashboards: adding, changing and removing panels, saving the settings and changing the time range now apply without reloading the page, keeping the chosen time range and the layout being edited. See Organize the layout.
- Personal settings: users now upload their own profile photo, in PNG, JPEG or WEBP. Without a photo, the avatar shows the name’s initials. The platform no longer queries Gravatar, an external avatar service tied to the email address.
Fixes
- Rules: a condition with
inornot inon a numeric field and a collection of numbers, such as$amount in [10, 20.5], resulted in an internal error. The comparison is now made by value, and100matches100.0. See Collections. - Metrics: when writing events fell behind, a 1-minute interval could be measured before receiving all its events and show a lower or empty value. Measurement now waits for the minute’s events to be written, and a 1-minute point may arrive a few minutes later.
- Monitors: after a read error, a monitor could skip part of a window’s segments and not open the alert, or delay by up to 1 hour the other monitors on the same query. A metric measurement rejected for excess queries could also be left without a value. All three problems are fixed.