Changelog

Streaming evaluation. Conditions are checked as events land rather than on a schedule. Nightly evaluation remains available for signals that depend on billing state, which still settles daily.

Signal backtesting. Before activating a signal, replay it against the last twelve months of history to see how often it would have fired and against which accounts.

Firing explanations. Every firing now links to the specific events, thresholds and time windows that produced it. Available in the UI and on the API as explanation.

Signal editor now shows the compiled SQL for any condition built in the visual builder, so your data team can review rather than rebuild.

Throttle configuration moved to the signal level, with per-channel overrides. The old workspace-wide throttle is migrated automatically.

The signals.evaluated_at field is deprecated in favour of signals.fired_at. The old field continues to return values until 1 February 2027.

Reverse ETL destinations. Write account_scores, segment_members and signal_events back to a schema you nominate, with full-refresh or incremental modes.

dbt source generation. Northwind emits a sources.yml for the tables it writes, so downstream models pick them up without hand-editing.

Warehouse sync now supports key-pair authentication for Snowflake in addition to username and password.

Sync failures now surface the underlying warehouse error rather than a generic message. Long overdue.

Fixed a case where a sync that timed out mid-batch could double-count events for accounts whose IDs sorted last.

Composite scores. Define your own weights across four component dimensions, or start from one of three templates for PLG, seat-based and usage-based businesses.

Score history. Every account keeps a daily score series for two years, so you can chart the slope rather than argue about the number.

Cohort comparison. Compare score distributions between two segments to test whether an onboarding change actually moved retention.

account.health_score now returns the composite score. Your previous values are available as account.usage_score with no change in definition.

Configurable session lifetime from 1 hour to 30 days, set per workspace. Idle timeout is now independent of absolute timeout.

Session tokens are now rotated on privilege change, so a role downgrade takes effect immediately rather than at next login.

Tightened the CSP on the application shell and removed the last two inline script exceptions.

Audit log export could omit entries from the final minute of the selected range. Fixed and backfilled for all affected workspaces.

Approval steps. Insert a review gate anywhere in a playbook. Approvers are notified in Slack and can approve, edit or reject inline.

Draft editing. Generated emails can be edited before sending, and edits feed back into the template so the next draft is closer.

Playbook runs now show a full execution trace with timing for each step, which makes debugging a stalled play far less mysterious.

Quiet hours are now timezone-aware per recipient rather than per workspace.

Account search moved to a dedicated index. P95 query time dropped from 1,240ms to 190ms across all workspaces.

Timelines stream in reverse chronological pages rather than loading the full history before first paint.

Reduced the application bundle by 38% by removing a charting dependency we were using for exactly one sparkline.

Timeline entries from Zendesk could appear with the ticket creation time rather than the event time when the two differed across a day boundary.

Keyboard navigation skipped the filter bar on the segments screen for Safari users.

Per-workspace residency. Selected at creation and immutable afterwards, so there is never ambiguity about where a record lives.

Sub-processor notifications by email and RSS, 30 days before any change takes effect.

Data export now includes a machine-readable manifest listing every table, row count and checksum.

HubSpot two-way sync with field-level mapping and conflict rules you configure rather than inherit.

Deal-stage signals. Trigger on usage changes within an open opportunity, which turns out to be the single most requested signal we have shipped.

Salesforce sync no longer retries indefinitely on a permanently invalid field mapping. It now fails loudly after three attempts and tells you which field.