Case study 06 — Anlytic
Automations that run with nobody watching
Scheduled actions driven by what a chart shows, without a browser open and without repeating themselves.
Scheduled
by interval or cron
Server-side
evaluation
Quiet
when nothing changed
The problem
People wanted their dashboards to act, not only display: tell me when this crosses a line, add a row when this returns something. That means evaluating on a schedule, with no session in existence, seeing exactly what the person saw when they set it up.
How it fits together
What I did
Rebuild the view
At run time the saved configuration is reconstructed server-side, so a scheduled run queries what the user actually configured rather than a default.
Evaluate, then decide
Conditions cover presence, absence and thresholds, combined with any or all matching.
Stay quiet when nothing changed
Results are compared against what was already acted on, so a schedule that ticks every minute does not send the same alert every minute.
The hard part
Reconstructing someone's saved view on the server, with no browser involved, so a scheduled run sees precisely what they saw. Everything else in the feature depends on that being faithful.
This suppresses repeats. It is not exactly-once delivery and is not described as such.
Stack
Written at the level the reasoning survives. Internal specifics are left out on purpose.