Project Rhystic Study

An internal booking system scheduled client appointments without ever asking whether anyone was free to take them. It accepted the booking, wrote it to a calendar, and stopped there — no capacity check, no warning, no second look.

So the first person to learn that a day had been overbooked was whoever opened their calendar and found it. There was no earlier signal, because none had been built. Leadership had no view of the problem at all until an overloaded schedule had already become somebody's emergency.

A monitoring layer built on top of the existing booking system rather than a replacement for it. Every thirty minutes during active scheduling hours, it polls calendars and compares what has been booked against real staffing availability.

It flags genuine capacity breaches only — never near-misses. Each one goes to an agent running on Claude Sonnet 5, which turns the raw breach into a concrete scheduling proposal and delivers it to leadership in a recurring digest. A separate notification flow fires the moment a new booking lands, telling the person whose calendar it landed on. Both paths feed one model-driven app, so everyone reads the same live view of scheduling health.

The person being booked went from no signal at all to knowing the moment it happened. Leadership went from discovering conflicts once they were already fires to reading a steady digest that arrives with a proposed resolution attached.

Both of those changes are the build. Flagging real breaches and never near-misses is precisely why the alert stays worth opening — a monitor that cried wolf would have been muted in a week, and that was a thresholding decision made in the automation, not a message sent afterward. Shipping a proposal instead of a bare alert is why leadership acts on it rather than adding it to a triage pile; the agent exists to close the distance between knowing and deciding.

And because it was built as a layer over the booking system instead of a replacement for it, nobody had to change how they book anything. There was no migration, no retraining, no new tool to adopt. The rollout cost almost nothing because the architecture had already absorbed it.

Capacity breaches now get caught and routed to leadership with a proposed fix, instead of surfacing only after someone notices an overbooked calendar.

The scheduling picture went from something reconstructed after the fact to something visible in real time, to everyone who needs it, while there is still room to act on it.