Start with the day people actually have

ShyftSync reads selected calendars already on the phone or takes a repeating schedule. A model combines those hours with the person’s usual sleep timing, commute, and preparation to estimate a timeline of work, rest, and free time.

The window finder keeps the stretches when both people are predicted to be free. It ranks a small selection, identifies a suggested part of each window, and leaves the choice to the people making the plan. Predicted sleep is excluded before ranking begins. A high score cannot buy its way past that constraint.

Each suggestion carries a short explanation. Choosing a plan adds it to the calendar; simply opening the app never books someone’s time.

One calculation, two implementations

The reference engine was built in Python so its assumptions, intermediate values, and behavior could be inspected. The iPhone uses a native Swift implementation, with the calculation performed on the device.

Saved synthetic cases connect the two. Checks compare the same time boundaries and numerical traces, rather than accepting two charts that happen to look alike. Rounding, daylight-saving transitions, and the difference between elapsed hours and local clock time all mattered in the port.

These checks establish implementation consistency. They do not establish that the model predicts an individual’s sleep accurately. Simulation, software correctness, and real-world validation answer different questions.

Share the answer people need

Partners need to know when they can make a plan. They do not need each other’s raw sleep records or personal explanations. Each phone prepares a limited availability payload and encrypts it for the linked phone. The relay passes sealed messages and cannot decrypt their contents.

Optional Apple Health sleep is processed locally. Raw samples and measured sleep intervals stay on the phone. Some predictions shared with a partner can reflect that local model, so the privacy boundary is described precisely: keeping raw records local is not the same as saying no derived information is shared.

The test suite checks that the limited payload still produces the expected shared windows, without sending the full personal timeline. Separate tests check that one person’s explanation does not appear as the other’s.

Let people stay in control

The interface has a week view, suggested times, and a shared olive tree. The garden recognizes protected rest and time together without taking growth away for working hours. A person can use the app alone, link with a partner, or choose whether to share predicted rest.

The current build runs on iPhones and has entered testing. Calendar import, shared planning, optional Health integration, and encrypted partner updates are implemented. Screenshots on the product page use the app’s synthetic demo schedules.

The next evidence has to come from independent evaluation and testing in everyday use. For now, the useful result is a working path from a model that can be inspected to a native app with explicit constraints and a narrow sharing boundary.