Where are Bean and Toffee?
Every answer here comes from a model trained on a few hundred real sightings, logged by hand over several months. It's a guess with numbers attached — not a camera, not a tracker.
How this works
Each sighting was logged with a timestamp and a spot — which side of the bed, whose desk, whose lap. The answer splits on one thing only: day or night. Within each half it's the frequency of what's actually been logged, so 10am and 4pm give the same answer, and so do 9pm and 3am.
That's deliberate, and it replaced a random forest that read the clock more finely. Tested properly — trained on the past, scored on what happened next — knowing the exact hour of the afternoon made the predictions worse, not better. Where these two settle during the day turns out not to be a function of the clock, or at least not one a few hundred sightings can demonstrate. Night versus day is a real difference. Two o'clock versus five isn't.
Recent sightings count for more than old ones, on a ninety-day half-life. That matters more than it sounds: a climbing wall went up in the office in July and promptly became the most popular spot in the house, so sightings from the spring describe a home that no longer quite exists. Ninety days is a deliberately gentle setting — it keeps the answers steady rather than letting a handful of new sightings rewrite the page, at the cost of taking a few weeks to notice when the cats genuinely change their minds.
The ~20 specific spots are grouped into rooms so the map stays readable, with the most likely exact spot shown as the smaller tag. Percentages are the model's confidence.
Not a prediction — the log. All sightings either cat has ever had recorded, on the same floor plan. Cold blue is a spot she has never been found in; red is where she actually sleeps. Both maps share one scale, so Bean and Toffee can be compared directly.