Day shift starts / ends — when your typical daytime shift runs (e.g. 5am–6pm). This is the window used as disruptive on entries checked "Night shift" — since during a run of nights, getting pulled into your normal daytime hours is what disrupts you. It is also the workday the Predicted line assumes for Monday to Friday until your log has enough history.
Night shift starts / ends — when your typical night shift runs — the hours in between your day shift (e.g. 6pm–6am). This is the window used as disruptive on everyday entries (not checked "Night shift"), and on scheduled primary or backup call days, hours worked from this time on are counted as Call automatically. On Saturdays and Sundays the Call box is also ticked for you when you go to check in, at any hour; switch it to Hospital or Clinic if you aren't on call. On other days that aren't on your call schedule, nothing is counted as Call unless you tick Call yourself.
Call day starts at — what hour your on-call shifts typically begin. The shaded Primary/Backup call boxes on the graph run a full 24 hours from this time, rather than midnight to midnight, so they actually line up with your real call day.
How the score works — Bandwidth used is one running score from 0 to 100%. Every hour on duty adds points, and it only comes back down while you're off duty. Whatever you haven't recovered carries into your next shift.
Points / hr worked — points added for every hour on duty, whenever it falls.
Extra pts / disruptive hr — added on top for each hour on duty inside whichever window applies to that entry (for everyday entries, from your night-shift start to your night-shift end).
Primary / backup call hrs — assumed hours worked on a scheduled call night, starting at your night-shift start time. Used for the Predicted line only, and only until you have logged 2 call nights of that kind.
How Predicted is built — from your own usual hours over the last 8 weeks. For each day of the week, and for primary and backup call days, it assumes you work the hours you worked on at least half of those days. Vacation days are left empty. A kind of day without enough history (3 logged weekdays, or 2 call nights) uses the Day shift and call-hours settings instead. The info button in the Predicted Call Schedule section shows what it is using right now.
Recovery pts / hr off — how fast the score falls while you're off duty. Recovery delay — optional: hours after checking out before recovery starts (getting home and to sleep). At 0, every hour off counts. The app doesn't know how much you actually slept — it only sees the gap.
Long-shift limit / Extra pts / hr past limit — once one continuous stretch runs past the limit, every further hour adds this many extra points on top of its normal rate. Back-to-back entries count as one stretch.
Graph smoothing — the graph is an estimate, so its lines are drawn as an exponential moving average of the score over this many hours instead of a sharp sawtooth. Higher is smoother, but the line lags further behind: at 6 hours the peak shows up several hours after a long shift ends and lower than the exact score. Set it to 0 to draw the exact score. The "Fucks Remaining" number always uses the exact score, not the smoothed line.
Why late hours of a long shift cost the most — two separate things stack. Overnight hours cost more because of when they happen (your body doesn't like being up at 3am regardless of duration). Hours past the long-shift limit cost more because of how long you've been going. Hour 20 of a 24-hour call is paying for both.
Time of day — the "Fucks Remaining" number also shifts a little with the clock, not just your logged hours: it reads a bit worse in the early morning (~5am) and a bit better in the evening (~8pm), since that's when most people's alertness is naturally lowest and highest regardless of how much they've worked. The graph itself doesn't include this — it's only applied to the live number.