Pomodoro Timer

**The system clock changing.** The version that shipped before September 2026 counted down against `Date.now()`, the wall clock, and the wall clock jumps: daylight saving moves it an hour twice a year and the time daemon steps it whenever the machine has drifted. Measured on 2026-09-04, a running 25 minute session given a one hour clock step did not just show a wrong number, it ENDED: the display went from 24:59 to 05:00 and the page moved straight to a short break, crediting a pomodoro nobody had worked. This version reads `performance.now()`, which is monotonic and is specified never to jump, and the same measurement leaves the session exactly where it was.

**A reload.** Nothing was written down. A session two and a half minutes into a one minute countdown went back to 01:00 on a reload, and the completed pomodoros, the minutes worked and both streaks were component state that vanished with it. Here the running session and today figures are written to this browser, picked up with the elapsed time taken off, and a saved value of the wrong shape is refused rather than breaking the page.

**A link somebody typed.** Every setting was read with `parseInt(x) || default`, which swallows a zero, truncates a decimal and accepts anything else. `?workDuration=0` opened at 25 minutes, `?workDuration=8.5` opened at 8, `?sessionsBeforeLongBreak=0` made every break a long one because a counter that must reach zero already has, and `?workDuration=-5` and `?workDuration=99999` rendered clocks reading "-5:00" and "99999:00", a session whose deadline is already past and one lasting sixty nine days. Every setting is now range checked and anything outside its range is clamped and named on the page.

**A background tab.** A browser throttles a background tab to about one timer callback a second, and to one a minute after five minutes, and no page can do anything about that. What a page can do is compute the remaining time from the clock rather than counting it down, so the reading is correct the moment it runs again, and say how late it noticed. Starved of every timer callback for four seconds, this timer reads the right remaining time to within half a second.

**And the cycle itself was already right,** which is worth recording rather than quietly claiming: with two sessions before a long break, stepping through gives Focus, Short Break, Focus, Long Break, and it does so both in the module and when the page is driven. A first measurement said it never reached a long break, and that was the test reading the three mode BUTTONS, all three of which carry a session name, rather than the selected one.

One thing this page cannot tell you, and says so: the 25 and 5 minute figures are conventions of Francesco Cirillo technique rather than a measured optimum. There is no document to check them against, so they are labelled as conventions where they appear.