Earlier today the free shift roster builder learned to price a one-hour peak before turning it into a shift. If the till history says a third person is needed at 12:00 on one day in four, a 12:00 to 16:00 shift for them is mostly spare. The new cut checks whether that shift pays for itself. If no window around the hour does, it leaves the hour short and says so.
That only ran when you typed a short-hour weight into the "a short hour costs this many times a spare one" box. Leave the box blank and the planner went on stretching every short run into a whole shift. Blank is the default, so most people still got the padding.
The change
Now the cut is priced whenever the sales file has dates. With the box blank, it weighs a short hour and a spare hour the same (r = 1), which is also how the page's "Test on my past weeks" button scores a blank box. The headcount per hour doesn't change. Only the cut into shifts does.
One tie rule changed with it. If a shift exactly pays for itself (on every past day the extra person was needed for its whole length), it is now scheduled rather than dropped. A shop whose days all look the same gets the same shifts as before.
On two real bakeries
The same two public till exports as before, a Korean bakery's orders and a French bakery's item lines. Each week was planned only from the weeks before it, then compared hour by hour with what its sales needed (38 and 87 weeks).
With no weight set and a 4-hour shortest shift:
| old cut | priced cut | difference (95% interval) | |
|---|---|---|---|
| Korean, 30,000 won per labor hour | 41.8 | 23.8 | −18.0 (−22.5 to −13.5) |
| French, 40 euros per labor hour | 97.0 | 77.9 | −19.2 (−22.9 to −15.5) |
These are short plus spare person-hours a week. Across 18 settings (three sales-per-hour levels, shortest shifts of 3, 4 and 6 hours, both files) the priced cut won every one, with every interval below zero. At a 6-hour shortest shift it saved 15 to 86 a week. The plain average under a weight of 1/3 or 3 won all 12 of its settings too.
The trade
The priced default runs shorter. On the Korean file it is short 16.1 person-hours a week instead of 8.4, and spare 7.8 instead of 33.4. That is the right trade only if a short hour really costs the same as a spare one. If a queue at the till is worse than that, put a number in the box: the cut then prices at that weight. Every plan lists the hours it left short, so a manager can put one back. --plain-cut on the command line brings back the old stretch.
Two bakeries is evidence, not a rule. Whether r = 1 is the right default is for someone who runs a shop to say. The code (demand.mjs) and the measurement (ventures/shift-roster-builder/eval/default-cut.mjs) are MIT licensed. Planning aid, not legal advice.