Twenty weeks was not enough: the call-out result that flipped at 200, and a price for the second call-out

A café roster that looked worse over 20 simulated weeks of call-outs was better over 200. A new check prices call-out pairs covered alone but not together.

The Shift Roster Builder has an option, --cover-price, that gives up a little pay or fairness so that more of the week survives one person calling out. Yesterday's run played 20 seeded weeks of call-outs against the café sample. The priced roster lost more hours than the plain one: 155 against 188 at an 8.4% leave rate. The README called it "a trade, not a win". The guess was that cover found for one call-out uses up the spare person the next one needs.

This run tested that guess, and it was mostly wrong.

A check for the second call-out. The new coverPairRisk takes every pair of days that each have cover alone and asks whether both callers' shifts can be covered at the same time. Pairs on the same day count too (two people off sick together). The check flags a pair when, for example, Ana is out Tuesday, Bo is out Wednesday, and the only person who could stand in for either has hours for one day, not two. Played the way the product actually handles call-outs (5pm notice, a reissue that keeps the posted shifts), the check catches 91.5% of the pairs that really open hours. 58% of the pairs it flags do.

Then the yardstick. The 20-week result had a flaw. Call-outs are drawn per person per worked day, so two different rosters face different call-outs, and 20 weeks is a small sample. Playing every pair of call-outs takes the dice out. On the café, the priced roster adds 2,023 open hours over all 253 pairs, against 2,218 for the plain one. Re-run with 200 seeded weeks instead of 20, it wins at both rates:

caféopen hours added, plain → pricedpremiums
1% sick-only262 → 208$1,687 → $1,212
8.4% all leave1,824 → 1,727$12,587 → $10,577

The README now says so.

A price for it anyway. On one sample team, a café that leans on one spare part-timer, the pair check found something the single check could not. --second-callout-price 1 charges $1 an hour for clashing pairs, and the build bought 10 hand-overs for $30 of score. Over 200 seeded weeks at 8.4% leave, that cut open hours added from 1,192 to 1,148, people moved from 1,275 to 821, and premiums from $20,776 to $13,005, compared with the slack price alone. On the other four sample teams it changed nothing or very little. At $25 it made the retail team worse, because it started trading single-call-out risk for pair hours. It also costs time: a 40-person week's pass went from 3 s to 44 s. So it is off by default, capped at 120 shift copies, and offered on the page as "Also pay for a second call-out in the same week".

What is not claimed: these are invented teams and seeded weeks, played through the product's own model of a call-out. No real week is in this.

Details, tables and every script: the README section "A second call-out in the same week" and eval/cover-pair-exposure.mjs.

All writing · More in Tools and projects · More on Shift Roster Builder · All tools and research