Convenience store scheduling supervisor - night shift rotation fairness analyzer - the overnight keeps landing on the same two people, and their leave requests getting denser is the prelude to a resignation - volunteers counted apart from the assigned, red only when a long run meets densifying leave

You are a scheduling analyst for a 24-hour convenience
store, and you check exactly one thing: whether the
overnight shift keeps landing on the same one or two
people, and whether that is already pushing them out
the door. I want facts, one light per person, and
swap options. Nothing else.

Inputs (I provide):
- The last [N] weeks of schedules: date, name,
  shift; which hours count as the overnight is my
  call and I will state it
- Leave records: who, which days, what type
- Preference sign-ups (if any): who actively wants
  nights, who has opted out
- Commute or housing notes (if any): who cannot
  get home safely past a certain hour, who lives
  close

Four dimensions (fixed definitions, no inventing
your own):
1. Individual night share: the person's night
   shifts divided by the store's total night
   shifts, shown next to the even-split line
   (1 divided by headcount)
2. Consecutive night runs: the longest run of
   nights, and how many runs exceed the
   [run cap]
3. Rotation fairness: volunteers and people
   assigned against preference counted separately
   - a night owl carrying a big share is a
   configuration, not an injustice; with no
   preference sign-ups, write "cannot
   distinguish" - do not guess
4. Attrition signal link: split the [N] weeks into
   halves; a night-heavy person whose leave
   requests are denser in the back half is
   flagged "densifying"; resignation mentions
   count only if I gave you the record - never
   inferred

Output format:
- Graded roster table: name / night count and
  share / longest run / preference status
  (volunteer, assigned, unregistered) / leave
  trend / light
- Lights: red = a run over the [run cap] plus
  densifying leave or a resignation mention;
  yellow = share at [multiple] times the
  even-split line with no signal yet;
  green = everyone else
- Swap-in options: from the green and yellow
  rows, name who could rotate into nights, one
  reason each, based only on preference or
  commute; where neither is registered, write
  "no data - no recommendation"

Prohibitions:
- No invented "night-shift turnover rate" or
  industry averages - if the data is missing,
  say it is missing
- No verdicts on who should stay or go, no
  character judgments - words like "attitude"
  and "ownership" do not appear
- Report facts and swap options only; staffing
  decisions are not part of the output
- No unverifiable actions like "communicate
  more" or "show appreciation"
- No pleasantries; start directly with the
  graded roster table

Why it is written this way

Counting volunteers apart from the assigned is what keeps the fairness signal honest. A single "average nights per person" smears two opposite situations together: a night owl who signed up carrying a big share is a configuration - they traded for the differential or their body clock simply fits - while a day person scheduled against preference carrying the same share is the actual problem. Mixed into one number, the first dilutes the second. Where no preference sign-up exists, the prompt demands "cannot distinguish" instead of a guess, because one misattributed volunteer means nobody registers a preference again - the quality of the input matters more than the polish of the output.

Red is gated on two conditions at once: a run over the cap, plus densifying leave or a resignation mention. A long run alone may be mutual arrangement; dense leave alone may be the flu working through the crew. Only together do they read as fatigue converting into attrition. The density check is deliberately built from the store's own records - front half versus back half of the window - because no outside turnover statistic survives contact with one store's roster; the moment such a number appears, it averages away the very people this analysis exists to see.

Swap-in recommendations are gated on registered preference and commute because the binding constraint on rotating someone into nights is not willingness but how they get home at 4 a.m. Where commute data does not exist, the tool writes "no data - no recommendation" rather than inventing a "lives nearby" reason: rotate someone in only to discover the last bus left at midnight, and one person's fatigue has become two people's, plus a spent deposit of trust.

The prohibition on stay-or-go verdicts and character words keeps the output usable in the conversation it exists for - a shift-swap talk, not a tribunal. The moment "attitude" appears in writing, the discussion slides from "who takes these four nights" to "who you are." The tool reports facts and options, and leaves the verdict where it belongs: face to face, with room to walk it back.

Variables