How it decides.

Four of them, as the grid draws them. The rest of the register is below.

Applied before the search. Each returns the reason you saw above.

• Only active staff enter the search

  Employment status

  Only ACTIVE staff enter the search at all.

  scheduler.py:224

• Trainees only fill shadow slots, and vice versa

  Training segregation

  Both directions are enforced, so a trainee cannot quietly become the coverage.

  scheduler.py:231-235

• Must hold the role the slot asks for

  Role held

  The person must hold the role the slot asks for.

  scheduler.py:238-243

• Must be assigned to that site

  Location assignment

  The person must be assigned to that site.

  scheduler.py:241-244

• Every certification the role needs, current

  Role-required certifications

  A lapsed certificate removes the person from the slot, not from a warning list.

  scheduler.py:246-250

• Inside their start and end dates

  Employment date window

  Both are checked per shift date, not once per roster.

  scheduler.py:256-264

• Week-on/week-off cycles, from an anchor date

  Multi-week roster pattern

  Longer cycles too, computed from a cycle length and an anchor date, so an off-week person is never offered.

  scheduler.py:266-275

• A specific date beats the usual pattern

  Date-specific override

  UNAVAILABLE blocks the day or just the overlapping hours. AVAILABLE replaces the recurring window for that date rather than adding to it.

  scheduler.py:279-317

• Shift sits entirely inside a declared window

  Recurring availability

  Partly-inside is not eligible.

  scheduler.py:319-334

• Full days, and half days by morning or afternoon

  Approved leave

  Half-day leave blocks only the matching half: AM blocks starts before noon, PM blocks starts from noon.

  scheduler.py:336-351

• Minimum and maximum length, overridable per site

  Shift length by employment type

  A per-location override outranks the org default, so a three-hour engagement minimum at one site does not have to become the rule everywhere.

  scheduler.py:353-362

• Curfews and shift caps, per age band, per country

  Young-worker rules

  Curfew start, curfew end and maximum shift length. Bands are configurable per organisation and default from the country preset.

  scheduler.py:364-394 · lib/minor-work-rules.ts

A roster breaking one of these is not scored badly. It cannot be represented.

• Minimum, ideal and maximum headcount per slot

  Staffing ceiling

  The maximum is a hard ceiling the model cannot exceed. Slots between ideal and maximum are allowed but scored against.

  scheduler.py:626-656 (CONSTRAINT 1)

• No mentor on the floor, no shadow shift

  Mentor pairing

  A trainee in a shadow slot requires a non-trainee working an overlapping shift at the same site and role.

  scheduler.py:668 (CONSTRAINT 1a)

• A slot can require several certified people

  Certification coverage

  Not merely one.

  scheduler.py:751 (CONSTRAINT 1b)

• A first-aider on the floor every operating hour

  Certification time-coverage

  Enforced by hour, not by day. Not simply someone certified rostered somewhere that day.

  scheduler.py:781 (CONSTRAINT 1c)

• One person cannot be in two places

  No overlapping shifts

  No two shifts that overlap in time on the same date.

  scheduler.py:846 (CONSTRAINT 2)

• Always hard, never a warning

  Maximum hours per day

  Not a preference, not a flag after the fact.

  scheduler.py:895 (CONSTRAINT 3)

• Week, fortnight or month, whichever you run

  Maximum hours per pay period

  Scaled to the organisation’s pay period rather than assumed weekly. A fortnightly cap is enforced fortnightly.

  scheduler.py:943 (CONSTRAINT 4)

• No close followed by an open

  Minimum rest between shifts

  Reads shifts outside the range being filled, so the gap is still enforced across the boundary into the week you already published.

  scheduler.py:1008 (CONSTRAINT 5)

• Counts days already worked before this roster

  Maximum consecutive days

  Boundary-aware, so a run does not reset when a new roster starts.

  scheduler.py:1079 (CONSTRAINT 6)

• Two days off means two together

  Minimum consecutive days off

  A rolling-window check, so two separate single days do not satisfy a requirement for two consecutive.

  scheduler.py:1139 (CONSTRAINT 6b)

Genuine trade-offs. A higher tier can never be outbid by a lower one.

• The core goal, and nothing outbids itMEDIUM

  Fill every slot

  The tier every other consideration sits beneath.

  scheduler.py:1714

• Allowed to the maximum, scored past idealMEDIUM

  Overstaffing above ideal

  Every slot past ideal costs the solver something.

  scheduler.py:1753

• Heavy reward for giving everyone a shiftMEDIUM

  Nobody sent home with nothing

  In hospitality a casual rostered zero hours is a retention problem before it is a scheduling one.

  scheduler.py:1794

• Shortfall against a contracted commitmentMEDIUM

  Minimum hours per period

  For staff whose agreement is expressed in hours.

  scheduler.py:1355 (CONSTRAINT 7)

• For agreements written in days, not hoursMEDIUM

  Minimum days per week

  Tracked separately from the hours commitment.

  scheduler.py:1206 (CONSTRAINT 8)

• Grouped days beat a fragmented weekMEDIUM

  Consecutive working days

  Prefers a run over single shifts scattered with gaps between them.

  scheduler.py:1311 (CONSTRAINT 9)

• Under-hours penalised three times as hard as overSOFT

  Asymmetric fairness

  Short-changing someone is treated as the worse error, and that is a constant in the source rather than a claim in a brochure.

  scheduler.py:1734-1735 · W_UNDER_TARGET = -3, W_OVER_TARGET = -1

• Multi-site staff stay where they usually workSOFT

  Home-site preference

  Where the rest of the roster allows it.

  scheduler.py:1750-1830