# Model and reporting note · v0.4

This is a fictional systems game, with substantial AI assistance. Consultation → optional imaging → care is an invented pathway, not a universal patient journey. The community scenario optionally adds specialist planning after care. Priority flags are scenario data, not clinical triage. Money, staffing skills, fatigue and outcomes are game parameters.

The pure simulation advances in fixed one-second ticks; rendering interpolates up to four quarter-grid walking segments. The authored walk is 0.90 seconds at 1.65 m/s. Runtime walk phase scales with actual modeled distance; yielding stops walking animation. Smaller routing nodes, occupied-position rejection, reserved movement trails and deterministic detours reduce crowd intersections. Separate entrance and exit paths prevent the shared-door jam observed in automated testing. This is conservative grid traffic, not a biomechanics or continuous crowd-dynamics solver. Characters still turn on grid segments; clinician prop contact and interior room access are approximations.

Each 2×2 room has a visible front threshold/ramp, capacity one, service duration, role skill, team requirement, upgrade level and optional decoration. Reservations cover walking, room entry, bed entry, service, bed exit and room exit. The doorway remains unavailable to corridor traffic throughout. New capacity cannot leapfrog a visit on the tick it finishes. Construction checks bounds, room overlaps, doors, entry/exit connectivity, people and reserved paths. Removal rejects reserved rooms and rooms used by returning clinicians. Decorative props use fixed sockets clear of circulation; they alter appearance and budget, not an unexplained performance score.

Clinicians have explicit compatible skills, a shift end, fatigue and a local personality tradeoff: quick thinkers work faster but accumulate more fatigue; careful listeners work slightly more slowly with less fatigue. Skills and traits are fictional, independent of appearances. Staff can be reassigned only after reserved work; a second room consumes a compatible available clinician. Recruitment is closed in the limited-team scenario. Protected breaks walk to a lounge or quiet bay, rest for 16 ticks, recover fatigue, then return. Their assigned module remains unavailable during travel and rest. Lounges improve recovery; there is no global service-speed bonus. One ambient discussion figure is scenery and contributes no capacity.

After each intermediate service, a handoff takes four ticks under bounded automation, or waits for explicit acknowledgement when automation is off. Communication support reduces that specific delay to one tick. Unsupported environmental access flags create nine waiting ticks; access support removes that obstacle. A community workshop investment changes only the next community shift’s modeled access delay to four ticks. It does not prevent disease, change intrinsic worth or assert a real intervention effect.

Waiting counts stationary and entrance-buffer waiting, including access delay; walking counts route and interior doorway time; collision yielding counts as waiting; handoff, equipment transfer and service have separate totals. Entering and leaving a room each take three ticks. Bed transfers take three ticks in each direction and use exported animation clips. Completed-only statistics omit people who departed or remain inside, so reports also expose all-arrival-so-far means, medians and p90, counts and access-cohort denominators. Experience is an unvalidated game indicator, not patient satisfaction. Conservation is arrivals = completed + departed + unfinished. Completion/departure is counted on reaching the exit. At 480 ticks or exhausted funds, unfinished visits are retained honestly.

All exogenous arrival times, flags and goals are precomputed from the seed. Construction, routing and policy do not consume arrival randomness. Extra user-triggered surges alter the tape explicitly and are logged. The lab captures a complete checkpoint and branches only while simulation time is unchanged. It replays identical arrivals and existing visits per pair. The primary pair uses the exact tape; four additional paired future-attribute variants provide sensitivity. The range is not a confidence interval. Changed layout, staff, cash and policy fields are identified. Workers preserve responsiveness. Model changes across versions invalidate unlabelled comparisons.

Five scenarios provide distinct constraints; sandbox opens all, while campaign unlocks the next scenario through stated multi-outcome mastery. Specialist planning unlocks in the community campaign stage. Goals require helped counts, departure limits, nonnegative funds and at least 40% team energy. These thresholds are tuning, not validated humane outcomes. No opaque combined morality score is used. Saved v0.2 visits are migrated without discarding IDs/counts/reservations; the migration action marks changed timing definitions and regenerated future v0.3 arrivals. The original Canvas save and /legacy/ remain untouched.

Device-local JSON contains version, seed, tape fingerprint, layout, decisions and timestamps, visits, assignments, timing definitions, unfinished people and results. No analytics or collection endpoint exists. Imported JSON is validated before replacing the current shift; nested undo snapshots are discarded. Malformed saves retain a recovery copy.

Operating costs every 15 ticks: 25 + 7 per room + 4 per clinician + 8 access support + 4 handoff support. Starting funds 8000; consultation 900, imaging 1500, care 1100, lounge 600, specialist planning 1400; upgrade 650; recruitment 800. Access support 500, handoff support 350, future community workshop 700. Completion contributes 390; room removal refunds 50%. These are fictional game economics.

New v0.4 visits classify each elapsed tick once. Historical v0.3 categories retain an explicit elapsedOffset, so category totals equal elapsed visit time plus that documented offset. Migration preserves IDs, outcomes and reservations while recalculating old paths and moving any newly blocked corridor position to the nearest clear node. The migration action is exported. Neighborhood arrivals use18-second ordinary spacing and a130–165second surge; the goal is12 completed with at most8 departures. The guide asks players to add consultation and then upgrade care. These parameters changed after adding real doorway-navigation time and platform clearance.