The Capacity Commons

The story

05:42 — twenty-one beds that do not answer the question

Freezing rain has closed one county road and slowed the main east-west route. An external regional forecast revises the 06:00–12:00 demand window for oxygen-capable adult medical service from five aggregate units to eight. In this scenario, a service unit is only the aggregate measure used for supported capacity in that category and interval; it is not a patient, room, or placement. Grid accepts the named forecast revision as an input. It neither creates nor validates the forecast.

The existing regional dashboard still reports twenty-one open beds. Eli Mercer, the overnight coordinator at the invented Lakebridge Medical Operations Coordination Center, sees a harder truth. Four service units are supported and time-qualified now. Three more are conditional on named operational gates. The rest are nominal rooms, in the wrong service category, outside the receiving window, unstaffed, or backed by stale evidence.

The region has eight units of declared demand and four it can currently defend. Three may become real. One is missing. Eli's screen does not say “no beds.” It says one unit short, three conditional, two evidence conflicts. Each state has a source revision, time boundary, and reason.

That language matters. A room is not the same as a supported service window. Authorization is not the same as completed work. At 05:42, even the conditional units are only possibilities: the operational owners have not committed, and the source systems have not reported their gates complete.

05:44 — the ward publishes a promise, not a spreadsheet

Alder 6, a respiratory-medical ward at Lakebridge University Hospital, runs its model inside the hospital network. Configured adapters receive locally permitted operational events: bed state, staffing coverage, pharmacy queue state, equipment confirmations, transport windows, environmental-services work, and the fact that accountable care teams have already made particular clinical decisions.

Grid does not make those decisions. Clinical clearance, the accepted discharge plan, patient or caregiver preference, and the selected acceptable provider remain externally owned, expiring predicates. If an owner revises or withdraws one, the affected capacity declaration becomes unavailable or unresolved.

Direct identifiers and row-level records remain within the hospital boundary under the authored design. That is a deployment and policy choice, not an automatic privacy guarantee. Alder 6 publishes a deliberately small aggregate promise: a service category, supported quantity, expected interval, validity time, evidence state, binding gate, and source-model revision.

Two rooms labeled “expected” on the old board become useful only if three chains finish in time: discharge medication, prescribed home-oxygen confirmation and delivery, and environmental-services turnaround. Rosa Alvarez, the ward flow coordinator, refuses to count them yet. “Expected” does not identify which dependency remains unresolved or who can resolve it.

05:47 — three ordinary facts fracture the plan

Within four minutes, a home-health provider cancels four morning visits because two staff members cannot travel. The regional oxygen supplier moves its earliest rural delivery window from 07:20 to 08:35. An environmental-services call-out lengthens Alder 6's isolation-room turnaround queue.

The ward model recalculates the affected dependency chain and republishes only the aggregate conclusion that changed. The hospital's capacity promise updates; the regional model then reevaluates its declared horizon. Raw ward datasets are not pooled into a regional solver.

Eli watches a green path turn amber, then unresolved. The regional explanation points to the hospital declaration. The hospital declaration points to Alder 6's oxygen-confirmation gate and the supplier's 05:47 revision. The coordinated experience follows model-local explanations and carried source coordinates; it does not pretend Grid automatically provides one universal cross-node proof.

The value is not merely that a screen updates. It is that authorized participants can see which accepted fact invalidated which represented conclusion. Four supported plus one conditional now leaves three unsupported inside the horizon.

05:52 — the bed problem becomes a coordination problem

The regional model evaluates three predefined alternatives over selected, declared inputs.

Alternative A opens a ten-bed overflow unit. The rooms exist, but the required staffing skills and oxygen support do not. The plan is infeasible in the required window.

Alternative B uses the eastern community hospital. It has physical space, but its relevant capability declaration expires at 08:00, and the ice route makes the last transport cycle miss the receiving window. Space outside the usable interval is not capacity.

Alternative C assembles a mobile capacity-release cell. It does not create beds or alter a clinical decision. It coordinates five administrative actions across shared dependencies blocking already authorized care transitions: move one discharge-pharmacy shift ninety minutes earlier; route the oxygen supplier's second vehicle from the western depot; use two predeclared, uncommitted community rapid-response slots after their owner confirms that existing service will not be displaced; temporarily move one environmental-services worker from a lower-pressure zone; and protect two uncommitted medical-transport windows before the dialysis peak. The scenario does not assign those actions one-for-one to beds or service units.

The composed planning and routing calculations test the represented precedence, resources, calendars, road windows, and uncertainty. Alternative C becomes conditionally feasible: four supported plus four conditional could meet the demand of eight. Four external owners still have to commit to the five actions before their gates expire.

06:03 — the model stops where authority begins

Hospital operations owns the pharmacy and environmental-services changes. The oxygen supplier controls its vehicle and delivery commitment. The community provider controls its rapid-response slots. EMS and the mobility providers control transport. Care teams retain clinical decisions, and established sending and receiving processes retain patient-specific acceptance.

Eli convenes the owners around one exact plan revision rather than four screenshots. Their surfaces emphasize different responsibilities but reference the same represented plan. At 06:03, configured integrations record the external commitments against that revision. The display says feasibility is conditional, commitment evidence is complete, external authorization is satisfied, and reported execution is in progress. It never says “clinically approved by Grid.”

Authorization does not manufacture capacity. Four units remain conditional until authoritative operational sources report that the gates actually completed. A message receipt proves only the recorded event, not physical execution or clinical effect.

06:07 — a bounded interruption keeps uncertainty visible

The wide-area link to Alder 6 drops for ninety seconds. The ward model keeps computing against the sources it can reach, while Eli sees the last delivered declaration with its observation and validity times. An authored freshness rule prevents that value from remaining current forever. When the link returns inside the configured delivery bounds, the pending aggregate is retried and the downstream conclusions recompute. This shows only bounded local continuity, not indefinite offline coordination or automatic conflict-free reconciliation after a long partition.

08:47 — a future bed becomes a credible service window

The relevant source systems report that all four conditional operational chains have completed. Rosa watches Alder 6's declaration change only after its gates are current. The hospital and regional dependency paths resolve. Supported capacity changes from four to eight only where the declarations' effective intervals overlap the portion of the 06:00–12:00 horizon that remains after 08:47. The update closes the represented shortfall within that overlap; it does not rewrite the unsupported earlier period or promise capacity outside those declarations.

Any patient-specific transfer remains in the established clinical, legal, and organizational workflow. Grid has not selected a destination, secured acceptance, or caused an outcome. The region in this scenario simply avoided confusing ten rooms with ten solutions. It found a small, bounded set of cross-organizational actions that could make supported capacity usable in time.

Three weeks later, analysts align the separately retained ward, hospital, and regional revisions. They see a recurring modeled association: scheduled dialysis transport consumes much of the small accessible vehicle pool on certain mornings, connecting care-transition delay, room turnover, hospital declarations, and the rural transfer queue. Humans test alternatives. The leading authored option is not another ward; it is a redesigned transport agreement that does not displace dialysis service.

Grid does not establish causation by itself. It makes the shared dependency inspectable. The ward model learns elapsed-time behavior; the institution learns which process to reconsider. Those are two different learning loops, held together by evidence and kept apart by authority.

Where the model stops

The model does not triage, place a patient, activate a transport, direct a clinician, or create capacity. It presents bounded alternatives to the people and institutions that already hold those responsibilities. Acknowledgment of a message is not evidence that a transfer occurred or that a patient outcome was achieved.

What remains to prove

A real evaluation would need approved aggregate contracts, privacy review, facility and transport integrations, policy enforcement, failure injection, usability testing, and retained evidence from the exact controlled configuration. The public Scenario demonstrates the question and the boundary—not deployment readiness.