The Block That Had No Department

The story

15:47 — one incident enters five systems

A pressure sensor and a Water dispatcher report a main break under Alder Street while Public Health's external heat status is active. Streets says the eastbound lane must close. Housing reports that Alder House's confirmed support arrangement remains adequate only through 17:12. Emergency Management opens its incident-coordination record.

Maya Okafor, Ward 9's service director, owns the block-level coordination record. Her goal is to assemble a current, reviewable plan that keeps the represented building-support commitment covered while Water repairs the break. She does not acquire authority over Water, Housing, Public Health, Emergency Management, or Streets by coordinating their facts.

Every record is valid within its own department. None answers the resident's complete question: what happens on this block, in what order, under whose authority, and with what evidence? The live clock is explicit: the confirmed support arrangement expires at 17:12 while the repair time remains uncertain.

Ward 9 does not inspect individual medical or disability records and does not infer whether residents are safe. Housing publishes only the aggregate operational predicates authorized for the purpose: the building's support state, how long it remains valid, whether the accessible transport interface is confirmed, and the source revision.

Water remains authoritative for the main, repair procedure, water-quality process, and restoration declaration. Streets owns traffic control and reopening. The other departments retain their corresponding responsibilities. The ward model composes those declarations; it cannot manufacture them.

Six weeks earlier — rehearse the rule, not the emergency

Before heat season, invented Bellwether runs an isolated exercise covering the same decision pattern: a critical-place water-main break, several externally supplied repair-duration cases, a closed arterial, an unavailable support vehicle, reduced crew calendars, and modeled and live connectivity interruptions.

The exercise exposes a policy defect. Under the old rules, every ward protects its own equipment even when another ward has a time-critical gap. City staff author a cross-ward alternative, a hard rule protecting any already active site, and department-owned authority lanes. A separate governed process approves the protocol. Simulation evidence does not release an action.

That rehearsal makes the live response faster without making it autonomous. The rule has been tested; the current facts, approvals, and execution evidence still have to earn each transition.

15:48 — estimate the shared gap locally

Ward 9 runs a local Grid model near its sources and operators to estimate elapsed minutes until Water reports the repair and required restoration gates complete. Its signed city base component returns 245 minutes, and its separately pinned local component adds 137. The estimate is intentionally narrow. It is not hydraulic engineering, failure diagnosis, a water-safety judgment, or an emergency-health decision.

Together they produce a represented completion time of 22:09, 297 minutes after confirmed building support ends at 17:12. Bellwether's policy is more cautious than the estimate: it requires an initial protection commitment through 22:30 and a new decision at 21:45 if restoration has not been declared. The longer 480-minute stress case remains visibly uncovered unless commitments are renewed.

The numbers are scenario planning inputs, not a guarantee. Maya sees the full distinction: support is current for eighty-four more minutes, the point estimate leaves a 297-minute gap, the policy horizon requires 318 minutes of coverage, and the required departmental authorities have not yet acted.

15:49 — selected values leave; local history stays

Ward 9 publishes only the values the city and department models need: repair estimate, support expiry, gap, protection horizon, required Water and support units, evidence state, observation and validity times, and the identities of the models used.

It does not publish resident records, row-level incident history, unselected sensor data, the training dataset, model weights, or the entire ward workbook. Water can use the time horizon. Housing can use the support gap. Streets can use the repair sequence and closure interval. The city-continuity model can use the resource quantities. One local conclusion becomes a shared dependency without becoming a central copy of local data.

15:51 — the city tests three ways to keep the promise

The city model compares three authored alternatives against current equipment, crew, depot, travel, commitment, and operating-rule declarations.

Alternative A waits for Ward 9's own resources. The repair may proceed, but Alder House support expires at 17:12. It fails both the point-estimate interval and the required protection horizon.

Alternative B moves the nearest Ward 9 mobile unit. It looks fast on the map, but a hard rule rejects it because the unit is already supporting an active cooling site. Bellwether may not keep one promise by silently breaking another.

Alternative C composes a cross-ward bundle: two externally certified Water units from Ward 11, one accessible support unit from Ward 7, and a Streets traffic-control crew. The represented bundle covers the block through 22:30, with a 21:45 extension gate for the longer case.

Grid has not found the morally correct answer. It has found one feasible bundle inside facts and policies people supplied. The nearest resource loses for an exact public-obligation reason. A different rule, stale fact, or unavailable crew could change the result.

The plan retains the Water and Streets sequence from incident classification and utility-locate state through traffic control, isolation, excavation, repair, and the externally owned flush, test, and restoration gates. The model can represent precedence and resources. It does not decide the engineering or water-quality outcome.

15:54 — four authority lanes

The proposed bundle reaches a bright boundary: evidence sufficient for review; feasible within authored bounds; authorization pending; execution not begun.

Maya cannot dispatch from the scenario surface. The Water duty manager approves the repair plan and use of Water resources. The city continuity duty officer approves cross-ward reallocation. The Streets traffic-control duty officer authorizes the traffic-control plan, closure state, and crew commitment. The Housing facilities manager confirms the building support interface.

At 16:03, the four authenticated decisions have been recorded against the exact revision. A separately qualified connector may now submit the predefined request to authoritative work-order or dispatch systems. Their technical acknowledgment establishes receipt only. It does not prove that a unit moved, traffic control was installed, a valve closed, water became safe, or a resident received service.

16:07 — ninety-six seconds without the city

The ward-to-city connection drops for ninety-six seconds. Ward 9 keeps evaluating local source changes because its model and state remain local, while the city retains the last delivered declaration with its observation and valid-through times. The link returns before the configured pending-delivery limit; revision and fingerprint checks reject stale or duplicate writes, and affected conclusions update. Nothing stays current merely because it is the newest available value, and this bounded interruption does not establish arbitrary partition recovery.

16:41 — approval becomes observable work

Water and support units report staged through their authoritative systems. Streets reports traffic control established. Housing reports the building interface ready. Thirty-one minutes before the original support limit, Maya watches three independently owned declarations become current.

The plan did not become true when it was approved. It became usable as authoritative work reports arrived.

At 21:45, Water has not declared restoration. A scheduled rule opens the extension decision rather than assuming resources will remain. Fresh commitments and the appropriate approval extend the support bundle to 23:50. At 22:21, Water's system reports the required repair and restoration gates complete. Actual elapsed time is 394 minutes—twelve minutes longer than the authored estimate. One case becomes evidence, not proof that the model was “right.”

Fourteen months later, governed completed records may support an inactive local candidate, subject to a city validation and release process Grid does not supply end to end. Analysts also see repeated delay in the serial handoff among utility locating, traffic control, and excavation authority, so Bellwether tests a protocol that prepares those independent lanes in parallel while preserving each release gate.

The local model produces an estimate. The connected model exposes a shared constraint. People change the rule that created it. Maya owns the complete coordination record: keeping the gap visible, routing the exact proposal to each responsible authority, and reconciling later evidence. She does not own any department's source facts, approvals, dispatch, execution, or outcome.

Where the model stops

The model can calculate, compare, and explain. It cannot spend public money, issue a health direction, dispatch emergency resources, close a street, approve a traffic-control plan, or send a crew. The operative workflow therefore names the accountable departmental and city-continuity authorities before an external instruction is eligible to be sent.

An external system's acknowledgment proves only that it received the submission. An authoritative execution report and a later outcome observation are separate records.

What remains to prove

This public Scenario makes the operating concept inspectable. A controlled evaluation would still need city-owned integrations, identity and policy controls, records obligations, negative-case tests, accessible operator surfaces, and retained evidence for the exact configuration. Until that work exists, the Scenario remains illustrative.