IP-safe project evidence study / R2 V2 / 2 November 2026
How can an operating application connect asset evidence, feasible maintenance and responsible authority? The project combined a synthetic scheduling investigation, a public fleet analysis, domain standards and a local authority-control exercise. These evidence populations answer different questions and remain separate.
| Model element | Declared setting |
|---|---|
| Assets | Three service nodes; at least two healthy per modeled slot |
| Work | Two chains: drain / fan replacement / return and drain / patch / return |
| Resources | Two technicians with different skills, one scarce kit and timed fan receipts |
| Time | Six-hour horizon in half-hour integer slots |
| Downtime | Continuous from drain start through return, including waiting |
| Unknown critical evidence | Hold rather than invented data or a claim of impossibility |
The direct planner constructed earliest-feasible chains in both orders. Richer workflows used ordinary mixed-integer optimization with separate checking. An exhaustive enumerator supplied a reference for the restricted fixtures. Its implementation was separate, but the same author selected the assumptions and wrote the implementations. [P1]
Review found that the original 24-row comparison contained only sixteen distinct modeled states and overlapped development states. The original record was retained, the exposure claim withdrawn, and a new confirmation set excluded previously used states. This is a concrete correction to the evidentiary design.
| Confirmation measure | Retained result |
|---|---|
| Distinct modeled states | 24 |
| Complete plans | 8 |
| Priority deferrals | 2 |
| Evidence holds | 3 |
| Finite-model impossibilities | 11 |
| Schedulable cases | 10 |
| Direct maintenance downtime | 93 half-hour node slots |
| Richer workflow downtime | 90 half-hour node slots |
| Additional completion | None |
| Final gain from retention or mathematical adaptation | None beyond conventional optimization |
In one confirmation case, patch work could not begin before slot eight. The direct plan drained the node at slot five, creating avoidable unavailable waiting. The optimizer delayed the drain to slot seven. All six tasks still finished at slot eleven, but downtime fell from eleven to nine node slots. This explains why the objective changed without additional completed work. [P1]
Across ten schedulable confirmation cases, three saved half-hour node slots equal 1.5 modeled node-hours. They are not measured service hours, repair savings or customer productivity. The experiment does not isolate Backpass from optimization and other workflow differences.
The native EMSM component refused the maintenance use under its gold-domain scope. A separately declared temporal-accounting adaptation improved on endpoint-only checks but tied conventional temporal accounting. Final schedules did not improve beyond the optimizer. Retaining that result avoids attributing an advantage to added complexity.
The retained NYC snapshot is dated October 30, 2026. It is descriptive administrative evidence, supplemented with year-to-date averages and annual context. The web data and report reflect the same underlying City evidence, not independent sources. [P2]
| Critical group | Target | Available | Shortfall |
|---|---|---|---|
| Collection trucks | 1,253 | 1,092 | 161 |
| Dual bin collection trucks | 581 | 489 | 92 |
| Sewer and water | 1,014 | 963 | 51 |
| Forestry | 155 | 133 | 22 |
| Material hauling | 224 | 207 | 17 |
| Traffic | 481 | 471 | 10 |
| Buses | 105 | 96 | 9 |
For each group, shortfall = max(target - available, 0). Gross shortfall sums positive gaps. Across all critical groups, gross shortfall was 378. Surpluses elsewhere totaled 201, leaving net shortfall 378 - 201 = 177. The two collection groups contributed 161 + 92 = 253, or 253 / 378 = 66.9% of gross shortfall.
An available ambulance cannot automatically substitute for an unavailable collection truck. The net total can therefore conceal operationally relevant group gaps. The public numbers support prioritizing investigation, with actual service demand and substitutes still required before scheduling repairs.
The citywide target minus available count was 27,082 - 25,887 = 1,195. Agency targets summed to 26,010, yielding an agency net gap of 123. The difference is 1,072 in target levels plus 123, not evidence of 1,072 uncounted vehicles. Agency and critical-group views overlap and must not be added.
| Annual measure | FY2025 | FY2026 |
|---|---|---|
| In-service rate | 87% | 87% |
| Days target achieved | 64% | 52% |
| Average age, months | 90.0 | 100.0 |
| Repair personnel | 1,301 | 1,266 |
| Nominal repair expenditure, millions | $327.0 | $358.7 |
A steady average can coexist with fewer days reaching the target. These aggregate measures suggest investigation of readiness pressure. They cannot identify the separate causal effects of staffing, parts delays, age or procurement. Fiscal 2026 ends before the October snapshot; annual and daily populations must not be subtracted as if contemporaneous. [P2]
An open-data endpoint returned 39 records, all dated July 15, 2022. The study excluded it from current-readiness calculations. Seasonal N/A rows were excluded rather than treated as zero availability. No daily time series or consecutive-shortage measure was reconstructed.
Thirteen synthetic checks used an unchanged existing authority component. They covered issuance, exact resource scope, commencement, replay, unknown reservations, reconciliation, revocation and conservation of cumulative allocation. All passed in the harness. The harness supplied outcomes and did not execute maintenance. All 135 operational acceptance cases remained NOT_RUN. [P3]
A shadow pilot should collect work orders, repair states, parts timestamps, skills, bay occupancy, route demand and observed service outcomes. Compare against a frozen conventional process. Advance only after the actual owners can inspect the evidence and the complete action route is qualified.
| Issue found | Implemented response |
|---|---|
| Development overlap in initial experiment | Withdrew the exposure claim and froze distinct confirmation states |
| Gross and net fleet gaps conflated | Reported 378 gross, 201 surplus and 177 net separately |
| Separate implementations implied independent science | Disclosed same-author model and evaluation exposure |
| New V2 interpretation could rewrite old evidence | Retained original experimental results unchanged |
| Local control tests could imply field admission | Kept 13 synthetic checks separate from 135 unrun operational cases |
The project illustrates evidence states, explicit constraints, bounded review and retained authority history working toward one operating application. It also shows a useful limit: the conventional optimizer explained the final schedule advantage. A credible Epistemic OS must preserve that conclusion even when a more expansive claim would be easier to promote.
[P1] Coordinating Physical Work Through Governed Decisions / Physical Work Research Monograph, sections on model, confirmation, mathematics and authority.
[P2] Where Municipal Fleet Availability Falls Short / Municipal Fleet Readiness Study, sections on concentration, target reconciliation, historical context and methods. Retained NYC Fleet Daily Service Report, October 30, 2026, and Fiscal 2026 Mayor’s Management Report, printed p.481.
[P3] Methods, Review and Human Decisions, plus thirteen-case local authority exercise. All supplied in V2G Physical World OS final archive. Tables transcribed from those documents and arithmetic checked in this expansion.
Underlying source snapshots, calculation records and selected test evidence can be reviewed under an appropriate NDA or contract. No live equipment control, independent field validation or measured service improvement is claimed.