Architectures/Demand · constraints · execution
Public reference architecture

RouteNet Optimization

A reproducible plan whose trade-offs and unassigned work remain explainable.

Keep source demand, locked work, constraint revisions, optimization objective, candidate solution and publication receipt distinct.

Know which component owns the state—and what it must never infer.

Every layer exposes an authoritative responsibility and an explicit non-authority boundary.

01
Orders / workforce / fleets

Demand and resources

OwnsWork identity, requirements, locks, capacity and availability

Must not inferMissing inputs are not silently defaulted

02
Routing / matrix / traffic

Road-cost context

OwnsProfile-specific travel cost and version

Must not inferMatrix assumption remains visible

03
SAAS-03

Optimizer

OwnsValidated candidate assignment under one objective/constraint set

Must not inferInfeasible work remains unassigned with reason

04
APP-05 / APP-06

Review and execution

OwnsOperator publication intent and field acceptance

Must not inferReference publication does not dispatch a driver

State advances through evidence—not optimistic UI.

Each transition names both the action and the identity or version evidence that makes it reproducible.

  1. 01

    Validate

    Reject duplicate demand and impossible source dimensions before solve.

    input revision · schema · validation issues
  2. 02

    Build costs

    Bind vehicle profile, territory, traffic window and matrix version.

    matrix/route versions · cost units
  3. 03

    Solve

    Apply capacities, skills, shifts, windows, locks and objective deterministically.

    solver/config seed · constraints · objective
  4. 04

    Verify

    Independently recheck every assignment, capacity, time and uniqueness invariant.

    check list · failures · result fingerprint
  5. 05

    Publish intent

    Record immutable plan revision for downstream review without claiming dispatch.

    plan ID/revision · idempotency · actor

Interfaces that a production program must own.

01

Planning input

Demand, resources, locks, windows, skills and revisions.

02

Cost matrix

Profiles, cells, traffic basis, units and validity.

03

Optimization result

Assignments, sequence, totals, violations and unassigned reasons.

04

Publication/execution

Plan revision, acceptance, actual events and replan lineage.

Failure states stay truthful and useful.

Oversized/ineligible work

Leave unassigned and name the exact violated capacity/skill rule.

Duplicate demand

Reject before solve to prevent double assignment.

Matrix/traffic unavailable

Use only an explicitly approved fallback basis or block planning.

Execution diverges

Preserve original plan and record actual/replan lineage separately.

Privilege follows the narrowest useful boundary.

  • Tenant/program boundary for orders, drivers and routes
  • Purpose-limited precise location and workforce data
  • Immutable input/result fingerprints for reproducibility
  • Idempotent publication and revision conflict
  • No customer ETA or dispatch claim without downstream authority

A green demo is not a production acceptance case.

  • 01Constraint catalogue and boundary-value fixtures
  • 02Determinism and independent-result verification
  • 03Scale/performance against declared planning windows
  • 04Unassigned explanation and locked-work tests
  • 05Plan-to-execution and replan lineage validation

Compose the system without collapsing product ownership.

Each product can be bought and operated independently while sharing identity, context and lifecycle contracts.

OEM program workshop

Turn the RouteNet Optimization reference into your program architecture.

Confirm target products, vehicle and cloud boundaries, source systems, contract versions, deployment, validation and lifecycle ownership.