FleetOps

FleetOps, better known as fleet operations or fleet management, is the work of running a group of vehicles so that each one is safe, legal and available when the business needs it, at the lowest cost over its whole life. It covers choosing and buying vehicles, assigning and dispatching them, fuelling or charging them, inspecting and maintaining them, keeping drivers safe and within working-time rules, and deciding when to replace or sell each vehicle.

Reference descriptions of fleet management list leasing and financing, maintenance, licensing and compliance, accident management, telematics, driver, speed and fuel management, health and safety, and resale.^[1] The discipline makes a few long-lived decisions well (which vehicle, how many, how long to keep it) and many routine ones (today’s routes, this week’s inspections) without letting any become a crash or a roadside prohibition.

Written for beginners, fleet staff, safety and finance leads, and engineers who integrate fleet data. Sources last verified 2026-10-23.

The practice is established. NAFA Fleet Management Association describes itself as serving “more than 3,500 fleet professionals” from government and corporate fleets and suppliers,^[2] and its Certified Automotive Fleet Manager (CAFM) credential is organised into eight disciplines: asset, business, financial, information, maintenance and risk management, professional development, and vehicle fuel management.^[3] Public employers write job classes for it: Pinellas County, Florida, has a “Division Manager, Fleet Operations” covering maintenance and repair, vehicle specifications, surplus disposal and the fuel programme.^[4] Transport regulations define much of the routine work (see Methods, rules and frameworks).

The one-word label is different. “FleetOps” names several unrelated products: Fleetbase’s open-source “Fleet-Ops” extension, described as “Open-source fleet, dispatch, transport management, and real-time logistics operations for Fleetbase”;^[5] an iOS fleet app published by Hypernym LTD;^[6] and a transportation and logistics software company whose site now announces “FleetOps is now Class8”.^[7] No body, standard or community uses “FleetOps” as a discipline name, so this page teaches the established practice under the Atlas’s label and refers to none of these products.

Without deliberate fleet operations, five things go wrong.

1. People get hurt. NIOSH reports that “motor vehicle crashes are the leading cause of work-related deaths in the U.S.” and that from 2011 to 2022 more than 21,000 US workers died in work-related motor vehicle crashes, 35% of all work-related deaths.^[8] Worldwide, the WHO counts about 1.16 million road traffic deaths a year, with speeding, impaired and distracted driving and unsafe vehicles among the risk factors.^[9]
2. The fleet is the wrong size. US DOE guidance notes that fleets collect specialised, rarely used or unsuitable vehicles over time;^[10] each still costs capital, insurance and depreciation.
3. Cost is judged at the wrong moment. The cheapest vehicle to buy can be the dearest to own, and one kept too long costs more in repairs and downtime than its replacement.
4. Maintenance becomes reactive. Breakdowns cost a tow, a missed job and sometimes a crash; missing inspection records are a compliance failure even if the vehicles are fine.
5. Working-time rules are broken by planning. A dispatcher who adds one more delivery can push a driver over a limit.

• Not logistics, purchasing, the device platform or robot fleets. Modes and carriers are SupplyChainOps, contracts ProcurementOps, telematics devices IoTOps, robots RoboOps; FleetOps runs the organisation’s own vehicles and drivers and decides what to buy, how many and when to replace. Autonomous road vehicles sit on the boundary (use cases).
• Not a software category. A telematics subscription does not create a maintenance programme or a safety culture.
• Not only trucks. Sales cars, vans, buses, ambulances and employees’ own cars used for work (the “grey fleet”) are fleets. The UK Health and Safety Executive states that “A grey fleet vehicle is owned and driven by a worker for business purposes” and that health and safety law covers these vehicles as well.^[11]

Fleet management has no founding date. Rules and technologies turned vehicles into data sources and working time into something auditable; these are the milestones this page relies on.

Mental model: a fleet converts capital into available, safe vehicle-hours. Every decision moves one of four dials:

1. How many vehicles (right-sizing, utilisation).
2. Which vehicles and for how long (specification, TCO, economic life, EVs).
3. How each vehicle-hour is used (dispatch, routing, fuel, idling).
4. How safe and legal each vehicle-hour is (maintenance, inspections, driver safety, working time).

The dials interact: fewer vehicles means less slack for breakdowns; an EV changes energy, maintenance and dispatch at once.

flowchart TB
  PR["ProcurementOps"]
  SC["SupplyChainOps"]
  IO["IoTOps"]
  FL["FleetOps: vehicles, drivers, cost"]
  PE["PeopleOps"]
  RK["RiskOps"]
  RO["RoboOps"]
  GR["GreenOps"]
  PR -->|"contracts"| FL
  SC -->|"demand"| FL
  IO -->|"device data"| FL
  FL -->|"training needs"| PE
  FL -->|"road risk"| RK
  FL -.->|"remote assistance"| RO
  FL -.->|"carbon method"| GR
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  classDef neutral fill:#f1f4f6,stroke:#5b7a87,color:#122029
  class FL core
  class PR,SC,IO,RO,GR,RK,PE neutral

No body publishes a “FleetOps manifesto”; this charter is this Atlas’s proposal. Mission: provide the vehicle-hours the organisation needs, safely and legally, at the lowest whole-life cost, with records that prove it.

Five building blocks appear in every fleet.

1. The asset register. One record per vehicle: identity, class, mission, owner, ownership type, odometer and engine hours, planned replacement date and status (columns in the reference implementation).
2. The driver register. Licences, categories and expiry dates, fitness requirements where they apply, training, and permitted vehicles. In the US, Driver Fitness is one of the safety categories carriers are measured on.^[24]
3. The work plan. Routes, schedules and bookings: where utilisation and working-time rules collide.
4. Energy. Fuel cards, tanks, chargers and tariffs: often the largest variable cost, and where fraud and waste hide.
5. Maintenance and inspection. Schedules, defects, work orders, parts, workshop capacity, and the records that prove it.

Telematics, fleet software and the cost model sit on top of these blocks; they do not replace them.

Titles vary with size (a 30-vehicle fleet may have one coordinator for all of this); the roles matter more.

A RACI for recurring decisions (R = responsible, A = accountable, C = consulted, I = informed), proposed by this Atlas:

The rule that matters most: whoever decides a defect is safe to drive with must not be measured on keeping the vehicle on the road.

flowchart TB
  A["Need: mission and duty cycle"] --> B["Specify: class, powertrain, upfit"]
  B --> C["Acquire: buy, lease or rent"]
  C --> D["Commission: register, upfit, telematics"]
  D --> E["Operate: assign, dispatch, fuel or charge"]
  E --> F["Inspect and maintain"]
  F --> E
  E --> G{"Replacement review: past economic life?"}
  G -->|"no"| E
  G -->|"yes"| H["Remarket or dispose: wipe data"]
  H --> A
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  classDef decision fill:#fdf3e3,stroke:#c9800f,color:#122029
  class A,B,C,D,E,F,H core
  class G decision

Easy to skip: commissioning telematics correctly (a device on the wrong ignition source gives wrong idle data for years); a scheduled replacement review; and disposal that removes the device, deletes paired phones and stored addresses, and closes fuel cards and toll tags.

Four cadences, proposed by this Atlas; regulated fleets may need some items more often.

Write down the escalations: safety-critical defect → off road until certified; a dispatch that would breach working time → back to the planner; injury collision → preserve data.

A logical architecture: small fleets run most boxes as features of one product; large ones integrate several.

flowchart LR
  subgraph SRC["Sources"]
    direction TB
    K["Dash cameras"]
    E["ELD or tachograph"]
    V["Vehicle gateway"]
    F["Fuel cards, tanks"]
    C["Chargers (OCPP)"]
  end
  subgraph CORE["System of record"]
    direction TB
    R["Registers"]
    T["Trips, events, duty"]
    M["Maintenance"]
    N["Energy, costs"]
  end
  subgraph USE["Decisions"]
    direction TB
    D["Dispatch, routing"]
    S["Safety coaching"]
    A["Compliance records"]
    P["Replace, right-size"]
  end
  K --> T
  E --> T
  V --> T
  V --> M
  F --> N
  C --> N
  T --> N
  R --> D
  T --> D
  T --> S
  T --> A
  M --> A
  T --> P
  M --> P
  N --> P
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  classDef neutral fill:#f1f4f6,stroke:#5b7a87,color:#122029
  class R,M,T,N core
  class V,E,K,F,C,D,S,P,A neutral

Trust boundaries. Telematics and camera data are personal data about drivers: keep live location to dispatch and safety, give finance aggregates, and treat API tokens as credentials. Chargers and the CSMS are internet-connected controls that can stop tomorrow’s work: keep chargers on their own network segment, off the corporate LAN; give CSMS users named accounts; and require an OCPP security profile that uses TLS. Profile 3 adds client-side certificates for the chargers,^[27] and a 2023 TU Wien study for Austria’s electricity association concluded that “OCPP 2.0.1, or OCPP-J 1.6 (JSON/Websocket) with security extensions when using OCPP security profile 3, are suitable for secure charging station control”.^[28]

Happy path: check, drive a planned route within working-time limits, refuel or charge, return, data lands. The exception that matters most is a defect:

sequenceDiagram
  autonumber
  participant A as App
  participant D as Driver
  participant X as Dispatch
  participant P as Planner
  participant W as Workshop
  D->>A: Brake hose cracked
  A->>P: Defect, safety-critical
  P->>X: Van off road
  X->>D: Use spare van
  P->>W: Priority work order
  W->>P: Repaired, certified
  P->>X: Van back in service

What the regulations ask of this flow, in two jurisdictions:

• US (interstate commercial motor vehicles). Drivers report defects in specified parts (brakes, steering, lights, tyres, wheels, coupling devices and others) at the end of each day’s work; no report is needed if none is found. Before the vehicle runs again, the carrier repairs safety-affecting defects and certifies the repair, keeping both for three months.^[29]
• Great Britain (HGVs and PSVs). The DVSA guide states that “The driver is always legally responsible for the condition of the vehicle while in use”, requires at least one walkaround check every day, and expects drivers’ defect reports to be kept for at least 15 months.^[30]

Other exceptions to design: a working-time limit reached mid-route (re-plan), a collision (preserve data), and a van that did not charge overnight (worked runbook).

Where fleet decisions are hardest and costliest: what a vehicle really costs, how many you need, how to maintain them, and when to replace them (including with an electric vehicle).

Total cost of ownership is “a financial estimate intended to help buyers and owners determine the direct and indirect costs of a product or service”; for a vehicle it includes depreciation, fuel, insurance, financing, repairs, fees and taxes, maintenance, and opportunity and downtime costs.^[22] TCO on its own has a flaw: two options kept for different periods cannot be compared by their totals. The fix is the equivalent annual cost (EAC), described as “the cost per year of owning and operating an asset over its entire lifespan”: take the net present value of all costs (net of resale) and divide it by the annuity factor for the keep period.^[23]

NPV(n)  = price + sum over years t = 1..n of  running_cost(t) / (1 + r)^t
                - resale(n) / (1 + r)^n
EAC(n)  = NPV(n) / A(n, r),   where A(n, r) = (1 - (1 + r)^-n) / r
Economic life = the n with the lowest EAC(n)

As the keep period grows, capital cost per year falls and running cost per year rises (maintenance, repairs and downtime grow with age). The sum is U-shaped; its bottom is the economic life. The chart shows Lab 1’s two illustrative vans.

%%{init: {"xyChart": {"width": 440, "height": 360}, "themeVariables": {"xyChart": {"plotColorPalette": "#c9800f, #16756f"}}}}%%
xychart-beta
  title "EAC per year (orange: diesel, teal: electric)"
  x-axis "Keep period (years)" [3, 4, 5, 6, 7, 8, 9, 10]
  y-axis "EAC per year" 12000 --> 16500
  line [13848, 13313, 13029, 12903, 12896, 12993, 13189, 13484]
  line [16165, 14803, 13957, 13402, 13041, 12828, 12737, 12758]

Three lessons that hold beyond these numbers: the bottom is flat (6 to 8 years differ by under 1% for the diesel; three years late costs a lot); high-capital, low-running-cost vehicles need longer keep periods, so a blanket five-year policy penalises EVs; and downtime belongs in the model, or unreliable old vehicles are kept too long.

The replacement rule for a vehicle you already own asks what one more year costs: the resale value you give up (resale at n − 1 grown by a year’s interest, minus resale at n) plus year n’s running cost. Replace when that exceeds the best replacement’s EAC at its own economic life. With Lab 1’s inputs, keeping the diesel van through year 7 costs about 12,848 and through year 8 about 13,858, against 12,737 a year for a new electric van charged at the depot: replacing at the end of year 6 or 7 is close to a draw, and keeping it into year 8 costs about 1,100 more than replacing. (Illustrative; same inputs plus the marginal-cost formula.)

Measure utilisation by days for pool and service vehicles, hours for all-day vehicles, distance where cost follows kilometres; each misleads alone (30 km every day is high on days, low on distance).

Right-sizing asks whether each vehicle should exist and is the right type. US DOE guidance reviews mission and drive cycle, utilisation, class fit and age, solicits driver input through surveys or meetings, and evaluates each vehicle’s business case for reassigning, replacing, eliminating or using a more fuel-efficient vehicle. It lists alternatives to owning a vehicle, such as motor pools, sharing across offices, stipends, public transport and short-term rental, and notes that “a 10% reduction in vehicle weight can improve fuel economy by 6%–7%”.^[10] US federal agencies must implement a structured vehicle allocation methodology, and GSA recommends repeating the study at least once every five years.^[16]

Sizing the fleet from data asks how many vehicles are in use at the same time on a busy day. Lab 2’s method:

1. From trip data, count the vehicles in use in each 15-minute slot of each day.
2. Take each day’s peak.
3. Choose a planning percentile of those daily peaks (the 90th, say: nine days in ten are covered).
4. Add a spare allowance for maintenance and breakdowns, sized to your measured off-road rate.
5. Cover the remaining peak days with rental, pool swaps or schedule changes, and price them.

In Lab 2’s data, eight of the ten vans (seven at the 90th percentile plus 10% spares) would cover nine days in ten. Low-use vans are candidates for review, not disposal: they may carry special equipment or already be the spares.

Time-based maintenance is criticised because equipment is “maintained whether it needs it or not”; predictive maintenance instead uses “the current condition of equipment” to time the work.^[32] That is not a reason to drop time-based inspections. Safety inspections are a legal and evidential baseline in many places: US carriers “must systematically inspect, repair, and maintain” every vehicle under their control and keep records that include “A means to indicate the nature and due date of the various inspection and maintenance operations”.^[33] The British DVSA guide says inspection frequency “would normally range between 4 to 13 weeks”, set by the deterioration expected for the vehicle’s age, load, terrain and use, and that inspections should follow a time-based pattern where practical, with mileage-based programmes “linked to time”.^[30]

The practical answer is layered: time-based safety inspections as the floor, condition-based work where data is good, and reactive repair only where failure is cheap and harmless.

Tyres are both a large cost and a safety item. In the US, front tyres on buses, trucks and truck tractors need at least 4/32 inch of tread and other tyres 2/32 inch;^[34] in Great Britain, since 1 February 2021 tyres over 10 years old may not be used on the front steered axles of HGVs, buses and coaches or on single wheels of minibuses.^[35] Record tread depth, pressure and date code at each inspection, and set replacement thresholds above the legal minimum.

flowchart TB
  Q1{"Failure safety-critical or legally inspected?"}
  Q1 -->|"yes"| PM["Time-based inspection as the floor"]
  Q1 -->|"no"| Q2{"Failure expensive or causes downtime?"}
  PM --> Q3{"Condition measurable and trusted?"}
  Q2 -->|"no"| RF["Run to failure, fix on report"]
  Q2 -->|"yes"| Q3
  Q3 -->|"yes"| CB["Add condition-based triggers"]
  Q3 -->|"no"| TB["Time or distance interval"]
  classDef decision fill:#fdf3e3,stroke:#c9800f,color:#122029
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  classDef risk fill:#fbeceb,stroke:#b4332b,color:#122029
  class Q1,Q2,Q3 decision
  class PM,CB,TB core
  class RF risk

Electrification changes the cost structure (more capital, less energy and maintenance), the operating model (charging constrains dispatch) and the risks (cold-weather range, grid lead times). It is a duty-cycle decision first.

What changes in the cost model. The US DOE notes “far fewer moving parts”, that “Brake wear is significantly reduced due to regenerative braking”, and common 8-year/100,000-mile battery warranties.^[36] Energy cost depends on where you charge (depot off-peak, public rapid, reimbursed home); Lab 1’s “Try this” shows how far a public price moves the break-even.

Range is a distribution, not a number. fueleconomy.gov reports that at 20 °F (−7 °C), with cabin heating, an EV’s fuel economy can drop roughly 39% and its range 41% in mixed driving, citing a 2019 AAA study; conventional cars lose about 15% in city driving at that temperature.^[37] Plan routes against winter range with a reserve, not against the brochure figure.

Charging is infrastructure. US DOE figures: Level 2 AC charging delivers 2.9 to 19.2 kW, about 25 miles (40 km) of range per hour; DC fast charging delivers up to 500 kW, roughly 100 to 200+ miles of range in 30 minutes, depending on the vehicle and state of charge.^[38] Depot vans with long dwell times usually manage on Level 2; trucks and double-shifted vehicles may need DC. DOE advises engaging utilities “early in the infrastructure planning process” and lists demand charges among operating costs.^[39] Managed (smart) charging spreads load to keep the site under its connection limit; OCPP supports smart charging with charge profiles and load balancing.^[26] Connectors matter for mixed fleets: SAE published J3400 (the North American Charging System) as a recommended practice in September 2024, after a technical information report in December 2023, while CCS remains common.^[20]

Policy is moving; date-stamp it. Example (as of 23 October 2026): a May 2025 settlement with 16 states required California’s CARB to propose repealing the high-priority and drayage parts of its Advanced Clean Fleets rule and to send an approved repeal to the Office of Administrative Law (OAL) by 31 August 2026.^[40] On 25 September 2025 CARB’s board voted to repeal the parts that apply to federal and private fleets, drayage included, and gave state and local government fleets more time: the 50% zero-emission purchase requirement moves back three years and the 100% requirement to 2030.^[41] Whether OAL approved the repeal could not be confirmed from an openable source (a trade report that it took effect on 10 September 2026 could not be fetched). Check current status before relying on any mandate or incentive.

flowchart TB
  A{"Winter daily distance within 70% of usable range?"}
  A -->|"no"| N1["Not yet: route redesign, PHEV or wait"]
  A -->|"yes"| B{"Parked at a site you control with dwell time to charge?"}
  B -->|"no"| C{"Reliable public or home charging at a known price?"}
  C -->|"no"| N2["Not yet: secure charging first"]
  C -->|"yes"| D
  B -->|"yes"| E{"Site power available in time?"}
  E -->|"no"| N3["Start utility request; pilot elsewhere"]
  E -->|"yes"| D{"EAC lower at your keep period and charging price?"}
  D -->|"no"| N4["Keep current powertrain; revisit next year"]
  D -->|"yes"| P["Pilot 5-10 vehicles, measure, then scale"]
  classDef decision fill:#fdf3e3,stroke:#c9800f,color:#122029
  classDef risk fill:#fbeceb,stroke:#b4332b,color:#122029
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  class A,B,C,D,E decision
  class N1,N2,N3,N4 risk
  class P core

Two playbooks deserve detail.

Fuel and energy reconciliation. Match each card transaction to the vehicle’s position at that time, and compare each vehicle’s litres (or kWh) per 100 km with its class median (rules in the reference implementation). US DOE estimates that US vehicles waste more than 6 billion gallons of fuel a year idling, about half of it by passenger vehicles, and says idling “can usually be reduced without compromising driver comfort or vehicle equipment operations”.^[42] DOE also cites analysis that aggressive driving lowers fuel economy by 15% to 30% at highway speeds and 10% to 40% in stop-and-go traffic.^[43] Smoother driving serves fuel and safety at once.

Driver safety programme. The UK HSE frames road risk as safe journey, safe driver or rider, and safe vehicle, and states that “Health and safety law applies to work activities on the road in the same way as it does on a fixed site”.^[11]

1. Safe driver: licence checks at hiring and at intervals; vehicle induction; fatigue and phone rules; coaching. US carriers must obtain each CMV driver’s motor vehicle record (MVR) at least once every 12 months and review it;^[44] in Great Britain, DVLA’s online service shows a driver’s entitlements, penalty points and disqualifications when the driver gives a check code, which must be used within 21 days and only once.^[45]
2. Safe vehicle: specification (including driver assistance), inspection and defect close-out. In the EU, Article 6 of Regulation (EU) 2019/2144 requires Intelligent Speed Assistance on motor vehicles of categories M and N, so vans and trucks as well as cars;^[46] the Commission’s summary gives July 2024 for all new cars and says “The driver is always in control and can easily override the ISA system”.^[19] Check the Regulation’s Annex II for the dates of heavier categories.
3. Safe journey: schedules drivable within limits; no calls to drivers at the wheel; weather plans.

On speed, the WHO states that “every 1% increase in mean speed produces a 4% increase in the fatal crash risk”.^[9] ISO 39001 offers a road-safety management-system frame.^[47] Collisions follow the checklist and feed the claims process in Cost, capacity and sustainability.

Two regimes as examples (as of October 2026; other countries and vehicle types differ):

Exceptions change the picture. In the US, drivers who stay within a 150 air-mile radius of their work reporting location, are released within 14 consecutive hours and have at least 10 hours off between shifts are exempt from the record-of-duty-status and ELD requirements; and a driver who meets adverse driving conditions may drive up to two additional hours to complete the run or reach a place offering safety.^[49] Paper logs may replace an ELD for drivers needing records on no more than 8 days in 30, some driveaway-towaway work and vehicles made before model year 2000.^[18] In the EU, tachographs are mandatory in new goods vehicles over 3.5 tonnes and new passenger vehicles for more than nine people, and from 1 July 2026 also in vehicles of 2.5 to 3.5 tonnes engaged in international hire-or-reward transport; smart tachographs also record the vehicle’s GNSS position at certain times.^[17]

Separate what is mandatory from what is merely useful.

Named tools are examples whose descriptions were checked on 2026-10-23, not recommendations.

Lock-in. Leased devices mean a reinstall campaign to switch, and history may export only in summary: test a full export and put export format and device ownership in the contract (with ProcurementOps).

Automation helps where decisions are frequent, rules clear and data reliable; it harms where it hides safety judgement or punishes drivers for sensor noise.

• Route optimisation is NP-hard (the vehicle routing problem generalises the travelling salesman problem).^[12] OR-Tools’ documentation says “Vehicle routing problems are inherently intractable” and that its solver “sometimes returns solutions that are good, but not optimal”.^[56] Heuristics (Lab 4) get much of the gain, and the objective matters as much as the algorithm: OR-Tools’ VRP example minimises the longest route, because minimising total distance alone can push all the work onto one vehicle.^[57]
• AI-detected driving events (phone use, close following) widen what can be coached and what is recorded about people; review each event before it reaches a score that affects pay or discipline.
• Predictive maintenance needs failure history most small fleets lack; start with fault-code alerts. Language models can summarise workshop notes, but not decide fitness to drive.
• Autonomous vehicles change who operates, not whether operations exist. Waymo describes “fleet response” agents who answer questions from the automated driver, which “is in control of the vehicle at all times” and “may come to a stop if it determines that is the safest course of action”; when needed, roadside assistance retrieves the vehicle manually.^[58] Remote assistance, retrieval, cleaning, charging and maintenance remain fleet operations.

Illustrative fleet: 120 vans, 15 rigid trucks, two depots, telematics with an API, an in-house workshop. Adapt intervals to your regulator and the manufacturer.

1. Maintenance schedule as data, reviewed and auditable, so software can compute due dates.

Example status: YAML syntax-checked by the validator; not executed. Intervals are illustrative; the file is data for your maintenance software, which applies the earliest trigger.

# pm-schedule.yaml: whichever trigger comes first raises the work order
version: 3
owner: maintenance-planner
reviewed: 2026-10-01
classes:
  van-3.5t:
    safety_inspection: {every_weeks: 10, every_km: null}
    service_a: {every_months: 12, every_km: 30000}
    brake_check: {every_weeks: 10}
  rigid-18t:
    safety_inspection: {every_weeks: 6}       # general haulage band, DVSA guide
    service_a: {every_months: 6, every_km: 40000}
    brake_performance_test: {at: [safety_inspection]}
overrides:                                    # applied on top of the class rules
  - when: {age_years_gte: 12}                 # older vehicles inspected more often
    set: {safety_inspection: {every_weeks: 6}}
tolerances:
  overdue_grace_days: 0                       # safety inspections: no grace
  service_grace_km: 1000
escalation:
  overdue_safety_inspection: vehicle_off_road
  overdue_service: notify_fleet_manager

2. Fuel and energy exception rules, applied to every transaction.

Example status: YAML syntax-checked. The when expressions are illustrative pseudo-code for a rule engine, not the syntax of a named product; thresholds are starting points to tune on your own data.

# fuel-exceptions.yaml
rules:
  - id: location-mismatch
    when: distance_card_site_to_vehicle_m > 500
    action: hold_for_review
  - id: over-capacity
    when: litres > tank_capacity_l * 1.05
    action: hold_for_review
  - id: rapid-refill
    when: km_since_last_fill < 100 and litres > 20
    action: review
  - id: wrong-product
    when: product not in vehicle.allowed_fuels
    action: hold_for_review
  - id: consumption-drift
    when: l_per_100km_30d > class_median * 1.25
    action: coach_or_inspect
review:
  owner: fleet-manager
  sla_hours: 72
  evidence: [card_transaction, vehicle_position, odometer]

3. Vehicle register columns. The minimum fields, as a CSV header:

unit_id,vin,registration,class,mission,cost_centre,depot,ownership,acquired_on,
acquisition_cost,odometer_km,engine_hours,telematics_device_id,planned_replacement,
status,disposed_on

4. Pulling trips from a telematics API. Traccar’s API (version 6.16.0 of its OpenAPI description) returns trips from GET /reports/trips for a time window given in ISO 8601 and at least one device or group; each trip has a distance in metres, speeds in knots and a duration in seconds, and the API accepts HTTP Basic or a bearer token.^[59] The script below writes trips in Lab 2’s format so the utilisation analysis can run on real data. Two details decide whether the numbers are right: Traccar returns times as ISO 8601 (its examples are UTC), so the script converts them to the fleet’s local time zone before cutting days and shift minutes; and Lab 2 counts working days, so the script numbers Monday-to-Friday dates only and reports how many there were.

Example status: executed on 2026-10-23 against a local mock server returning trips in the documented format, across the 25 October 2026 clock change (Python 3.8 with a test stand-in for zoneinfo; the script needs Python 3.9+). Not run against a real Traccar server. Use a read-only Traccar user limited to the group you analyse, and give its token an expiry date (POST /session/token accepts an expiration).^[59]

"""Pull trips from a Traccar server into the Lab 2 CSV format (Python 3.9+).

Usage: traccar_trips.py GROUP_ID FROM TO TIMEZONE > trips.csv
FROM and TO are ISO 8601 times such as 2026-10-05T00:00:00Z; TIMEZONE is the
fleet's IANA zone such as Europe/London. Reads TRACCAR_URL (for example
https://tracking.example.com/api) and TRACCAR_TOKEN from the environment. Use a
read-only Traccar user limited to the group you analyse: the token grants
everything that user can see.

Days are numbered by local working day (Monday-Friday) because Lab 2 counts
working days; weekend trips are skipped. Remove public holidays by hand. The
number of working days is printed to stderr: pass it to lab2_utilisation.py.
"""
import csv
import json
import os
import sys
import urllib.parse
import urllib.request
from datetime import datetime, timedelta
from zoneinfo import ZoneInfo

BASE = os.environ["TRACCAR_URL"].rstrip("/")
TOKEN = os.environ["TRACCAR_TOKEN"]
GROUP_ID, FROM, TO, TZ = sys.argv[1], sys.argv[2], sys.argv[3], ZoneInfo(sys.argv[4])


def get(path, **params):
    url = f"{BASE}{path}?{urllib.parse.urlencode(params, doseq=True)}"
    req = urllib.request.Request(url, headers={"Authorization": f"Bearer {TOKEN}",
                                               "Accept": "application/json"})
    with urllib.request.urlopen(req, timeout=60) as resp:
        return json.load(resp)


def local(ts):
    """Traccar times are ISO 8601 (UTC 'Z' in its examples); convert to fleet time."""
    return datetime.fromisoformat(ts.replace("Z", "+00:00")).astimezone(TZ)


first, last = local(FROM).date(), local(TO).date()
dates = [first + timedelta(days=i) for i in range((last - first).days + 1)]
workday = {d: n + 1 for n, d in enumerate(d for d in dates if d.weekday() < 5)}

trips = get("/reports/trips", groupId=GROUP_ID, **{"from": FROM, "to": TO})
w = csv.writer(sys.stdout)
w.writerow(["vehicle", "day", "start_min", "end_min", "km", "idle_min"])
skipped = 0
for t in trips:
    s, e = local(t["startTime"]), local(t["endTime"])
    if s.date() not in workday or e <= s:
        skipped += 1
        continue
    same_day = e.date() == s.date()
    end_min = e.hour * 60 + e.minute if same_day else 24 * 60  # cut trips at midnight
    w.writerow([t["deviceName"], workday[s.date()], s.hour * 60 + s.minute, end_min,
                round(t["distance"] / 1000, 1),
                0])  # trips carry no idle time; see /reports/summary engineHours
print(f"working days: {len(workday)}; trips skipped (weekend or zero length): {skipped}",
      file=sys.stderr)

Run it with the URL and token in the environment, never on the command line where they end up in shell history:

export TRACCAR_URL="https://tracking.example.com/api"
read -rs TRACCAR_TOKEN && export TRACCAR_TOKEN
python3 traccar_trips.py 7 2026-10-05T00:00:00Z 2026-10-30T23:59:59Z \
  Europe/London > trips.csv
# stderr reports "working days: 20; ..." - pass that number to Lab 2
python3 lab2_utilisation.py trips.csv 20

The trips report has no idle field, so the script writes zero; Traccar’s summary report adds engine hours for the period, and engine hours minus trip time gives a rough idle estimate.^[59] Trips that cross midnight are cut at 24:00, and public holidays must be removed by hand; Lab 2 stops with an error if a day number exceeds the working days you pass, so days% can never exceed 100.

Four labs, beginner to advanced, in Python’s standard library; no accounts, no cost.

Example status: all four labs verified by execution on 2026-10-23 (Python 3.8.10, Linux). All data is illustrative.

Objective. Compare a diesel van and an electric van on equivalent annual cost for keep periods of 1 to 10 years, find each one’s economic life, and find the annual distance at which the electric van breaks even. Prerequisites. Python 3.8 or later. Cost and risk. None.

Save the inputs as vans.json (one currency; energy price per litre or per kWh; consumption per 100 km).

{
  "discount_rate": 0.06,
  "km_per_year": 25000,
  "vehicles": {
    "diesel_van": {
      "price": 38000, "charger": 0,
      "use_per_100km": 8.5, "energy_price": 1.60,
      "maint_year1": 900, "maint_growth": 0.25,
      "downtime_days_year1": 2, "downtime_cost_per_day": 250,
      "insurance": 1400,
      "depr_year1": 0.25, "depr_after": 0.12
    },
    "electric_van": {
      "price": 52000, "charger": 3000,
      "use_per_100km": 24.0, "energy_price": 0.22,
      "maint_year1": 550, "maint_growth": 0.25,
      "downtime_days_year1": 2, "downtime_cost_per_day": 250,
      "insurance": 1600,
      "depr_year1": 0.30, "depr_after": 0.13
    }
  }
}

Save this as lab1_tco.py:

"""Lab 1: total cost of ownership and equivalent annual cost (Python 3.8+).

Compares a diesel van and a battery-electric van over keep periods of 1-10 years
and finds each one's economic life (the keep period with the lowest EAC).
All inputs are illustrative; replace them with your own quotes and data.
"""
import json
import sys

def eac(npv_cost, rate, years):
    """Equivalent annual cost: NPV of costs divided by the annuity factor."""
    annuity = (1 - (1 + rate) ** -years) / rate
    return npv_cost / annuity

def npv_of_keeping(v, years, rate, km_per_year):
    """NPV of buying now, running for `years`, selling at the end."""
    cost = v["price"] + v.get("charger", 0)
    for t in range(1, years + 1):
        energy = km_per_year / 100 * v["use_per_100km"] * v["energy_price"]
        maint = v["maint_year1"] * (1 + v["maint_growth"]) ** (t - 1)
        days_down = v["downtime_days_year1"] * (1 + v["maint_growth"]) ** (t - 1)
        down = days_down * v["downtime_cost_per_day"]
        yearly = energy + maint + down + v["insurance"]
        cost += yearly / (1 + rate) ** t
    resale = v["price"] * (1 - v["depr_year1"]) * (1 - v["depr_after"]) ** (years - 1)
    cost -= resale / (1 + rate) ** years
    return cost

def table(vehicles, rate, km):
    rows = {}
    for name, v in vehicles.items():
        rows[name] = [eac(npv_of_keeping(v, n, rate, km), rate, n) for n in range(1, 11)]
    return rows

def main():
    cfg = json.load(open(sys.argv[1] if len(sys.argv) > 1 else "vans.json"))
    rate, km = cfg["discount_rate"], cfg["km_per_year"]
    rows = table(cfg["vehicles"], rate, km)
    names = list(rows)
    print("keep_years " + " ".join(f"{n:>12}" for n in names))
    for i in range(10):
        print(f"{i + 1:>10} " + " ".join(f"{rows[n][i]:>12,.0f}" for n in names))
    for n in names:
        best = min(range(10), key=lambda i: rows[n][i])
        print(f"economic life {n}: {best + 1} years, EAC {rows[n][best]:,.0f} per year")
    # Break-even distance at a 7-year keep: smallest km/yr where EV EAC <= diesel EAC.
    ice, ev = names[0], names[1]
    for k in range(5000, 80001, 500):
        r = table(cfg["vehicles"], rate, k)
        if r[ev][6] <= r[ice][6]:
            print(f"break-even at 7-year keep: about {k:,} km per year")
            break
    else:
        print("no break-even below 80,000 km per year")

if __name__ == "__main__":
    main()

Steps: mkdir fleet-labs && cd fleet-labs, save both files, then run python3 lab1_tco.py vans.json.

Expected output.

keep_years   diesel_van electric_van
         1       17,980       25,870
         2       14,922       18,724
         3       13,848       16,165
         4       13,313       14,803
         5       13,029       13,957
         6       12,903       13,402
         7       12,896       13,041
         8       12,993       12,828
         9       13,189       12,737
        10       13,484       12,758
economic life diesel_van: 7 years, EAC 12,896 per year
economic life electric_van: 9 years, EAC 12,737 per year
break-even at 7-year keep: about 27,000 km per year

Validation. By hand, the diesel van kept one year: running cost = 25,000 km × 8.5 L/100 km × 1.60 = 3,400 fuel + 900 maintenance + 500 downtime + 1,400 insurance = 6,200; resale = 38,000 × 0.75 = 28,500. NPV = 38,000 + (6,200 − 28,500) / 1.06 = 16,962; the one-year annuity factor is 1/1.06, so EAC = 16,962 × 1.06 = 17,980, which matches the first row.

Try this. Set the electric van’s energy_price to 0.45 (a public rapid-charging price) and rerun: its best EAC rises to 14,117 and the 7-year break-even moves to about 79,500 km a year. Charging location can matter more than the vehicle. Changing km_per_year to 15,000 or 40,000 does not move the economic lives: the model ties maintenance to age, not distance, a limitation to fix before real use.

Troubleshooting. KeyError means a field is missing from vans.json; every vehicle needs every field. A ZeroDivisionError means a discount rate of 0: use a small positive rate. Teardown. Delete the fleet-labs folder when you finish all labs.

Objective. From a trip log, compute each van’s utilisation by days, distance, engine hours and idle share, then estimate the fleet size needed to cover nine days in ten. Prerequisites. Python 3.8+. Cost and risk. None.

Save this generator as make_trips.py (it writes a seeded, illustrative trip log for 10 vans over 20 working days):

"""Lab 2 data: write an illustrative trip log for 10 vans over 20 working days."""
import csv
import random
import sys

random.seed(42)
USE_PROB = [0.95, 0.9, 0.9, 0.85, 0.8, 0.75, 0.6, 0.35, 0.25, 0.15]
with open(sys.argv[1] if len(sys.argv) > 1 else "trips.csv", "w", newline="") as f:
    w = csv.writer(f)
    w.writerow(["vehicle", "day", "start_min", "end_min", "km", "idle_min"])
    for day in range(1, 21):
        for i, p in enumerate(USE_PROB):
            if random.random() > p:
                continue
            t = random.randint(7 * 60, 9 * 60)          # first departure 07:00-09:00
            for _ in range(random.randint(1, 4)):
                dur = random.randint(45, 180)
                end = min(t + dur, 18 * 60)
                if end - t < 20:
                    break
                km = round((end - t) / 60 * random.uniform(18, 35), 1)
                idle = round((end - t) * random.uniform(0.05, 0.30))
                w.writerow([f"V{i + 1:02d}", day, t, end, km, idle])
                t = end + random.randint(20, 90)        # time on site between trips
                if t >= 17 * 60:
                    break

Save the analysis as lab2_utilisation.py:

"""Lab 2: utilisation, idling and a right-sizing estimate (Python 3.8+)."""
import csv
import math
import sys
from collections import defaultdict

DAYS = int(sys.argv[2]) if len(sys.argv) > 2 else 20   # working days in the sample
SHIFT_MIN = 11 * 60       # 07:00-18:00 shift window
MIN_DAYS_PCT = 50         # policy threshold: review vans used on fewer days than this
SPARE_PCT = 10            # policy: spare vehicles on top of the planning peak

trips = list(csv.DictReader(open(sys.argv[1] if len(sys.argv) > 1 else "trips.csv")))
by_v = defaultdict(lambda: {"days": set(), "km": 0.0, "eng": 0, "idle": 0})
busy = defaultdict(set)   # (day, 15-minute slot) -> vehicles in use
for r in trips:
    v, d, s, e = r["vehicle"], int(r["day"]), int(r["start_min"]), int(r["end_min"])
    if not 1 <= d <= DAYS or not 0 <= s < e <= 24 * 60:
        sys.exit(f"bad row {r}: day must be 1..{DAYS} and start < end within the day")
    x = by_v[v]
    x["days"].add(d)
    x["km"] += float(r["km"])
    x["eng"] += e - s
    x["idle"] += int(r["idle_min"])
    for slot in range(s // 15, math.ceil(e / 15)):
        busy[(d, slot)].add(v)

print(f"{'van':4} {'days%':>6} {'km':>7} {'eng_h':>6} "
      f"{'use%shift':>11} {'idle%':>6}  flag")
for v in sorted(by_v):
    x = by_v[v]
    days_pct = 100 * len(x["days"]) / DAYS
    shift_pct = 100 * x["eng"] / (DAYS * SHIFT_MIN)
    idle_pct = 100 * x["idle"] / x["eng"]
    flag = "review" if days_pct < MIN_DAYS_PCT else ""
    print(f"{v:4} {days_pct:6.0f} {x['km']:7.0f} {x['eng'] / 60:6.1f} "
          f"{shift_pct:11.0f} {idle_pct:6.1f}  {flag}")

peaks = sorted(max((len(vs) for (d, _), vs in busy.items() if d == day), default=0)
               for day in range(1, DAYS + 1))
p90 = peaks[math.ceil(0.9 * len(peaks)) - 1]
need = math.ceil(p90 * (1 + SPARE_PCT / 100))
mid = peaks[len(peaks) // 2]
print(f"daily peak vans in use: min {peaks[0]}, median {mid}, max {peaks[-1]}")
print(f"90th percentile of daily peaks: {p90}; with {SPARE_PCT}% spares: {need} vans "
      f"(fleet today: {len(by_v)})")

Run python3 make_trips.py trips.csv and then python3 lab2_utilisation.py trips.csv (an optional second argument gives the number of working days; the default is 20). use%shift is engine-on trip time, idling included, as a share of the 07:00-18:00 shift window.

Expected output (the generator writes 326 trips plus a header):

van   days%      km  eng_h   use%shift  idle%  flag
V01      95    2275   84.5          38   19.6  
V02      80    2035   75.5          34   17.2  
V03      85    1804   69.6          32   16.3  
V04      95    2372   87.0          40   19.6  
V05      65    1447   51.7          24   19.6  
V06      85    2188   82.8          38   16.9  
V07      75    2034   73.4          33   18.3  
V08      30     718   26.5          12   19.9  review
V09      25     720   27.2          12   18.4  review
V10       5     185    7.0           3   13.0  review
daily peak vans in use: min 4, median 7, max 8
90th percentile of daily peaks: 7; with 10% spares: 8 vans (fleet today: 10)

Validation. V10 has a 15% daily chance of use, so one day (5%) is within random spread; the 90th percentile (7) lies between the median (7) and the maximum (8), as it must.

Interpretation. Ask what the three flagged vans carry and who uses them, then price eight vans plus peak rental.

Troubleshooting. bad row means a day number above the working days you passed, or a trip whose end is not after its start; fix the data rather than the check. With real data from the Traccar script, idle% is zero (no idle data). Teardown. Delete trips.csv.

Objective. Flag breaches of the 14-hour window, 11-hour driving limit, 30-minute break and 70-hour/8-day limit in a duty log (49 CFR 395.3).^[48] Prerequisites. Python 3.8+. Cost and risk. None; a teaching tool that ignores sleeper-berth splits, adverse conditions, short-haul and the 34-hour restart.

Save the illustrative log as duty_log.csv (statuses: OFF off duty, SB sleeper berth, D driving, ON on duty not driving; the last row only marks the end of the log):

driver,time,status
D-017,2026-10-05T00:00,OFF
D-017,2026-10-05T06:00,ON
D-017,2026-10-05T06:30,D
D-017,2026-10-05T10:30,ON
D-017,2026-10-05T11:00,D
D-017,2026-10-05T15:00,ON
D-017,2026-10-05T15:30,D
D-017,2026-10-05T17:30,OFF
D-017,2026-10-06T05:00,ON
D-017,2026-10-06T05:15,D
D-017,2026-10-06T13:45,OFF
D-017,2026-10-06T14:15,D
D-017,2026-10-06T16:00,OFF
D-017,2026-10-07T04:00,ON
D-017,2026-10-07T04:30,D
D-017,2026-10-07T09:00,OFF
D-017,2026-10-07T09:30,D
D-017,2026-10-07T12:00,ON
D-017,2026-10-07T15:00,D
D-017,2026-10-07T18:00,ON
D-017,2026-10-07T18:30,D
D-017,2026-10-07T19:30,OFF
D-017,2026-10-08T12:00,OFF

Save the checker as lab3_hos.py:

"""Lab 3: a simplified US property-carrying hours-of-service checker (Python 3.8+).

Teaching tool only. It checks the 10-hour reset, the 14-hour window, the 11-hour
driving limit, the 30-minute break after 8 hours of driving and the 70-hour/8-day
limit. It ignores sleeper-berth splits, adverse driving conditions, the short-haul
exception and the 34-hour restart, so it must never be used for compliance.
"""
import csv
import sys
from datetime import datetime, timedelta

OFF = {"OFF", "SB"}

def minutes(events):
    """Expand status-change events into one status per minute."""
    out = []
    for (t0, s0), (t1, _) in zip(events, events[1:]):
        n = int((t1 - t0).total_seconds() // 60)
        out += [(t0 + timedelta(minutes=i), s0) for i in range(n)]
    return out

def check(timeline):
    found, off_run, shift_start = [], 10 * 60, None
    drive_shift = drive_since_break = nondrive_run = 0
    history = []                      # 1 if on duty (D or ON) for each minute, for 70/8
    for t, s in timeline:
        on = s not in OFF
        history.append(1 if on else 0)
        if not on:
            off_run += 1
            nondrive_run += 1
            if nondrive_run >= 30:
                drive_since_break = 0
            continue
        if off_run >= 10 * 60:        # on duty after 10 hours off: a new shift
            shift_start, drive_shift, drive_since_break = t, 0, 0
        off_run = 0
        if s != "D":
            nondrive_run += 1
            if nondrive_run >= 30:
                drive_since_break = 0
            continue
        nondrive_run = 0
        drive_shift += 1
        drive_since_break += 1
        rules = [
            ("14-hour window",
             shift_start is not None and t >= shift_start + timedelta(hours=14)),
            ("11-hour driving limit", drive_shift > 11 * 60),
            ("30-minute break required", drive_since_break > 8 * 60),
            ("70 hours in 8 days", sum(history[-8 * 1440:]) > 70 * 60),
        ]
        for name, broken in rules:
            new = (not found or found[-1][1] != name
                   or t - found[-1][2] > timedelta(minutes=1))
            if broken and new:
                found.append([t, name, t])
            elif broken:
                found[-1][2] = t
    return found

def main():
    path = sys.argv[1] if len(sys.argv) > 1 else "duty_log.csv"
    rows = list(csv.DictReader(open(path)))
    events = [(datetime.fromisoformat(r["time"]), r["status"]) for r in rows]
    for start, rule, last in check(minutes(events)):
        mins = int((last - start).total_seconds() // 60) + 1
        print(f"{start:%Y-%m-%d %H:%M}  {rule:26} {mins:4d} min of driving")
    print("done")

if __name__ == "__main__":
    main()

Run python3 lab3_hos.py duty_log.csv.

Expected output.

2026-10-06 13:15  30-minute break required     30 min of driving
2026-10-07 18:30  14-hour window               60 min of driving
done

Validation. 6 October: driving 05:15–13:45 is 8.5 hours without a 30-minute break, so the last 30 minutes are flagged. 7 October: on duty at 04:00, so the window closes at 18:00 and the hour from 18:30 is flagged. 5 October: the 30-minute on-duty periods count as breaks under the rule,^[48] and 10 hours of driving is inside the limit.

Troubleshooting. ValueError: Invalid isoformat string means a timestamp is not in YYYY-MM-DDTHH:MM form. The checker treats the time before the first row as a full 10-hour rest, so a log that starts mid-shift opens its first shift at the first on-duty row and can miss a 14-hour breach; start logs at the beginning of a rest period. Teardown. Delete the files.

Objective. Route three 32-unit vans from one depot to 12 stops and compare a naive plan with a heuristic one. Prerequisites. Python 3.8+. Cost and risk. None.

Save the stops as stops.csv (coordinates in km from the depot, demand in units):

stop,x_km,y_km,demand
DEPOT,0,0,0
S01,4,9,8
S02,-6,7,6
S03,9,-3,10
S04,-8,-5,7
S05,2,14,5
S06,12,6,9
S07,-11,2,4
S08,5,-10,6
S09,-3,-12,8
S10,14,-8,5
S11,-13,11,7
S12,7,4,6

Save the planner as lab4_routes.py:

"""Lab 4: capacitated routing with Clarke-Wright savings and 2-opt (Python 3.8+).

Distances are straight lines times a road factor; real routing needs a road network
and usually a solver such as OR-Tools. The point is to compare a naive plan with a
heuristic one and to see why routing is a search problem, not a formula.
"""
import csv
import math
import sys

CAPACITY = 32       # units per van
ROAD_FACTOR = 1.3   # straight-line to road distance, illustrative

rows = list(csv.DictReader(open(sys.argv[1] if len(sys.argv) > 1 else "stops.csv")))
pts = {r["stop"]: (float(r["x_km"]), float(r["y_km"])) for r in rows}
dem = {r["stop"]: int(r["demand"]) for r in rows}
stops = [r["stop"] for r in rows if r["stop"] != "DEPOT"]

def d(a, b):
    (x1, y1), (x2, y2) = pts[a], pts[b]
    return ROAD_FACTOR * math.hypot(x1 - x2, y1 - y2)

def length(route):
    path = ["DEPOT"] + route + ["DEPOT"]
    return sum(d(a, b) for a, b in zip(path, path[1:]))

def naive():
    """Fill vans in stop-number order: what a spreadsheet plan often does."""
    routes, cur, load = [], [], 0
    for s in stops:
        if load + dem[s] > CAPACITY:
            routes.append(cur)
            cur, load = [], 0
        cur.append(s)
        load += dem[s]
    return routes + [cur]

def savings():
    routes = {s: [s] for s in stops}            # start: one route per stop
    owner = {s: s for s in stops}
    pairs = sorted(((d("DEPOT", i) + d("DEPOT", j) - d(i, j), i, j)
                    for i in stops for j in stops if i < j), reverse=True)
    for sav, i, j in pairs:
        ri, rj = owner[i], owner[j]
        if ri == rj or sav <= 0:
            continue
        a, b = routes[ri], routes[rj]
        if sum(dem[s] for s in a + b) > CAPACITY:
            continue
        # i and j must be route ends; orient so that i ends a and j starts b.
        if a[-1] != i:
            a = a[::-1]
        if b[0] != j:
            b = b[::-1]
        if a[-1] != i or b[0] != j:
            continue
        routes[ri] = a + b
        del routes[rj]
        for s in routes[ri]:
            owner[s] = ri
    return list(routes.values())

def two_opt(route):
    best, improved = route[:], True
    while improved:
        improved = False
        for i in range(len(best) - 1):
            for k in range(i + 1, len(best)):
                cand = best[:i] + best[i:k + 1][::-1] + best[k + 1:]
                if length(cand) < length(best) - 1e-9:
                    best, improved = cand, True
    return best

def show(name, routes):
    print(name)
    for r in routes:
        load = sum(dem[s] for s in r)
        print(f"  load {load:2d}  {length(r):6.1f} km  " + " > ".join(r))
    print(f"  total {sum(length(r) for r in routes):.1f} km with {len(routes)} vans")

show("naive (stop order)", naive())
show("savings + 2-opt", [two_opt(r) for r in savings()])

Run python3 lab4_routes.py stops.csv.

Expected output.

naive (stop order)
  load 31    84.0 km  S01 > S02 > S03 > S04
  load 32   118.2 km  S05 > S06 > S07 > S08 > S09
  load 18   101.9 km  S10 > S11 > S12
  total 304.1 km with 3 vans
savings + 2-opt
  load 28    53.9 km  S01 > S05 > S06 > S12
  load 24    56.6 km  S04 > S07 > S11 > S02
  load 29    60.3 km  S03 > S10 > S08 > S09
  total 170.9 km with 3 vans

Validation. Every stop appears once per plan; no load exceeds 32. Demand is 81 units, so at least three vans are needed (81 ÷ 32 = 2.53, rounded up). The heuristic plan is 133.2 km (44%) shorter because the naive one criss-crosses the area.

Extend it. Add time windows and the savings method needs repair steps; a solver such as OR-Tools, with capacities, time windows and dropped visits, earns its place there.^[56] Troubleshooting. If a route exceeds capacity, check that CAPACITY is at least the largest single demand. Teardown. Delete the fleet-labs folder.

The first three are illustrative scenarios; the fourth is a real, sourced operating model.

Municipal right-sizing, step by step. (1) Put every vehicle in the register with its department and mission. (2) Pull twelve months of trips so that seasonal work shows. (3) Run Lab 2’s peak-percentile method per vehicle class, not per department. (4) Ask drivers and supervisors about each low-use vehicle before deciding. (5) Move survivors into a bookable pool, price short-term rental for peaks, and dispose of the rest. Owner: the fleet manager, with each department head signing off its list. Pitfall: snow ploughs and emergency reserves look idle by design; tag them as mission-reserved before the study.

Last-mile electrification, step by step. (1) Screen each route’s winter distance against usable range with a reserve. (2) Price energy at the depot tariff and model EAC (Lab 1). (3) Ask the utility for the connection capacity the overnight load needs; the answer sets the timetable. (4) Configure managed charging so that the site stays under its limit. (5) Pilot with the routes that have the most margin, through one winter. Owner: the fleet manager, with dispatch and facilities. Pitfall: a summer-only pilot proves nothing about January.

Other pitfalls: cameras before policy and coaching; routing that optimises kilometres while the business measures jobs done.

Every fleet metric needs a denominator that reflects exposure. Formulas are this page’s definitions unless sourced; write yours down, because the same name means different things in different fleets.

Leading and lagging. Collisions and prohibitions lag; PM compliance, defect close-out, hours breaches and reviewed safety events lead, and you can act on them this week.

Worked calculation (illustrative fleet of 135 vehicles, one year).

• Cost per km = (1,215,000 + 620,000 + 310,000 + 205,000 + 50,000) ÷ 3,750,000 = 2,400,000 ÷ 3,750,000 = 0.64 per km.
• Collision rate = 9 ÷ 3.75 = 2.4 per million km. Depot B: 2 ÷ 0.8 = 2.5 per million km, which looks the same as the fleet. But with only two events, the exact 95% Poisson interval for the count is 0.24 to 7.2, so depot B’s rate could be anywhere from about 0.3 to 9.0 per million km. A league table of depots on one year’s collisions ranks noise.
• PM compliance = 389 ÷ 412 = 94.4%; list the 23 late ones and check for safety inspections.
• Availability = (2,700 − 94) ÷ 2,700 = 96.5%: the 3.5% off road sizes the spare allowance.

Safety and compliance obligations are covered under the practices above. Retention is part of compliance: US carriers keep inspection reports three months, maintenance records one year (plus six months after disposal) and duty-status records six months;^[29][33][18] the DVSA guide asks for 15 months.^[30] Deleting too early fails as surely as keeping everything forever.

Privacy. Telematics and cameras make the vehicle a monitored workplace. The UK Information Commissioner’s Office is specific (as of October 2026; it says this guidance is under review after the Data (Use and Access) Act):^[51]

• “Telematics data which records the activities of drivers is personal information and is subject to data protection law.”
• On vehicles also used privately, “you will rarely be able to justify monitoring during private use”; a company car can have tracking the driver can disable outside work.
• “You must inform workers and passengers of any vehicle monitoring.”
• Where tachographs are legally required, an employer can rely on the lawful basis of legal obligation.
• Driver monitoring, “such as monitoring driver behaviour and driving style, or using cameras or audio”, is harder to justify, and the ICO says you must carry out a DPIA for it, as for any tool that “uses analytics to make inferences, predictions, or decisions about drivers”. Tracking work vehicles is not on that list, but it still needs a lawful basis, transparency and proportionality; this page recommends a DPIA for it as well.
• Inward-facing cameras capture driver and passengers; audio is higher risk, so switch it off by default and trigger it only in exceptional circumstances.

In the EU, GDPR Article 35 requires a DPIA where processing “is likely to result in a high risk”, and each supervisory authority publishes a list of operations that need one:^[61] check your national list rather than the ICO’s. Elsewhere the law differs, but a defensible camera default recurs: outward-facing, event-triggered clips, audio off, driver-facing only for a documented risk, short retention unless linked to an incident, access limited to safety staff.

Security. Telematics accounts can locate every vehicle and sometimes immobilise it: named least-privilege accounts, MFA for administrators, read-only analytics users with expiring tokens, prompt removal of leavers. Treat the CSMS the same way (see trust boundaries); device security belongs to IoTOps.

Cost drivers. Capital (depreciation or lease), energy, maintenance, insurance and downtime are the largest lines; NAFA’s certification material calls fuel “the second largest fleet vehicle expense”.^[3] The levers, roughly in order of size: fewer vehicles, the right class and replacement age, energy (consumption, idling, tariff), then maintenance efficiency. Whether a lease sits on the balance sheet is an accounting question: involve finance before choosing lease over buy.

Claims and insurance. NAFA’s risk-management discipline covers insurance, subrogation, training, safety and crash management, and notes that not every loss can be insured.^[3] In practice: report every collision the same day, recover costs from at-fault third parties (subrogation), record repair cost and downtime per claim, and feed causes into coaching and vehicle specification. How much loss to retain (a higher excess) is a finance decision that the fleet’s collision rates should inform.

Chargeback and funding. A fleet can be funded from a central budget, or charge user departments a monthly rate that also builds a replacement fund. For such a fund, Government Fleet’s guidance advises building salvage values into the rate, returning sale proceeds to the fund, using the vehicle’s actual cost, and keeping the rate for the vehicle’s life instead of resetting it each budget cycle.^[62] A full-cost rate gives departments a reason to return under-used vehicles; a centrally funded fleet gives them none (this page’s view).

Capacity is vehicle-hours in the right place: drivers, vehicles, workshop bays and, for EVs, charging power after 18:00.

Sustainability. Fuel burned in owned or controlled vehicles is a direct (Scope 1) emission; purchased electricity is indirect (Scope 2).^[60] For electricity the organisation buys, for example at its depots, electrification moves emissions from Scope 1 to Scope 2, resized by the grid mix; classify reimbursed home and public charging by your inventory’s boundary rules. US carriers and shippers can benchmark freight emissions through EPA’s SmartWay.^[52] The carbon method itself is in GreenOps.

• Record audits. Each quarter, trace one inspection and one defect for ten sampled vehicles from schedule to certification.
• External signals. Roadside results and regulator scores (such as FMCSA’s safety categories)^[24] are the regulator’s view of your fleet; treat every prohibition as an incident.
• Data-quality checks. Silent devices, odometer jumps, driverless trips, unmatched fuel: control failures, not just bad data.
• Driver feedback. Drivers know which vehicles are unreliable and which routes cannot be driven legally.
• Model back-testing. Yearly, compare cost-model forecasts (maintenance by age, resale) with actuals.

flowchart TB
  A["05:30 shortfall found: 11 vans low"] --> B["Triage: energy needed vs on board, per route"]
  B --> C{"Enough energy with route swaps?"}
  C -->|"yes"| D["Swap routes: long routes to full vans"]
  C -->|"no"| E["Rapid charge, spare vehicles, defer jobs"]
  D --> F["Diagnose: CSMS, chargers, site limit"]
  E --> F
  F --> G{"Cause found?"}
  G -->|"CSMS or network"| H["Restore link; check charge profiles"]
  G -->|"site load limit"| I["Rebalance profiles; check tariff windows"]
  G -->|"charger or vehicle fault"| J["Work order; tag units"]
  H --> K["Post-incident review and prevention"]
  I --> K
  J --> K
  classDef risk fill:#fbeceb,stroke:#b4332b,color:#122029
  classDef decision fill:#fdf3e3,stroke:#c9800f,color:#122029
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  class A risk
  class C,G decision
  class B,D,E,F,H,I,J,K core

1. Protect today’s work (05:30–06:15).

1. Export each van’s state of charge and each route’s distance; convert to energy with the class’s seasonal kWh per 100 km plus a reserve (this page suggests 20%).
2. Give the longest routes to the fullest vans; short routes often fit vans at 50–60%.
3. For the remaining gap: top up on the fastest chargers, plan a mid-route rapid charge, use spares or rentals, or defer low-priority jobs. Record the cost.

2. Diagnose (from 06:15).

1. Every charger or a subset? All points to the CSMS, network link or site supply; a subset to a circuit, charger or vehicles.
2. Did sessions start? None: check the CSMS link and expired credentials. Started but delivered little: check charge profiles.
3. Was the site limit binding? Smart charging caps total power under the connection limit through charge profiles;^[26] a new building load, profile change or tariff-window change cuts every van’s share.
4. Vehicle side? A scheduled-charging setting or charge-port fault shows on one van only.

3. Fix and prevent.

• Alert at 01:00 on projected shortfall (energy so far plus power × hours left, versus tomorrow’s need).
• Keep a fallback: rapid chargers on the routes and two spare vans charged first.
• Put charge-profile and tariff changes under change control, with a test night.
• Review: a new cause (fix it) or an accepted risk (record the decision and its cost)?

A deliberate fleet function is worth it when vehicle cost is material or vehicle failure stops the service; people drive for work regularly; regulations apply (commercial vehicles, passenger transport, dangerous goods); or vehicle decisions are made without data.

When not to build it, or not yet:

• A handful of cars. A policy, a lease with maintenance and licence checks usually suffice. Before buying telematics, compare its yearly cost (subscription × units, plus the time to use it) with the fuel, time or claims it must save; at small scale it often fails that test (this Atlas’s rule of thumb, not a measured threshold).
• Transport is better bought than run. If carriers can do the work at a known price, outsource (with SupplyChainOps) and keep oversight of safety requirements, service levels and data.
• Mostly grey fleet. Then the job is a grey-fleet policy and checks, perhaps with a pool, not a fleet department.
• No decision depends on it. Collecting telematics with no plan to act on it adds privacy risk without benefit.

Hidden responsibilities: records retention, data protection assessments, telematics change control, driver consultation, and a replacement budget that survives a bad year.

Going electric is in the EV decision tree; here, how to source vehicles and maintenance.

Acquisition model by vehicle group.

In-house versus outsourced maintenance. Illustrative weights; set yours before scoring each option 1 (poor) to 5 (good).

Arithmetic: in-house 0.30 × 4 + 0.25 × 3 + 0.20 × 4 + 0.15 × 2 + 0.10 × 2 = 3.25; outsourced 0.30 × 3 + 0.25 × 4 + 0.20 × 3 + 0.15 × 4 + 0.10 × 4 = 3.50. Moving 10 points of weight from flexibility to downtime flips it (3.45 against 3.40): when a decision is this sensitive, run a trial (outsource one depot for six months) and compare measured downtime and cost.

This maturity model is this Atlas’s proposal, not an industry standard; move up one level at a time.

flowchart TB
  L1["1 Reactive: spreadsheets, fix on failure"] --> L2["2 Recorded: register, PM schedule, defects"]
  L2 --> L3["3 Measured: KPIs per km, coaching"]
  L3 --> L4["4 Optimised: EAC, right-sizing, routing"]
  L4 --> L5["5 Adaptive: condition-based, energy-managed"]
  classDef neutral fill:#f1f4f6,stroke:#5b7a87,color:#122029
  classDef core fill:#e8f3f2,stroke:#16756f,color:#122029
  class L1,L2 neutral
  class L3,L4,L5 core

Most fleets gain more from levels 1 to 3 than from anything at level 5.

For a fleet manager starting at level 1 or 2.

Dependencies: finance (cost data), HR (driver records, consultation), procurement (contracts), IT or IoT (telematics). Quick wins: closing old defects, cancelling unused fuel cards, returning clearly redundant vehicles.

Commissioning a new vehicle

• Registered, insured, taxed; documents in the register
• Telematics installed, mapped to the unit, ignition and odometer verified on a test drive
• Upfit (racking, lighting, equipment) inspected and recorded
• PM schedule created from the class template
• Driver induction on the vehicle done and recorded

Operating readiness for electric vehicles

• Routes screened against winter range with a reserve
• Charging capacity confirmed for the overnight window, with managed charging configured
• Chargers OCPP-compliant; CSMS alerts on failed sessions and projected shortfall
• Fallback plan (rapid chargers on route, spare vehicles) documented
• Workshop staff trained for high-voltage safety; tools available

Monthly governance review

• KPI pack reviewed, with actions and owners
• Overdue PMs and defects open more than 48 hours explained
• Fuel and energy exceptions closed or escalated
• Safety events reviewed; coaching done; repeat events escalated
• Data-quality report clean (silent devices, unassigned trips)

Privacy and data audit (annual)

• Each data feed has a purpose, lawful basis, retention and access list
• Monitoring policy shared with drivers and current
• DPIA reviewed where required
• Audio recording off by default; driver-facing cameras justified where used
• Data removed from disposed vehicles; leavers’ accounts removed

After a collision

• People cared for; emergency services called where needed
• Telematics and camera data preserved before routine deletion
• Report, photos, statements, duty-status record collected
• Preventability decided against the written definition
• Actions (coaching, route, vehicle) recorded and tracked to close

One-page fleet policy (outline).

FLEET POLICY                                   Owner: Fleet manager   Review: annual
1. Scope: vehicles owned, leased, rented, and own vehicles used for work
2. Eligibility: who may have an assigned vehicle; pool use for others
3. Drivers: licence checks (on hire, every 6 months), training, fitness to drive
4. Vehicle use: business use, private use rules, passengers, phone use, alcohol
5. Checks: daily walkaround; defect reporting; who signs off repairs
6. Working time: limits that apply; planner and driver responsibilities
7. Fuel and charging: cards, limits, home charging reimbursement, idling
8. Monitoring: data collected, purpose, retention, who sees it, drivers' rights
9. Collisions: what to do, reporting deadline, investigation
10. Replacement: review cycle, criteria (EAC, condition, mission)

Defect report and close-out (fields).

DEFECT REPORT
Vehicle (unit, registration):        Odometer:        Date/time:
Reported by:                         Found during: walkaround / trip / inspection
Item:  brakes | steering | lights | tyres | wheels | mirrors | wipers | coupling
       | horn | emergency equipment | other: __________
Description:
Safety-critical?  yes / no           Vehicle off road?  yes / no
REPAIR (workshop)
Work order:        Action taken:                     Parts:
Repaired by:       Certified safe by:                Date/time:
Driver review of repair (next driver):  signature / app confirmation

Replacement review record (one per vehicle group).

REPLACEMENT REVIEW                           Group: ______   Date: ______
Units in group: __   Average age: __   Average km: ______
Current EAC curve (attach model): economic life = __ years
Condition: open defects __, repeat defects __, downtime days (12 mo) __
Mission change expected? ______   Utilisation (days %) __
Options considered: like-for-like / different class / EV / lease / remove
Decision: ______   Rationale: ______   Budget year: ____
Approved by: ______ (fleet)   ______ (finance)

Telematics data decision record.

DATA DECISION RECORD                        ID: TEL-___   Date: ______
Feed: location / driving events / video / audio / fuel / engine data
Decision it supports: ______________________________
Lawful basis or legal requirement: ______   DPIA needed? yes / no
Who can see it: ______   Retention: ______   Private-use handling: ______
Drivers informed how: ______   Review date: ______

Is FleetOps a recognised discipline? The practice is, as fleet management or fleet operations, with an association and certification (NAFA, CAFM)^[3] and regulations that define much of the work. The one-word “FleetOps” is mainly a product name (see Term status).

How is this different from logistics or supply chain management? Supply chain decides what moves where, by which mode and carrier; fleet operations runs the organisation’s own vehicles and drivers. A field-service or utility fleet can be large and move few goods.

Do we need telematics? Only if a decision depends on it: sizing the fleet, scheduling maintenance by engine hours, coaching drivers, reconciling fuel, or proving working time. Write those decisions down first, then apply the rules in Safety, privacy and compliance.

When should we replace a vehicle? When one more year costs more than the best replacement’s EAC (the replacement rule). Bring it forward for a mission change or repeated serious defects; push it back only with evidence of lower running costs.

Are electric vans cheaper? It depends on duty cycle, charging price and keep period: in Lab 1 the same van wins with depot charging and loses on public charging. Run the model with your own numbers and test winter range.^[37]

Do hours-of-service rules apply to our vans? It depends on jurisdiction, weight, use and exceptions such as the US short-haul radius;^[49] in the EU, tachographs now reach some 2.5–3.5 t vans on international hire-or-reward work.^[17] Check with your regulator; this is not legal advice.

• Established now: telematics with engine data; electronic duty-status records; outward and AI-assisted dash cameras; managed depot charging over OCPP;^[26] route optimisation solvers.^[56]
• Spreading, with conditions: condition-based maintenance from fault codes; dispatch that plans around state of charge; Plug & Charge and bidirectional charging under ISO 15118,^[63] where vehicle and charger support it.
• Regulatory change to watch: data access rights for connected vehicles in the EU (the Data Act has applied since 12 September 2025);^[21] tachograph scope extending to light vehicles in international transport;^[17] shifting zero-emission fleet mandates, such as the partial repeal of California’s Advanced Clean Fleets rule.^[41]
• Automation: remote assistance for automated vehicles^[58] moves fleet work towards exceptions, recovery and depots.
• Speculative or vendor-driven: predictive maintenance for small fleets; vehicle-to-grid revenue in an EV business case. Treat these as hypotheses to test.

Option A: replace all 40 now with electric vans; five charge on the depot’s spare capacity and 35 at public rapid chargers until the depot upgrade is done. Option B: keep the diesels one more year, order the depot upgrade now, run a five-van electric pilot on the depot’s existing spare capacity, and replace the rest at the end of year 7.

Assumptions (from the replacement rule and Lab 1): a diesel’s seventh year costs about 12,848 and its eighth about 13,858; a depot-charged electric van’s EAC is about 12,737; public charging at 0.45 instead of 0.22 per kWh adds about 1,380 a year per van (6,000 kWh × 0.23).

Reasoning.

1. Option A meets the target now, but the electric van is cheaper only with depot charging. Over the 12 to 15 months of the upgrade, 35 vans would pay about 1,380 more each per year for energy (about 48,000 to 60,000 in total), plus driver time at public chargers that the model does not count.
2. Option B’s extra diesel year (12,848) costs almost the same as a depot-charged electric van (12,737). Waiting one year is cheap; waiting two (13,858) is not.
3. The pilot measures what the model lacks: winter energy use, charging behaviour, driver feedback.
4. Uncertainty: faster diesel resale decline or a slipping utility date hurt B; rising public-charging prices hurt A.

Decision. Option B, with a trigger: if the upgrade slips past month 15, replace the oldest diesels with electric vans on managed charging that fits existing capacity rather than running all diesels into year 8. The attractive choice is wrong because charging infrastructure, not the vehicle, sets the cost.

After this page you should be able to: explain fleet operations and its handoffs; calculate EAC and an economic life; estimate fleet size from trip data; design a layered maintenance strategy and defect close-out; plan within working-time rules; design a safety and monitoring programme defensible on privacy; and decide whether a vehicle group is ready to electrify.

Fleet operations is not glamorous and not new, and that is its strength: the problems are understood, the rules written down, and the economics fit in a spreadsheet. Its failures are failures of discipline: vehicles nobody owns, inspections that slip, defects that get driven, rules planners cannot see, and buying on price.

The newer pressures (telematics, cameras, electrification, automation) do not change this. They make the data richer, the privacy questions sharper and the cost model more sensitive to where and when energy is bought. A fleet that already knows its utilisation, its economic lives and its defect close-out times can absorb those changes with evidence. A fleet that does not will find that the new technology mostly measures its old problems more precisely.

Changelog

• 2026-10-23: First version. Research, ledger (research/fleetops-ledger.md), four labs executed, nine diagrams. Not yet independently reviewed.
• 2026-10-23 (review round 1): ICO DPIA scope corrected and EU rule added; Class8 quotation made verbatim; Traccar script now converts to local time and numbers working days, Lab 2 takes the day count and rejects bad rows; CARB status from CARB’s own release; ISA vehicle categories; OCPP security, licence checks, tyres, claims and chargeback added; glossary extended; sequence and chart diagrams revised; repetition cut (reviews/fleetops-r1-response.md).
• 2026-10-23, post-score polish: AFLEET footnote dated December 2023 (DOE/EE-2798); TU Wien conclusion quoted verbatim from the abstract; reference architecture gains an energy and cost record (fuel cards, tanks and chargers feed it; the gateway feeds maintenance); EAC chart narrowed to fit a phone; repeated public-charging result in the EV section replaced by a link to Lab 1 (reviews/fleetops-polish.md).