Skip to content
FICUS by Numascale, home
FIG. A1Applications

Applications

Thirteen products. Always aware.

Pick a device. See its Reflex software, numbers and evidence status.1

Select a device · scroll for more

Eye tracking: line drawing with the sensor positions marked in orangeEye tracking: line drawing with the sensor positions marked in orange

FICUS · Eye-tracking Reflex

402 nJ per X/Y

● Sensors · Event camera at the nose pad, facing the eye · 640×480

Measured · nRF54LM20

Wearables

Eye tracking

Where is the pupil, right now?

Reflex software finds pupil X/Y on the microcontroller.2

402 nJX/Y · energy-first2

39.5 µsX/Y · energy-first

18.4 pxmean error, accuracy-first

Compare both profiles2

Lowest energy and fastest: 402 nJ and 39.46 µs per X/Y update. Lower mean and median error than all three references; P95 error slightly higher than Axon and TensorFlow Lite.

FICUS Energy-first: how much faster, how much less energy and how much lower the error is than Neuton, Axon and TensorFlow Lite. Cells marked worse are FICUS losses.
FICUS vsNeutonNordicCPUAxonNordicNPUTF LiteOpen sourceCPU
Faster per X/YFICUS 39.46 µs41.23 times faster40.60 times faster160.98 times faster
Less energy per X/YFICUS 402 nJ96.57% less energy, 29.12 times less96.83% less energy, 31.58 times less99.26% less energy, 135.89 times less
Lower mean errorFICUS 21.40 px45.19% lower mean error25.97% lower mean error25.32% lower mean error
Lower median errorFICUS 15.71 px55.76% lower median error40.63% lower median error39.70% lower median error
Lower P95 errorFICUS 61.82 px22.72% lower P95 error5.67% higher P95 error, FICUS is worse5.76% higher P95 error, FICUS is worse

Each cell is the FICUS margin over that model; ember, marked worse, where FICUS loses. Speedups are the published factors; percentages are computed from the measured table.

Measured · nRF54LM20B · Nov 20262

1 / 13

Illustrations · not product photographs9

FIG. A2Eye tracking, replayed

Eye tracking, replayed

Five models. One pupil.

Two FICUS profiles. Neuton. Axon NPU. TensorFlow Lite.5

Slowed · 4 observations/s · not runtime footage

The ring is the annotated pupil centre, the cross is each model's estimate, and the number under a panel is that frame's error in pixels. The energy and time in each header are the measured mean per update on the board, not per frame.

31.6×

Less energy vs Axon NPU2

0.47%

CPU at 120 updates/s · PROJECTION3

360

Validation observations5

Measured · nRF54LM20B · Nov 20262

FIG. A3Why gaze matters

Why eye tracking

Look. Ask. Be understood.

Gaze tells AI glasses which thing you mean.6

  1. 01 · The missing input

    Voice says what. Gaze says which.

    Voice gives intent. Gaze gives attention.6

  2. 02 · Always on

    It runs every frame, all day.

    Stay aware without draining the cell.6

  3. 03 · FICUS

    The same function, as software.

    FICUS Reflex software runs on a standard Cortex-M33.2

More glasses. More gaze.7

+263%

AI-glasses shipments · H1 2026 YoY7

27.3M

Display-less glasses · 2030 FORECAST7

12.2M

Display glasses · 2030 FORECAST7

10.4M

Mixed-reality headsets · 2030 FORECAST7

FIG. A4One function, many products

Beyond glasses

One function, seven sectors.

A glance can guide work, play and care.8

  • AI glasses: line drawing with the eye cameras marked in orangeAI glasses: line drawing with the eye cameras marked in orange

    Consumer

    Measured task

    AI glasses

    Which sign, label or face?

    Replay ↑2

    Gaze guides requests and notifications.8

  • VR and gaming: line drawing with the eye cameras marked in orangeVR and gaming: line drawing with the eye cameras marked in orange

    Entertainment

    Exploring · same near-eye input

    VR and gaming

    Why render detail you cannot see?

    Detail, menus and aiming follow gaze.8

  • Guided work and inspection: line drawing with the eye cameras marked in orangeGuided work and inspection: line drawing with the eye cameras marked in orange

    Manufacturing

    Exploring · same near-eye input

    Guided work and inspection

    Next step, with both hands busy?

    Gaze advances steps and confirms inspection.8

  • Rugged AR on site: line drawing with the eye cameras marked in orangeRugged AR on site: line drawing with the eye cameras marked in orange

    Construction

    Exploring · same near-eye input

    Rugged AR on site

    Plans and help, with gloves on?

    A glance opens the hard-hat visor.8

  • Neurology and monitoring: line drawing with the eye cameras marked in orangeNeurology and monitoring: line drawing with the eye cameras marked in orange

    Medical

    Exploring · same near-eye input

    Neurology and monitoring

    What can eye movements reveal?

    Track movements for clinical research.8

  • Attention studies: line drawing with the eye cameras marked in orangeAttention studies: line drawing with the eye cameras marked in orange

    Research

    Exploring · same near-eye input

    Attention studies

    Where do shoppers, drivers and athletes look?

    Map attention beyond the lab.8

  • Driver distraction warning: line drawing with the eye cameras marked in orangeDriver distraction warning: line drawing with the eye cameras marked in orange

    Automotive

    Exploring · different optics

    Driver distraction warning

    Eyes off the road?

    Gaze can trigger a distraction warning.8

Illustrations · not product photographs8

FIG. A5What the pain costs

Small failures. Large costs.

Published pain benchmarks. Not FICUS results or customer losses.9

€78,068

per fracture, first-year care14

US$2.3M

per hour, large car plant15

US$22,434

per person over 10 years35

All 36 scenes · costs and sources
  1. Norway

    ▲ On the charger

    €78,068 per fracture, first-year care14

    Mean hip-fracture care cost, standardised pathway

    Source 14 ↗

  2. Rayong, Thailand

    ▲ Asleep between checks

    US$2.3M per hour, large car plant15

    Unplanned downtime cost of a large automotive plant

    Source 15 ↗

  3. Jurong Island, Singapore

    ▲ Nobody watching

    ~US$500k per hour, large oil & gas site16

    Downtime cost at large oil and gas facilities

    Source 16 ↗

  4. Pilbara, Australia

    ▲ Nobody watching

    A$25,000 per hour of stopped belt17

    Iron-ore conveyor downtime cost

    Source 17 ↗

  5. Kazakhstan

    ▲ Asleep between checks

    US$600M for one derailment settlement18

    East Palestine derailment class-action settlement

    Source 18 ↗

  6. Hanoi, Vietnam

    ▲ Cries wolf

    ~US$19B per year in Vietnam19

    Economic cost of road deaths and serious injuries

    Source 19 ↗

  7. Rotterdam, Netherlands

    ▲ Asleep between checks

    US$10–12M per crane, up to 2-year lead time20

    Procurement cost of a ship-to-shore crane

    Source 20 ↗

  8. Catania, Italy

    ▲ Nobody watching

    US$126,000 per vacuum-pump incident21

    Chamber pump failure with a 24-hour outage

    Source 21 ↗

  9. Kigali, Rwanda

    ▲ Cries wolf

    US$66,660 per closed alarm-response claim22

    Mean indemnity for unanswered equipment warning alarms

    Source 22 ↗

  10. Mexico City, Mexico

    ▲ Nobody watching

    US$39B per year worldwide23

    Estimated annual cost/value of non-revenue water

    Source 23 ↗

  11. New South Wales, Australia

    ▲ Too late

    US$1.925B in accrued wildfire claims24

    PG&E's liability for the 2021 Dixie Fire

    Source 24 ↗

  12. Shenzhen, China

    ▲ Nobody watching

    US$258.3M NYC losses in 202325

    Upper estimate of e-mobility battery fire losses

    Source 25 ↗

  13. Johor, Malaysia

    ▲ Nobody watching

    >US$1M for 1 in 5 major outages26

    Share of recent major outages that cost over US$1M

    Source 26 ↗

  14. Ouarzazate, Morocco

    ▲ Too late

    US$58.4M per solar hail claim27

    Average insured solar hail loss claim

    Source 27 ↗

  15. Kolwezi, DR Congo

    ▲ Too late

    US$150,000 per hour, large copper mill28

    Contained-metal value a large copper SAG mill processes per hour

    Source 28 ↗

  16. Taoyuan, Taiwan

    ▲ Too late

    US$1M per server shipment29

    Value of one high-value server shipment

    Source 29 ↗

  17. N1 highway, South Africa

    ▲ Nobody watching

    US$46,765 per non-injury truck crash30

    Estimated economic cost of a non-injury large-truck crash

    Source 30 ↗

  18. Quebec, Canada

    ▲ Nobody watching

    >US$30,000 per freeze-damage claim31

    Average insurer payment for freeze-related water damage

    Source 31 ↗

  19. West Java, Indonesia

    ▲ Asleep between checks

    US$59M per heavy-industry plant, per year32

    Annual unplanned downtime cost of a large heavy-industry plant

    Source 32 ↗

  20. Manila, Philippines

    ▲ Asleep between checks

    ≈US$5M to replace one substation33

    Estimated replacement cost after a substation fire

    Source 33 ↗

  21. São Paulo, Brazil

    ▲ Nobody watching

    US$6,386 per AF-related stroke admission34

    Median hospital cost of an AF-related ischemic stroke

    Source 34 ↗

  22. In conversation

    ▲ Left to the user

    US$22,434 per person over 10 years35

    Excess healthcare cost linked to untreated hearing loss

    Source 35 ↗

  23. Changwon, South Korea

    ▲ Too late

    US$150,000 per hour, top end for SMEs36

    Downtime cost for small and medium manufacturers

    Source 36 ↗

  24. In the factory

    ▲ Asleep between checks

    US$253M per large plant, per year37

    Annual unplanned downtime cost of an average large plant

    Source 37 ↗

  25. Rivers State, Nigeria

    ▲ Nobody watching

    ~US$3M revenue lost in two days38

    Revenue lost during two days of evacuation downtime

    Source 38 ↗

  26. Hamburg, Germany

    ▲ Nobody watching

    ~US$35B per year, biopharma39

    Losses from failures in temperature-controlled logistics

    Source 39 ↗

  27. Madrid, Spain

    ▲ Nobody watching

    >US$361B global occupational heat loss40

    Heat-related deaths, incapacity and treatment costs

    Source 40 ↗

  28. Mumbai, India

    ▲ Cries wolf

    US$8,761 per range/cooktop fire41

    Average home range/cooktop fire property loss

    Source 41 ↗

  29. Norway

    ▲ Easily fooled

    €20,700 per stolen insured car42

    Average cost to motor insurers of a car theft

    Source 42 ↗

  30. Lagos, Nigeria

    ▲ Nobody watching

    US$1.41B per year in US construction43

    Struck-by injury workers' compensation costs

    Source 43 ↗

  31. North Sea

    ▲ Needs a crew

    US$100k–1M+ per offshore blade repair44

    Offshore blade repair once damage has escalated

    Source 44 ↗

  32. Atacama, Chile

    ▲ Too late

    US$1.54M per work-related death45

    Average economic cost of a fatal workplace injury

    Source 45 ↗

  33. Kumasi, Ghana

    ▲ Too late

    US$48,000 per medically consulted injury46

    Economic cost of a medically consulted work injury

    Source 46 ↗

  34. Cairo, Egypt

    ▲ Nobody watching

    US$6,884 per child asthma admission47

    Mean hospital service delivery cost of an asthma admission

    Source 47 ↗

  35. Europe

    ▲ Asleep between checks

    £1,172.20 per van-day off the road48

    Average UK business cost of van downtime

    Source 48 ↗

  36. Malawi

    ▲ Too late

    A$2,300 per cold-chain breach49

    Average vaccine value lost in a cold-chain breach

    Source 49 ↗

AI-generated scenes · public benchmarks9

FIG. A6Reading the evidence

Know what each number proves.

Next step · Reflexes for Things

Bring a model and a recording.

We compile Reflex software, verify it, then measure on your target.

Device makers

Pilot a Reflex

Pick two or three workloads. Measure Instinct against your current model.

Start a pilot →

Engineers

Try Reflex Lab

Explore Reflexes, compose Instinct, profile your CSV in the browser.

Open the Lab →

Investors

Investor materials

Round, plan and data room. Approved investors only.

Request access →

Notes

  1. 1Nordic evidence v6.3 · five Reflexes vs Neuton/TensorFlow Lite · nRF54LM20, CPU only · C1: features in RAM → class · PPK2 charge × assumed constant 3.0 V · sensing/feature extraction/radio/sleep/regulator losses excluded.
  2. 2Eye tracking: separate run · nRF54LM20B, Cortex-M33 at 128 MHz, Nov 2026 · prepared count map → pupil X/Y, each model's frontend included · charge × operator-declared 3.0 V · sensor/illumination/raw-event accumulation excluded · reused validation, descriptive accuracy · energy-first: lowest energy/time, 41.23× faster/29.12× less energy vs Neuton, 41–161× faster vs Neuton/Axon NPU/TensorFlow Lite · lower mean/median error, slightly higher P95 than Axon/TensorFlow Lite · accuracy-first: 916 nJ, 18.42 px mean/53.94 px P95, better than all three, 18–71× faster · active power: 10.2 vs 7.2–8.6 mW · matrix: published speed factors, percentages from measured table, ember/worse = FICUS loss · Nordic owns Neuton/Axon NPU · TensorFlow Lite: open-source framework.
  3. 3Compute-only projections, not product/battery-life measurements · energy × rate · 225 mAh CR2032 × 3 V · five-Reflex geomean basis, 100 decisions/s, no specific-product prediction · sensing/features/radio/sleep/regulator losses excluded, measured in pilot · eye tracking at 120/s: 0.048 mW if nothing else draws power · CPU share = measured time × 120/s · not a measured 120 Hz product.
  4. 4Historical: earlier FICUS studies, own datasets/targets · separate from Nordic run · pilots: no Reflex yet, built from maker's current model.
  5. 5360 validation observations · 60 × 6 recordings · chronology, source order, no interpolation · reused INI-30, 5,396 rows, descriptive accuracy, not a fresh test · headers: measured mean/update, not frame costs · ring: annotation, cross: estimate, panel: error in px · P. Bonazzi et al., Zenodo, 2024 · CC BY 4.0 · rendering/overlays: FICUS.
  6. 6Ganzin Technology via PR Newswire Asia, 15 Jun 2026 · natural interaction, context, personalised AI · trackers: 60–240 updates/s · AURORA IIE: about ¼ NPU-solution power · consumer module: NPU, up to 60 Hz · Ganzin Technology, accessed 14 Nov 2026 · Ganzin products, not measured against FICUS · complete trackers include sensor/illumination.
  7. 7Counterpoint Research, reported by Technology.org, 12 Nov 2026 · H1 2026, year-on-year · IDC, 15 Jun 2026 · 2030 forecasts from 2026: display-less 13.6M, display 3.0M (41.9%/year), mixed reality 3.2M · Forecasts count devices, not eye trackers · not all glasses have eye tracking · no attach rate assumed.
  8. 8
    Seven sectors · only INI-30 pupil-position task measuredOnly the pupil-position task on INI-30 near-eye event data is measured · other products need their own model, compiled Reflex and sensor measurement. · Ganzin Technology, accessed 14 Nov 2026 (consumer/work/site) · Singh et al., 2023: about 3° foveation, tracker energy can cancel savings · Giacomini et al., Journal of Neurology, 2026: saccades/fixation/pursuit, multiple sclerosis/Parkinson's/Alzheimer's/ALS, no clinical validation here · Ganzin Technology, accessed 14 Nov 2026: 120 Hz, 32 g · Seeing Machines, accessed 14 Nov 2026: EU, every newly registered vehicle from 7 July 2026, gaze direction/duration, cabin camera needs its own Reflex · illustrations, not product photographs.
  9. 936/42 scenes have fitting figures · others unpriced · AI-generated illustrations, not demonstrations/product photographs · public benchmarks, not FICUS results/customer losses · vendor/insurer surveys labelled · source currency/year, no inflation adjustment · collected 15 Nov 2026.
  10. 10
    Smart glasses · historical study details9 KB Reflex · unseen people, 0 false events · Every sample watched · main chip/camera/display sleep · 89% sure decisions, no main chip, FICUS glasses test · today's glasses: 2–5 h, about half display-off draw from main-chip idle.
  11. 11
    Measured Reflexes · unequal accuracy · validation/contextFrozen artifacts, unequal accuracy · runtime comparisons, not matched-accuracy rankings or fresh-data validation · Watches and rings: 99.84%, historical panel · Magic Wand · IMU features · 1.6 µs · CPU sleeps afterwards · Electric toothbrush: 100.00%, historical panel · Smallest measured Reflex · 0.54 µs, nRF54LM20 CPU · customer-brush integration validated in pilot · TensorFlow Lite artifact: 24% historical accuracy · Electric motors: 100.00%, historical panel · Class can gate radio alerts · sensing/radio excluded · historical May 2026: 100% of 10 CWRU bearing-fault classes, 2.0 µs, 4.7 nJ · Gearboxes: 98.88%, historical panel · White-box explanation map for maintenance sign-off · official GearXAI: 1st of 20 teams · IJCAI-ECAI 2026 · 72.6 KB · Asset trackers: 100.00%, historical panel · Local handling between radio reports · 163/163 replay rows: 156 known-session + 7 observer · unseen-session accuracy not established · full-device energy/battery life measured in pilot.
  12. 12
    Batteries, drones, vessels · historical study detailsEV battery packs: Regression Reflexes · 4 datasets · 6k–23k params · 17–31 µJ/decision · Nano-drones: 92-byte false-alarm gate on airframe · 3,793 false alerts cut · 94.5% targets kept · Autonomous vessels: Own controller · 276 B–2.4 KB · 1.4–13× faster than Axon NPU · 9–49 µs · historical FICUS studies.
  13. 13
    Hearing, earbuds, buildings · pilot contextHearing aids: Every frame, same cell beside hearing DSP · first market: value/microwatt and certification pull · Earbuds: Presence/voice gating lets Bluetooth audio sleep · software licences until FICUS-BLE-WIFI ships · Smart buildings: Battery sensors decide locally, quiet until context changes · current-model pilots, not measured results.
  14. 14
    Haugan et al. 2021 · Source/scopeMean hip-fracture care cost, standardised pathway · Trondheim, age ≥65 · 2008–13 cohort · 2012 euros · total care, not fall-detection savings · Haugan et al. — Health care services and costs after hip fracture, comparing conventional versus standardised care: A retrospective study with 12-month follow-up, Injury, 2021 ↗.
  15. 15
    Siemens 2024 · Source/scopeUnplanned downtime cost of a large automotive plant · Vendor survey · large plants, all causes · not line-audited loss · Siemens, The True Cost of Downtime 2024 ↗.
  16. 16
    Siemens 2022 · Source/scopeDowntime cost at large oil and gas facilities · 2021–22 survey, high oil prices · Siemens 2023: near US$125k per hour · all causes · Siemens/Senseye, The True Cost of Downtime 2022 ↗.
  17. 17
    Smart Idler · Source/scopeIron-ore conveyor downtime cost · Supplier's illustrative assumption · not site-audited rate · Smart Idler, Conveyor downtime: quantifying the true cost of idler failure ↗.
  18. 18
    Norfolk Southern 2025 · Source/scopeEast Palestine derailment class-action settlement · US hazardous cargo · settlement only, not total loss/typical Kazakhstan derailment · Norfolk Southern, 2025 Annual Report, Note 19, page K87, 2025 annual report (published 2026; settlement agreed 2024) ↗.
  19. 19
    Asian Transport Observatory 2025 · Source/scopeEconomic cost of road deaths and serious injuries · Approximate 2021 Vietnam societal cost · all road users, not crash-detector-preventable losses · Asian Transport Observatory / APRSO / IRF, Viet Nam Road Safety Profile 2025 ↗.
  20. 20
    TT Club 2026 · Source/scopeProcurement cost of a ship-to-shore crane · Asset value at risk · not one failure's cost · TT Club, Critical asset series: ship-to-shore crane, 2026 ↗.
  21. 21
    Pfeiffer Vacuum · Source/scopeChamber pump failure with a 24-hour outage · Supplier's worked example · not fab average · 24-hour outage · Pfeiffer Vacuum, ARGOS semiconductor services ↗.
  22. 22
    CNA / NSO 2015 · Source/scopeMean indemnity for unanswered equipment warning alarms · US insurer, 2010–14 · one claim · equipment alarms, not proven alarm fatigue/Rwanda costs · CNA and Nurses Service Organization — Nurse Professional Liability Exposures: 2015 Claim Report Update, Table 10 ↗.
  23. 23
    Liemberger & Wyatt 2019 · Source/scopeEstimated annual cost/value of non-revenue water · Global estimate, all non-revenue water · not Mexico City leakage/sinkhole repairs · Liemberger and Wyatt, Water Supply (IWA Publishing), Quantifying the global non-revenue water problem, 2019 (online 2018) ↗.
  24. 24
    PG&E 2025 · Source/scopePG&E's liability for the 2021 Dixie Fire · 2021 Dixie Fire, California, USA · end-2024 accrual before insurance/recoveries · lower-bound estimate, not final cost · PG&E Corporation and Pacific Gas and Electric Company, FY2024 Form 10-K, 2025 ↗.
  25. 25
    UL Standards & Engagement 2025 · Source/scopeUpper estimate of e-mobility battery fire losses · NYC, US, not Shenzhen · upper estimate includes monetized deaths/injuries, not insurer payouts · UL Standards & Engagement / Oxford Economics — The Impact of E-Mobility Battery Fires in NYC, 2025 ↗.
  26. 26
    Uptime 2026 · Source/scopeShare of recent major outages that cost over US$1M · Operator survey, all causes · not mechanical failures alone · Uptime Institute, Annual Outage Analysis 2026 ↗.
  27. 27
    GCube 2023 · Source/scopeAverage insured solar hail loss claim · GCube five-year hail sample · geography unspecified · not Morocco sandstorm/tracker-only loss · GCube Insurance — Hail No! report press release (now hosted by Tokio Marine GX), 2023 ↗.
  28. 28
    RME · Source/scopeContained-metal value a large copper SAG mill processes per hour · Supplier's standard contained-metal metric · not site-audited loss · Russell Mineral Equipment, Mill relining benchmarking scorecard ↗.
  29. 29
    Tive / GEODIS · Source/scopeValue of one high-value server shipment · Customer-story cargo value at risk · not audited damage loss · Tive, GEODIS customer story ↗.
  30. 30
    FMCSA 2024 · Source/scopeEstimated economic cost of a non-injury large-truck crash · US, not South Africa · 2022 dollars, all causes, not fatigue-only · includes delay, not fleet-only losses · US Federal Motor Carrier Safety Administration, Crash Cost Methodology Report, Table 3, 2024 ↗.
  31. 31
    State Farm 2026 · Source/scopeAverage insurer payment for freeze-related water damage · US State Farm, 2024–June 2025 · includes other freeze damage · not Quebec/heat-pump repair costs · State Farm, Frozen Pipes? Not So Nice!, 2026 (2024–June 2025 data) ↗.
  32. 32
    Siemens 2024 · Source/scopeAnnual unplanned downtime cost of a large heavy-industry plant · Vendor survey · large plants, all causes · not quarry loss · Siemens, The True Cost of Downtime 2024 ↗.
  33. 33
    RI DPUC 2023 · Source/scopeEstimated replacement cost after a substation fire · Rhode Island, USA, 2022 fire · utility estimate, not final invoice/Manila outage losses · Rhode Island Division of Public Utilities & Carriers, Nasonville Substation Fire Investigation, p. 15, 2023 ↗.
  34. 34
    Vieira et al. 2019 · Source/scopeMedian hospital cost of an AF-related ischemic stroke · Joinville private hospitals, 2016–17 · 7 AF strokes · median, 2016 PPP US$, not nominal dollars · Vieira et al. — The cost of stroke in private hospitals in Brazil: a one-year prospective study, Arquivos de Neuro-Psiquiatria, 2019 ↗.
  35. 35
    Reed et al. 2019 · Source/scopeExcess healthcare cost linked to untreated hearing loss · US insured adults 50+ · 2015 dollars · vs no hearing loss · association, not proven hearing-aid savings · Reed et al., JAMA Otolaryngology–Head & Neck Surgery, 2019 ↗.
  36. 36
    Siemens 2024 · Source/scopeDowntime cost for small and medium manufacturers · Siemens upper end · not median · Siemens, The True Cost of Downtime 2024 ↗.
  37. 37
    Siemens 2024 · Source/scopeAnnual unplanned downtime cost of an average large plant · Vendor survey · automotive/FMCG/heavy industry/oil & gas · all causes · Siemens, The True Cost of Downtime 2024 ↗.
  38. 38
    Seplat 2026 · Source/scopeRevenue lost during two days of evacuation downtime · Evacuation downtime · corrosion/vandalism leaks · not one flange's bill · Seplat Energy, FY2025 results, 2026 ↗.
  39. 39
    IQVIA 2019 · Source/scopeLosses from failures in temperature-controlled logistics · Industry-wide 2019 estimate · lost product/replacement/wasted logistics/root-cause costs · IQVIA Institute, reported by Air Cargo News, 2019 ↗.
  40. 40
    ILO 2024 · Source/scopeHeat-related deaths, incapacity and treatment costs · Global model, not Spain · health-economic loss, not employer bill · annual basis unspecified · International Labour Organization, Heat at work: Implications for safety and health, section 2.6, 2024 ↗.
  41. 41
    NFPA 2023 · Source/scopeAverage home range/cooktop fire property loss · US 2017–2021, not Mumbai · property-only damage, not disabled-alarm-attributable loss · NFPA — Home Cooking Fires: Supporting Tables, Table 13, 2023 ↗.
  42. 42
    GDV 2025 · Source/scopeAverage cost to motor insurers of a car theft · Germany, 2024 · approximate insured-car theft cost, all methods · not Norway/relay-only attacks · German Insurance Association (GDV), Car Theft Report 2024, 2025 (2024 data) ↗.
  43. 43
    Liberty Mutual 2025 · Source/scopeStruck-by injury workers' compensation costs · US 2022 · nonfatal, >5 days off work · not Lagos/per-incident cost · Liberty Mutual, 2025 Workplace Safety Index: Construction, 2025 (2022 data) ↗.
  44. 44
    ONYX Insight · Source/scopeOffshore blade repair once damage has escalated · Escalated blade damage · supplier examples, not averages · ONYX Insight, Cut turbine maintenance costs ↗.
  45. 45
    NSC 2024 · Source/scopeAverage economic cost of a fatal workplace injury · US 2024 all-industry societal average · not Chile/mining-specific/employer-only bill · National Safety Council, Injury Facts: Work Injury Costs, 2024 (estimate year) ↗.
  46. 46
    NSC 2024 · Source/scopeEconomic cost of a medically consulted work injury · US 2024 all-industry societal average · not Ghana/power-tool-specific/employer-only bill · National Safety Council, Injury Facts: Work Injury Costs, 2024 (estimate year) ↗.
  47. 47
    Barry et al. 2025 · Source/scopeMean hospital service delivery cost of an asthma admission · US ages 5–17 · 2016–2019 data, 2020 USD · all asthma admissions, not inhaler-error-only costs · Barry et al., Respiratory Medicine: Socioeconomic disparities in inpatient outcomes among US pediatric asthma hospitalizations, 2025 ↗.
  48. 48
    Mercedes-Benz Vans UK 2026 · Source/scopeAverage UK business cost of van downtime · Mercedes-Benz vendor survey · 500 UK decision-makers · all van downtime, not EV-/diagnosis-only · Mercedes-Benz Vans UK, Uptime Advantage report (vendor survey by Opinium), 2026 ↗.
  49. 49
    RACGP 2019 · Source/scopeAverage vaccine value lost in a cold-chain breach · NSW, Australia, 2019, revised procedures · government data via RACGP · not Malawi/road-only · Royal Australian College of General Practitioners, newsGP: Tip of the cold chain iceberg? Vaccines worth $26 million lost, 2019 ↗.