What it is
It reads a stream of measurements and says one thing: something changed here, investigate now. It does not diagnose, declare an emergency or set policy.
- Deterministic. No neural network, no training, no weights, no random seed in the detection path. Identical input produces identical output on every run, by anyone, on any machine. Every alert traces to the rule that fired it.
- Air-gapped capable. No cloud, no GPU, no external service. It runs in a browser on ordinary hardware.
- Sealed. The instrument, scoring code, reference data and measurement record are hash-frozen, so any evaluation is provably against the same artefact.
The five layers
| Layer | Name | Watches for |
|---|---|---|
| 1 | Violation | The reading crossed the defined action limit |
| 2 | Spike | A sharp statistical excursion above baseline |
| 3 | Wave | A slower wave building above baseline |
| 4 | Disruption | The signal’s rhythm losing phase lock: a change in structure, not magnitude |
| 5 | Coherence | Strict confirmation of phase-lock and harmonic resonance together |
Layers 2 to 5 work on readings below the action limit. They are the early-warning function. A parallel spatial engine separates a localised source from regional drift.
Flagship result: New Delhi PM2.5
Real recorded data
27,820 consecutive hourly records from one New Delhi station, the largest continuous series in the portfolio.
Air Mode uses a trailing 30-day baseline. It uses only data already seen, never future data.
Environmental contamination records
Real recorded data
Every one of these is real recorded regulatory or monitoring data.
| Dataset | Contaminant | Medium | Records | Missed | False alarms | F1 |
|---|---|---|---|---|---|---|
| Flint, Michigan lead crisis | Lead | Drinking water | 31 | 0 | 0 | 1.000 |
| Bethpage TCE plume, Long Island | Trichloroethylene | Groundwater | 6 | 0 | 0 | 1.000 |
| East Palestine derailment, Ohio | Vinyl chloride | Air / soil | 33 | 0 | 0 | 1.000 |
| Daily asbestos exposure events | Asbestos | Air | 3,499 | 0 | 0 | 1.000 |
Flint, Bethpage and East Palestine were re-run at eight sensitivity settings. F1 was 1.000 at every one.
Early warning without the threshold
Real recorded data
On the 2011 Yap Island dengue outbreak (185 daily observations, 978 real reported cases), with the action-limit layer switched off entirely, the remaining layers first flagged the epidemic 70 days before its peak, at 7 cumulative cases. With the threshold layer on, the flag comes 85 days before peak.
On the Texas coast, the spatial engine was given 4,254 GPS-tagged beach surveys of plastic pellets with no labels and no tuning. It singled out the Galveston Bay / Houston Ship Channel stretch as the localised source.
Across the whole portfolio
Real data and synthetic data are kept apart and labelled, in the benchmark report and here.
| Kind of data | Tests | Records | Missed events |
|---|---|---|---|
| Real recorded data | 14 | 5,546 | 0 |
| Synthetic stress series | 8 | 11,200 | 0 |
Every result shown above on this page is from real recorded data.
Instrument v5.0, locked 9 September 2026. Covered by U.S. Non-Provisional Patent Application 19/700,318, pending.
Read more
Licensing & collaboration
Offered for licensing, not for sale. Independent validation, engineering collaboration and research partnerships are welcome.
[email protected]Inventor: Paul Samuel Guarino, Independent Researcher & Inventor