Alert logic
Fifty feet is not an event
On a platform, in a shop, on a bus, fifty feet happens hundreds of times a day to everyone. What makes an encounter meaningful is how long it lasts, whether it has happened before, and where it happens.
The design decision that matters most in this whole document is that distance alone never produces a dispatch. Everything else follows from that.
An automatic call to emergency services on every threshold crossing does three things, in this order: it floods dispatch, it desensitises the dispatchers, and it gets the programme switched off — usually after an incident where a real alert was treated as another false one. The graduated ladder exists to keep the escalated tier rare enough to be believed.
The ladder
| Tier | What happens | What it takes to get here | Who learns |
|---|---|---|---|
| T0 · No event | Nothing is recorded and nobody is told. | Outside the threshold, or no usable position fix. | Nobody |
| T1 · Log | Written to the band’s local log. Silent. | Inside the threshold, transient dwell. | Nobody, unless the log is later reviewed |
| T2 · Notify | Guardian device notified quietly. No sound, no police. | Sustained dwell short of the alert threshold, or any encounter on a degraded fix. | Guardian |
| T3 · Alert | Audible and haptic on the guardian’s device; supervising officer’s queue updated. | Sustained dwell, or a repeat pattern, or an exclusion zone. | Guardian, supervising officer |
| T4 · Dispatch | Referred to law enforcement as actionable, with the encounter record attached. | Sustained dwell and either an exclusion zone or a third repeat encounter. | Guardian, officer, dispatch |
Where the noise is
The audible alert belongs to the guardian’s device, not to the child’s band and not to the surrounding air. A band that makes a noise announces the child as a child under protection, marks them out to every other child in the room, and — per the threat model — points at the monitored adult. A guardian phone that buzzes does none of those things.
Tier classifier
The actual classifier from the specification, running here—
This is not a mock-up of the classifier. It is SP.classify() from /assets/engine.js, the same function the check page runs against its ten test vectors.
Why the ladder exists
The room where a bad threshold gets you ignored
Dispatchers triage. A source that is wrong nine times out of ten is a source they learn to skip, and they are right to.
Dispatch load model
Naive proximity dialling against the graduated ladder—
reduction in referrals from putting dwell, repeat and zone tests in front of the dispatch tier.
Capacity is dispatchers × 480 minutes × three shifts. Crossings per subject per day is the model’s softest input — it depends entirely on urban density and on how many children carry a band — so it is a control rather than a constant. The point the model makes is not the exact figure. It is the order of magnitude between the two policies, which holds across the whole range of the sliders.
Rules that are load-bearing
- A position you do not have cannot justify anything. A lost fix produces T0 — and the gap itself is logged, so an outage is auditable rather than invisible.
- A degraded fix caps at Notify. No dispatch on a position the system is not confident about.
- Dense venues triple the dwell requirement. Incidental proximity in a shop or on a platform is the norm, not the signal.
- Repeats count more than distance. Two encounters in thirty days is a pattern. Being three metres away instead of nine is not.
- T4 needs two independent reasons. Sustained dwell plus either a designated zone or a third repeat. Never proximity alone, at any distance, for any duration.
- The assessed risk tier moves the result by one step, never more. It is an input, not an override — see due process for why tiering has to come from a validated assessment rather than an offence label.