The self-clearing warning
Tell the person, not the street
A warning gives someone the chance to leave before anything else happens. It also creates the one fact that matters more than proximity ever did: that they were told, and chose what to do next.
The idea this page is built on came in plainly: put a warning alarm where the monitored person can hear it, tell them to leave, and if they leave, that is a good outcome. If they do not, now there is a problem.
That instinct is right, and this design document adopts it. Three reasons, in order of how much they matter.
- It converts an alert from an accusation into a compliance opportunity. Every other tier on the alert ladder tells somebody else about the subject. This one tells the subject, first, and gives them a way to end the encounter themselves.
- It is markedly less punitive, and that is a legal question, not a courtesy. Whether a nominally civil scheme is treated as punishment turns on its effects. Seven factors are used to decide whether a nominally civil scheme is punitive in effect. Two of them - whether the scheme imposes an affirmative disability or restraint, and whether it is excessive relative to its purpose - read very differently for a private message a person can act on than for a public alarm they cannot. The legal architecture page runs that scoring.
- Non-compliance after a warning is far stronger evidence than presence. Presence is ambiguous and, on the numbers this document keeps returning to, usually innocent. Remaining after being told to leave is not ambiguous. It is a decision, made with notice, at a known time, recorded. For a supervising officer or a court that is a materially different fact, and it is the real prize in the whole idea.
The flaw, and it is the most dangerous idea on this site
The wording in the original ask was "you are within fifty feet of a child." That sentence is a child detector.
It confirms that a child is present, to the one person in the world who must never be told that. Say it once and it is a mistake. Build it into a device and it is a sensor: a person can walk a street, a park, a shopping center, and read the alarm's silence and speech as a survey of where children are. The whole two-sided architecture is designed so the child side broadcasts nothing [33][34][35] - and then a warning phrased that way hands the same information back through the front door, out loud, to the subject.
This is the inversion risk of the child-side device, turned around and pointed at the person being monitored. It is not a wording problem to be fixed in review. It is the difference between a protective system and a targeting one, so it is fixed structurally, below.
The fix: the zone triggers it, never the child
Protected zones are public knowledge. A map of schools, playgrounds, daycares and pools is published, and it has to be - a person cannot comply with a boundary they are not allowed to see, and due process requires the boundary be knowable before it is enforced. A warning that fires on entering a published place therefore discloses nothing that a map does not already disclose.
A warning that fires on a child being there discloses the one thing the system exists to protect. So the trigger is the zone, and the machine below has no input for presence at all.
| Rule | What it means | Why it is structural rather than editorial |
|---|---|---|
| W-1 Zone triggers, presence never does | The warning is raised by crossing into a published protected zone and holding it. Nothing about who else is there reaches the warning path. | The state machine takes zone membership and the clock. There is no presence field to read, so there is no branch that could ever depend on one. |
| W-2 No presence, direction, distance, bearing or count | The message says where the subject is. It never says who is near, how near, which way, or how many. | Every string the machine can emit is swept against those three classes by a test on this page. |
| W-3 Identical behavior either way | Wording, timing, repetition, escalation and the written record are byte-identical whether or not a child is present. | If any of them varied, the variation itself would be the signal. Wording alone is not enough - a warning that fired faster when a child was there would leak just as badly. |
| W-4 Private delivery only | Haptic on the subject's own device plus text on that device's screen. Never a speaker, never a light, never anything a bystander can perceive. | An audible alarm in a public place is the de-anonymization mechanism. The threat model puts a number on it: in an ordinary crowd, with a handful of people paying attention, an audible alert identifies the subject to the street more often than not. |
| W-5 The zone may be named, the reason may not | "Zone: Riverside Primary School" is fine. "Zone: Riverside Primary School, occupied" is not. | The name is on the published map. The occupancy is not, and never becomes so. |
| W-6 The grace clock is a function of the subject's own history | It shortens with the subject's prior warnings in the trailing day, and with nothing else. | A clock that varied with anything outside the subject would be a side channel carrying exactly what W-1 removed. |
| W-7 Warnings are logged as issued, whether or not anyone was there | The record says a zone was entered and what the subject did. It never records who else was present, because the warning path never learned it. | A record that noted presence would reconstruct the detector after the fact, in discovery. |
The exact wording
Four lines, fixed. Only the zone's published name and the grace clock vary, and the subject already knows both.
You are inside a protected zone. Leave the zone now.
This warning is sent every time you enter this zone. It does not tell you whether anyone is here.
Zone: Riverside Primary School. You have 60 seconds.
Leaving clears this. Remaining after this message is recorded as a decision.What the second line is doing
It is not politeness. It is the countermeasure. A subject who receives this message and reasons "the alarm went off, so a child must be here" has drawn the inference the whole design exists to block - and the sentence tells them, correctly, that the inference does not hold. The message is boilerplate, and saying so out loud is what makes it safe.
It also happens to be true in the ordinary case. Most warnings will fire on empty grounds, at times when nobody is there at all.
What is deliberately absent
- No statement that anyone is present, and no statement that nobody is.
- No direction, bearing, distance or count.
- No variation in urgency, tone or repetition between an occupied zone and an empty one.
- No sound and no light that anybody else can perceive.
- No identification of the subject to anyone in the street, at any tier, ever.
The warning cannot be false
It makes no claim that could be false. It reports the subject's own position against a published boundary - geography, not detection. This is why it can be a great deal more frequent than anything on the alert ladder without producing the alert fatigue that has wrecked deployed monitoring programs.
95.50% of monthly GPS monitoring alerts in Cook County were false, up from 57.79% under radio frequency.Hwang, Maniar, Smith & Nivens (2026), "False Alerts Related to GPS Electronic Monitoring in Illinois Cook County Sheriff's Office," Criminal Justice Review; urban GNSS error, Satellite Navigation (2025)
Driven by GPS drift; the worst zip codes were distinguished by tall buildings. [17]
The cost is real and this document states it plainly: the design deliberately accepts warnings that are, from the subject's point of view, indistinguishable from false ones. A person will be buzzed walking across an empty school field at night. That is not a defect being tolerated. It is the mechanism. If the warning could be distinguished from a false one, it would be a detector.
The warning state machine
Drive it: enter the zone, watch the grace period, leave or stayWhat the subject is shown, verbatim
You are inside a protected zone. Leave the zone now. This warning is sent every time you enter this zone. It does not tell you whether anyone is here. Zone: Riverside Primary School. You have 60 seconds. Leaving clears this. Remaining after this message is recorded as a decision.
Timeline
- 10 swarning haptic + subject device
Issued on zone entry.
The record this writes
A record is written on every resolved warning, and the subject can obtain it.
- The trigger is the zone, not a child.
This is the state machine itself, from /assets/bracelet.js, folding a timeline of zone-membership samples into a transcript. The samples handed to it carry a presence flag on every tick; the machine never reads it. The check page runs the same machine against its vector table, including a sweep of 300 paired runs whose only difference is that flag.
The clock, clearing, and what escalation is
| Stage | Timing | Delivery | What it means |
|---|---|---|---|
| Entry dwell | 10 s continuously inside | Nothing | Absorbs boundary jitter and people walking past a corner of the zone. Below this, no warning is issued at all. |
| Warning | At the end of the entry dwell | Haptic, then text on the subject's own device | The four fixed lines. The grace clock starts. |
| Grace | 60 s on a first entry; 45, 30, then a floor of 15 s as prior warnings accumulate in the trailing day | A second haptic at 30 s remaining | Time to leave. It shortens for a repeat visitor because a person who has been told twice today does not need to be told at length a third time. |
| Exit | Any time before grace runs out | Haptic | The grace clock stops. Clearance is confirmed after 60 s continuously outside, so stepping over the line and back does not buy a fresh grace period - the clock resumes where it stopped. |
| Cleared | On confirmed exit | Haptic and a written record | Nothing is referred anywhere. A compliance record is written and the subject can obtain it. |
| Escalation | When grace runs out with the subject still inside | Haptic to the subject; a queue entry for the supervising officer | Tier 1 - the officer queue. Not audible, not public, not a dispatch. The fact carried forward is that the subject remained after notice, and for how long. |
| Pattern | From the third warning in the trailing day | The same queue entry, flagged | Repeated entry is a supervision matter, addressed by a person, not by a louder alarm. |
Cleared is a record worth having
A cleared warning is written down and it belongs to the subject as much as to the agency. Someone under supervision who can show a run of warnings entered and cleared within seconds has evidence of compliance that no self-report produces. That is worth something at a review hearing, and it is worth something to the person.
The record names a zone, a time, a grace period and what the subject did. It does not name, count or describe anyone else, because the warning path never learned anything to name.
Escalation is quiet, and stays quiet
Non-compliance moves the encounter onto the ordinary ladder at the officer-queue tier. It does not skip tiers, it does not make a sound, and it does not identify the subject to anybody in the street. Audible response remains gated behind cross-modality agreement, sustained dwell, zone classification and a human adjudicator, exactly as before.
What the warning adds is not force. It is the difference between "was near" and "was told, and stayed."
Where the audit is not checked, the design is theatre
A warning that escalates into a queue nobody reads is worse than no warning, because it manufactures a record of notice that was never acted on. The published field evidence on this is not encouraging.
80% of alerts were never checked by officers in a Tennessee audit.Karsten & West, Brookings (2017) [29]
So the warning tier carries the same operational requirement as every other tier on this site: a named response time, a named person, and a published figure for how many queue entries were actually reviewed. That belongs in operations, and it is a condition of the design rather than a nice-to-have.
What this does not do
- It does not stop anybody. A person who intends harm can ignore a buzz on their wrist. The warning changes what is known and recorded; it does not change what is physically possible.
- It does not detect a child, and must never be built so that it could. Everything on this page is an argument for keeping it that way under pressure - and there will be pressure, because a presence-aware warning would look more responsive in a demonstration.
- It does not reach the great majority of harm. Most abuse is by family and acquaintances in private homes, where there is no zone to enter and no boundary to cross. See coverage.
- It is not self-executing law. A warning can only be a condition of supervision if an order says so, and the order has to say what happens when it is ignored. That is drafting, not engineering.
This page is part of a design document. Nothing described here has been built, and the wording, timings and thresholds above are proposals to be tested, not measurements of a deployed system.
How this works
1:08 · narrated · captioned · nothing here is sent anywhere when you press play. Prefer to read it? The transcript is below.
Transcript
| At | Section | What it says |
|---|---|---|
| 0:02 | What it is | A warning gives the monitored person a chance to leave before anything else happens. It is delivered privately, to their own device. Nobody in the street hears it or sees it. |
| 0:12 | The flaw | The wording in the original ask was, you are within fifty feet of a child. That sentence is a child detector, handed to the one person who must never be told. |
| 0:22 | The fix | So the trigger is the zone and never a child. Protected places are on a published map, so a warning on entry discloses nothing the map does not. There is no input for presence. |
| 0:34 | The wording | Four fixed lines. Only the zone name and the grace clock vary. One line says the warning is sent every time, and that it does not tell you whether anyone is here. |
| 0:42 | Leaving and staying | Leaving clears it, and the cleared record belongs to the subject as evidence of compliance. Remaining after notice is a recorded decision, and it goes quietly to one officer. |
| 0:54 | The limit | It stops nobody. A person who intends harm can ignore a buzz on their wrist. What the warning changes is what is known and recorded, not what is physically possible. |