Safe PerimeterBorn Between 2 Generals

Specification · Draft 1.0

Safe Perimeter, in full

69 numbered requirements across nine sections. MUST and MUST NOT are normative; SHOULD is a strong default an adopter may depart from with a written reason. This page prints.

How to read this

Each requirement carries a reference of the form SP-n so it can be cited in a procurement document, a statute drafting note, or a complaint. Each section names the page that argues for it — the reasoning is not repeated here.

This is a design document. No system described here has been built, procured, piloted or deployed.

Requirements
69
Normative (MUST / MUST NOT)
59
Sections
9
Prohibitions
13

1. Scope and claim

Argued at /coverage/

RefLevelRequirement
SP-1MUSTThe system is described as closing a specific gap — sustained, unexplained physical proximity between a child and an adult under active monitoring, in places a fixed geofence cannot see — and never as a solution to child sexual abuse generally.
SP-2MUSTPublished material states that the large majority of child sexual abuse is committed by family members and acquaintances, and that most people who commit these offences have no prior conviction and are therefore invisible to the system.
SP-3MUSTPublished material states that the system has no effect on online grooming.
SP-4MUST NOTAny coverage figure be published without the assumptions that produced it.
SP-5SHOULDCoverage be re-estimated annually against the jurisdiction’s own case data rather than national averages.

2. Enrolment of monitored subjects

Argued at /due-process/

RefLevelRequirement
SP-6MUSTMonitoring be imposed prospectively, at sentencing, as an express component of the sentence, with the defendant on notice before plea.
SP-7MUSTAny application to a person already convicted proceed by individualised determination: a validated risk assessment, notice, a hearing, the right to counsel, and a written finding.
SP-8MUST NOTMonitoring be applied to any class of existing registrants by blanket administrative reclassification.
SP-9MUSTTiering derive from a validated actuarial risk assessment together with offence characteristics, and never from an offence label alone.
SP-10MUSTThe subject be entitled to challenge the assessment and to an independent evaluation.
SP-11MUSTEvery imposition carry a defined term, with the burden on the state to justify continuation at each review.
SP-12SHOULDThe highest tier be sized to the supervision capacity that actually exists, not to the population that qualifies on paper.

3. Enrolment of children

Argued at /legal/

RefLevelRequirement
SP-13MUSTA child be enrolled only by a guardian with verified parental authority, verified against the estranged-parent threat specifically.
SP-14MUSTEnrolment end automatically at a defined age and never convert into an adult account.
SP-15MUST NOTA child’s band be required as a condition of attending school, receiving services, or entering a public place.
SP-16SHOULDThe views of a child old enough to hold one be recorded at enrolment, because a device a child did not agree to is a device they will disable.

4. Alert logic

Argued at /alert-tiers/

RefLevelRequirement
SP-17MUSTThe response ladder have at least the five tiers defined in this specification, with distinct recipients at each.
SP-18MUST NOTAny tier above Log be reached on separation distance alone, at any distance, for any duration.
SP-19MUSTSustained dwell be required for every tier above Notify.
SP-20MUSTDispatch require sustained dwell together with either a legally designated exclusion zone or a third repeat encounter within thirty days.
SP-21MUSTA lost position fix produce no event, and the gap itself be recorded so the outage is auditable.
SP-22MUSTA degraded position fix cap the encounter at Notify.
SP-23MUSTDwell thresholds be raised in dense venues, where incidental proximity is the norm.
SP-24MUST NOTThe assessed risk tier move an outcome by more than one step in either direction.
SP-25MUST NOTThe child’s band emit any audible or visible alert.
SP-26MUSTAudible alerting be confined to the guardian’s own device, at tier 3 and above.
SP-27MUST NOTAny alert be placed automatically to emergency services on a proximity threshold.
SP-28SHOULDLadder thresholds be tunable per jurisdiction and every change be versioned against the encounters it affects.

5. Data protection and matching

Argued at /data-flow/

RefLevelRequirement
SP-29MUSTProximity matching be performed on the child’s device against a published list of pseudonymous keys.
SP-30MUST NOTAny position of any party be transmitted to a server at any time.
SP-31MUSTThe child’s band be receive-only: no advertised service, no response to probes, and no identifier transmitted.
SP-32MUSTBroadcast identifiers rotate at least every fifteen minutes and be unlinkable across days without the key that generated them.
SP-33MUSTThe band hold no identity for the child, so that a found device tells the finder nothing.
SP-34MUSTPosition history on the band be limited to a rolling window of no more than 72 hours and never be exportable or synchronised.
SP-35MUSTThe pairing secret be wiped on tamper and on guardian revocation.
SP-36MUSTThe published key list be partitioned by region wherever the full list exceeds what a band can hold.
SP-37MUSTPublished material state plainly that publishing the key list creates a proximity oracle any third party can build a receiver for, and that this capability did not previously exist.
SP-38MUST NOTA guardian be able to learn the identity of the other party to an encounter.
SP-39SHOULDKey distribution be rate-limited and client-attested, accepting that neither eliminates the oracle.

6. Devices

Argued at /devices/

RefLevelRequirement
SP-40MUSTThe subject’s device detect strap breach by continuity, wear by skin contact and temperature, and being worn by motion coherence.
SP-41MUSTThe subject’s device carry its own network radio and not depend on any device the subject controls.
SP-42MUSTBoth devices buffer events they cannot transmit and upload them on reconnection.
SP-43MUSTEvery buffered event be labelled with its true age when delivered.
SP-44MUST NOTA buffered event be presented as current.
SP-45MUSTLoss of all received signals in a location with prior coverage be recorded as an event in its own right.
SP-46MUST NOTAny implanted device be used. A passive implant has no power, no positioning and no radio range, and cannot perform this function.
SP-47SHOULDThe child’s band achieve a five to seven day charge interval; a band charged daily will not be worn.
SP-48SHOULDThe child’s band carry no screen, name or branding that marks the wearer out.

7. Power and availability

Argued at /operations/

RefLevelRequirement
SP-49MUSTThe charging obligation be stated in the subject’s conditions with a specific window and duration.
SP-50MUSTA dead battery on the subject’s device be defined in the order as a compliance event with a written response.
SP-51MUST NOTA dead battery on a child’s band ever be treated as a violation by anyone.
SP-52MUSTBoth devices warn before power loss, and the guardian be told when a band goes offline.
SP-53SHOULDRepeated discharge patterns be addressed through supervision rather than through the device.

8. Due process, audit and exit

Argued at /due-process/

RefLevelRequirement
SP-54MUSTEvery tier 3 and tier 4 event produce a record the subject is entitled to obtain, including the classifier inputs that produced it.
SP-55MUST NOTThat record disclose the child’s identity or location history.
SP-56MUSTA written challenge route exist with a deadline binding on the authority as well as on the subject.
SP-57MUSTFix quality, device health and firmware version at the time of an encounter be disclosed to a subject contesting it.
SP-58MUSTAn overturned alert be removed from the repeat count that feeds the ladder.
SP-59MUSTEvery read of the encounter store be written to an append-only audit log.
SP-60MUSTA subject be able to obtain their own portion of that audit log.
SP-61MUSTA petition for removal exist, with published criteria and published grant rates.
SP-62MUSTRecords be destroyed on exit, with a certificate issued to the subject.
SP-63MUSTAlert accuracy be published in aggregate at least quarterly, classified by cause.
SP-64SHOULDMedian time from alert to human action be tracked weekly as the programme’s leading failure indicator.

9. Operations

Argued at /operations/

RefLevelRequirement
SP-65MUSTFirmware be rolled out in stages and the version recorded against every encounter.
SP-66MUSTStaleness of a band’s copy of the key list be a monitored condition rather than a silent degradation.
SP-67MUSTEnrolment, tier, term and audit history transfer with a subject who moves between jurisdictions.
SP-68MUSTThe programme end for a subject who moves somewhere the order is unenforceable, and say so rather than claim a coverage it no longer has.
SP-69SHOULDFleet health — reporting, silent over 24 hours, strap events — be reviewed daily.

Open questions this document does not answer

A specification that claims to have settled everything is not finished, it is unexamined. These are open:

  • Radio choice. Bluetooth LE gives range and battery life but is trivially detectable; ultra-wideband gives far better distance accuracy at a power cost a small band may not survive. The dwell thresholds in this document assume distance accuracy the radio may not deliver, and that assumption has not been tested.
  • The proximity oracle. Local matching requires publishing keys, and publishing keys lets anyone build a detector. Regional partitioning and attestation narrow it. Nothing eliminates it without moving matching to a server, which reintroduces the central location database. Whoever adopts this has to choose knowingly.
  • Whether the child-side device is adopted at all. Every figure in the coverage model is multiplied by the share of children carrying a working, charged band. That share is not a technical parameter, and no amount of engineering sets it.
  • Cost. Nothing in this document is costed. Hardware, network, monitoring centre staffing, hearings and appeals are all real and none is estimated here, because an invented number would be worse than an absent one.
  • Whether the assessment instrument is good enough. The tier feeds the ladder. Validated actuarial instruments have known and published error rates, and this design has not been tested against what those errors do to alert quality.