Referral evidence
Current, proportionate information approved by the responsible authority.
CareMatch structures the evidence needed to decide whether a home can safely meet this child’s assessed needs. Only eligible homes receive a comparative fit score. It does not make the placement decision.
A referral records age, placement type, needs, current risks and safeguards, education, therapeutic and cultural needs, family contact, location, accessibility and required staffing. A verified home profile records its registration, vacancies, care model, environment, workforce, transport, location and pseudonymised existing-child compatibility information.
Current, proportionate information approved by the responsible authority.
Timestamped capability and vacancy information supplied by an authorised provider.
Only necessary pseudonymised risks, vulnerabilities, triggers and safeguards are shown. Identities are never shared.
Before ranking, CareMatch asks whether this home can safely meet this child’s needs in everyday care and foreseeable crises without creating an unacceptable risk to the child or the existing group.
Suitable, Suitable with conditions, No-go, or Insufficient information. Missing evidence pauses the decision; it is not treated as proof that a home is unsafe or safe.
This design follows the official principle that a registered person should accept a placement only when the home can respond effectively to assessed needs after considering the impact on children already living there. See the Department for Education guide to the Children’s Homes Regulations and Ofsted’s children’s homes inspection framework.
A score cannot compensate for a failed gate. Homes marked No-go or Insufficient information are not ranked. The starting weights below are a pilot configuration and must be governed, versioned and tested.
For each eligible home, CareMatch adds each factor score multiplied by its active weight, then divides by the total weight. The result is a comparative 0–100 score. It is not a probability of placement success.
The referral and home snapshots, rule version, factor scores, ranking, exclusions, user, timestamp, final decision and override reason are retained as an append-only history.
Acceptance, start date, duration, stability, incidents, breakdowns, reasons, successful move-on and commissioner/provider feedback are stored in a structured outcome record. The current MVP does not automatically learn from or change its rules using that data.
Any later machine-learning model would require a separate lawful basis and DPIA, representative data, outcome definitions, bias and performance testing, independent validation, change control and ongoing human oversight.