The engine ranks and shows its working. Every candidate arrives with its score broken into named dimensions, one sentence you can read aloud, and — if it is blocked — the reason it is blocked, still on the screen.
Open any score and it breaks into the dimensions that produced it, each with its own points, its own ceiling and the detail behind it. Then it says the two things a number cannot:
Strongest signal: a son the same age who plays in the same league. Watch: both parents work late Tuesdays — and he told us he values shared dinners.
A recommendation with no caveat is one nobody can argue with, and this is a decision about where a fifteen-year-old sleeps.
3.08 The score opened and argued with. Weights are tunable per organisation — what matters to you is not what matters to the programme down the road.
No black box and no “proprietary algorithm”. These are the dimensions for a host family, with the default weight each one carries. Your organisation can retune them.
An allergy, a full school, a smoking household against a non-smoking student: these are not worse matches, they are wrong matches. So they are never quietly discounted into a lower score — they are marked blocked, with the reason, and they stay visible.
And you can still choose them. Staff who know something the database does not are the reason the override exists — on the record, with a sentence.
Seven blocking reasons for a family, each with its own stated sentence: pet allergy, a pet-free request, smoking household, no private room, a diet the family cannot cater for, no open bed, and a commute past the family’s own limit.
Choosing anything other than the top suggestion requires a reason. Not as a UI nicety — the database refuses the write without one. Every committed step is stored with the score at the time and a frozen snapshot of the alternatives it beat.
The placement, the decision with its frozen alternatives, a timeline event, and an audit entry.
The student, the host family and the coordinator are notified in the same transaction.
They are a corpus a human reads before any weight changes.
The application exists and the student is an applicant.
Accepted by the programme, not yet in matching.
In the workspace. This is the column with a real work-in-progress limit.
A family is assigned and the offer is out.
Everyone has said yes.
Flights are booked; the student has not landed.
In country. Where the care work happens.
Seven detectors run over your own data. A sparkline whose numbers were typed by whoever built the screen is worse than no sparkline — it teaches you to trust a shape that means nothing.
A signal, a clear, or dormant — the honest third state, where the detector says which data it needs, that you are not recording it, and draws nothing.
“Students nobody has spoken to in a while” uses your organisation’s own median gap, not a number from a manual written for somebody else.
Every signal names the pattern in one line a person would say aloud, shows the window it looked at and the exact rows behind it, and offers one move. It is class C0 — a notice. It never acts and never mails anybody. It closes itself when the trend reverses, because the signals are recomputed on every read.
Placement & Care and Network include 30 days beside your current process. Access does not include placement operations.
