The Evidence Engine · under the hood
Exactly how the guide learns — and protects you while it does.
For the people who like to know how the engine runs: here's the real architecture. Kaiārahi learns from language and choices, never from your private life — designed with Aotearoa's Algorithm Charter principles and published Māori data-governance guidance in view from the first line of code. Our own formal reviews — a Privacy Impact Assessment and a Māori data-governance review — complete before any demographic collection begins.
01 · The pipeline
Every interaction runs the same gauntlet before a single thing is kept. Most of the time, nothing is kept at all.
The evidence record never holds your raw prompt. What's kept is redacted text, a one-way pseudonymous ID, and a keyed hash — never your name, never your identity. Capture only ever runs for a learner with a current consent on record — on by default for adult accounts and off in one tap, and opt-in only for 16–17s and the te reo / Tūahiwi programme. Two things are deliberately different, and we would rather tell you than let you find out: the work you save to My Playbook, and your answers during an assessment run, are kept as you wrote them — that is what makes them any use to you. Neither becomes part of the evidence record. Your saved work is yours to delete, piece by piece or all at once, whenever you like; your assessment answers go if you withdraw from the research; and everything goes when you close your account.
02 · The guardrails
These are written into the code as non-negotiable — they beat build convenience, every time.
Capture runs only when it has been deliberately switched on — a governance decision, not a code default — and it fails closed if anything it depends on is missing.
Nothing is captured for anyone without a current, active consent on record.
Personal details are stripped before anything is stored or scored. It over-redacts rather than risk a leak.
Records carry a one-way ID, never your email or identity. They can't be traced back to you.
Devices get passed around. One tap says "not me right now" and capture pauses — because attributing someone else's work to a learner would corrupt the very thing the evidence is for. Nobody is logged out, and we never ask who the other person is.
A programme cohort's records are New Zealand–resident data and live on our own Hearth — encrypted before anything is written down, decrypted only on shore, never parked overseas "for now". Built and tested end-to-end — including the erasure that runs when someone leaves.
✦ Designed with the Algorithm Charter for Aotearoa and published Māori data-governance guidance in view — our own formal reviews complete before demographic collection begins.
“We measure capability, not people. We learn from language and choices — never from a person’s private life — and what we’re trusted with, we hold with care.”
The Kaiārahi data covenant
03 · The measure
Not clicks, not time-on-page. We look at the quality of a person's thinking with AI — across four dimensions.
Goal clarity, context, constraints — a clear, well-shaped ask.
Iterating, breaking a problem down, asking good follow-ups.
Does the prompt actually set up a genuinely useful result?
Privacy, checking the answer, cultural and ethical care.
Scored blind to who you are, and human-reviewed — never silently by a machine. It never rewards length, and never penalises plain English or te reo Māori. Capability, not verbosity.
04 · The journey
Your level is earned, not bought — and it's what opens up the create tier as you grow.
Your band is built from real activity — experiments run, things made — and, where it exists, a measured capability score from the reviewed index above. The measure leads; activity keeps it moving even before anything is scored.
05 · Where it lives
It runs responsibly in the cloud today, and it's built to come home.
Runs on Cloudflare with data minimised, redacted and pseudonymous. A conscious, careful interim — never the destination.
Local-first, sovereign hosting. The whole store moves home by design — one clean swap, no scramble. Data stays where it belongs.
✦ Māori data is handled with care at every step — never a commodity, never sold.
06 · Why it matters
Careful measurement isn’t just good housekeeping — it’s how we show, with real evidence, that this actually works.
When we can demonstrate that ordinary people — not the already-confident — walk out genuinely more capable with AI, we can bring Kaiārahi to far more of them: through schools, workplaces, and the kind of public and government programmes that reach people who’d never otherwise get the chance. Your practice is the evidence that opens those doors — and because it’s gathered cleanly, it stands up to scrutiny.
Measured growth in how well people brief, judge and direct AI — not vanity metrics.
Proof that people come back from feeling left behind — the conviction this whole kaupapa is built on.
Evidence gathered to a standard that holds up to government and academic scrutiny.
No black box, no hand-waving — just careful, accountable work in service of helping people. If you'd like to be part of it, you'd be in good hands.
Our promise is simple: we’ll only ever ask for what genuinely helps, we’ll guard it with real care, and if you switch it off you lose nothing. That’s the deal — and we’ll keep it.
Kaiārahi