PROJECT H.
A longitudinal behavioural inference engine with cryptographic audit properties.
In plain terms: it watches how a person learns, through everyday conversation and the things they make, and builds a private, tamper-proof record of how they grow. No grades. No scores. The record is theirs.
Not “who should this child become?” but “what is this child already remembering?”
What this is.
A system that observes how one person learns, through everyday conversation and the things they make, and turns it into a private, verifiable record of how they grew. It runs today for one eleven-year-old, home-educated across two countries.
Parents and educators who want a real account of a child's development instead of a report card. Developers and researchers who want the engine underneath it: open, forkable, and honest about its own uncertainty.
A grade flattens a person into a number, and the number outlives the person it described. This measures richer things, holds them as narrative and evidence, and hands the whole record to the learner.
One boundary. Everything hashes at the door.
Everything a learner says or makes enters here, and gets fingerprinted on the way in so it can never be quietly altered later.
Forty small evidenced decisions. Not one giant guess.
Instead of one AI guessing at everything at once, many small judges each answer a single question, and only when the learner's own words prove it.
An open vocabulary, canonicalised by meaning.
Skills are not picked from a fixed list. The system names what it sees in the learner's own language, then merges the ones that mean the same thing.
Skills and values are living dimensions, not a fixed list. The engine names what it observes in the learner's own language, then collapses synonyms into canonical concepts, so persistence and perseverance become one node. A new concept mints as provisional and promotes to stable at three distinct observations. Embeddings run locally, in-process, 384 dimensions: nothing leaves the box.
Design-spec canonicalisation: attach when cos(e_new, e_existing) ≥ 0.86, else mint provisional. As-built, an LLM matcher performs the collapse by meaning.
25 tenets. Zero scores.
Twenty-five human qualities, followed over years. They surface as stories and evidence, never as a number, a grade or a ranking against anyone else.
One root. Change a byte, break the proof.
The record can be proven untampered years later, the same way a bank or a browser proves a certificate, without anyone having to take it on trust.
Hers for life. Verifiable offline, forever.
Learner-owned. Private. Exportable.
The learner holds the record, not a school or a company. It exports in open credential formats and can be checked by anyone, offline, with no account and no permission.
The engine knows no tenet names.
The machinery and the philosophy are separate files. The engine measures whatever a config tells it to, so your values are a document you write, not code you fork.
One engine. Any philosophy.
One family's philosophy runs on it today. Anyone else's could: a forest school, an adult learning to paint again, a team onboarding. Same engine, different beliefs.
Bring your own pack.
booting…