Watch good teaching. Get evidence back on your own. Have someone you can admit confusion to. None of the three were available inside a school.
WHO GETS WHAT
The four people from the front page, and which part of Ranius works for them. Same mechanism, different doors.
Children out of school, or not learning to read
Lessons, published
Every lesson is open, so where you live and what your school can manage stop deciding what you get. A few hours a week is enough to start, and children pick a teacher who fits. Being enrolled somewhere is not a condition.
People who hold a licence but do not teach
Somewhere to teach at all
Until now, almost the only route back was a full-time school post. On Ranius you can teach once a week while keeping the job you have. A licence is not an entry condition — but the experience behind it obviously counts.
People with no licence who want to teach
The route from applying to certified
A way in that is not four years to a licence. Apply → train → coach → certify, in about three months. You can start the moment you want to, and stopping partway is designed to cost you little.
Teachers already inside a school
Assessment, and a staff room
Timestamped evidence and the children's voice come back on your own teaching. A team of four to eight, every week, with teachers around the world. The only comparison is your own past — we never rank teachers.
The four are illustrative composites, not real individuals.
BECOMING A TEACHER
We keep writing "apply → train → coach → certify, about three months" across this site. Here is what is actually inside it.
Apply
Tell us the subject you want to teach and the hours you can give. We ask whether you hold a licence, but it is not a condition of entry. No CV.
To pass Nothing. Everyone who applies moves on.
Train
Watch a published lesson with one dimension chosen in advance. Record your own lesson against that same dimension. AI returns what happened and when, second by second. Watch, do, get it back — repeated.
To pass Until you meet the rubric. There is no cap on attempts.
AI carries this part. It is where a shortage of people used to stop everything.
Coach
A team of four to eight meets every week. You bring the lesson that did not work. The point is to have someone you can admit confusion to.
To pass Nothing. This exists to stop people leaving, not to filter them out.
This part is never given to AI. Handing it over would empty it of meaning.
Certify
Two external raters watch a lesson against the same rubric used in training. The reason for the decision comes back as timestamped evidence.
To pass Both raters at or above the bar. Only then do you stand in front of children.
AI takes only the part that a shortage of people used to block. Someone to admit confusion to, and the certification decision, stay with people.
Durations and thresholds are design values. Whether this route produces teaching equivalent to a licensed teacher is what the December 2026 A/B measures. How many teachers one mentor can carry is being measured now.
HOW YOU LEARN
We keep saying "watch, respond, hold each other up" without saying what you actually do. Here is one week of training, as it runs.
Watch
Pick one published lesson. You choose the dimension before you go in — "how much of the time children are talking", "what happens after a child says they do not understand" — one dimension, the one you will try that week.
Why Watch a lesson with no dimension in mind and all you keep is "that was good". Narrow it to one and it becomes something you can try yourself.
Do
Try that same dimension in your own lesson. The lesson is recorded. It is fine if it does not work. This step collects lessons that were attempted, not lessons that went well.
Why Watching alone changes nothing. Once a week, you do it yourself.
Get it back
AI watches the recording and returns timestamped evidence against the dimension you chose. Not a score — when something happened, and what. The judgement stays with you.
Why "Wait longer" is not something you can act on. "You waited three seconds" is.
What the AI returns (example)
08:12 A child said they did not understand. Three seconds later you gave the answer.
19:40 Same situation. This time you waited twelve seconds, and another child began explaining.
whole Children talked for 18% of the lesson (last time 12%). The comparison is your own previous lesson.
Talk
The team meets. You bring the lesson that did not work. You do not walk through the AI notes — you talk about why you did what you did. Nobody hands out answers.
Why Leave this out and people quit. The second most common cause behind teachers taking mental health leave is relationships at work, at 23.2%, and 63.4% of them had been at that school under three years. Having nobody to admit confusion to bites before any question of ability.
Ask the children
The children who took the lesson answer a few questions. Not to rank teachers — to pick up what an adult watching cannot see.
Why A single classroom observation has a reliability of 0.14–0.37: the same lesson gets a different verdict depending on who watches. Student surveys sit at 0.66, steadier than adult observation.
What is a "dimension" (rubric), really?
The lens you watch through — "today I only look at this". It is not a score sheet. The same one is used in training and in certification, so nobody gets failed against a bar they never saw. Its content is what we are working out with an assessment specialist.
Does AI do everything?
No. AI carries only the watching and the timestamped response — the part a shortage of people used to block. Someone to admit confusion to, and the certification decision, stay with people. Hand those to AI and the job stops being one people can stay in.
Are teachers ranked?
No. The comparison is your own previous lesson. The moment teachers are lined up against each other, nobody brings the lesson that went badly. And if nobody brings it, none of this works.
How many hours a week?
30 minutes watching, one lesson of your own, 60 minutes with the team — two to three hours a week. Small enough to sit alongside another job. A design value; we measure it in December.
You repeat that week for about three months, up to certification. None of it is exotic. It is what a school could never make time for, moved outside and actually run.
Sources: MET Project (single observation reliability 0.14–0.37; student survey 0.66) / MEXT FY2024 personnel administration survey (second cause "workplace relationships" 23.2%; 63.4% of those on leave were under three years at that school). Days, durations and example dimensions are design values, measured in the December 2026 A/B.