A transcript says where a child studied. Not what they were taught.
When a child joins from another school system, age and the transfer certificate fix the year. The course, the set and the support inside it are decided by a person who cannot see what the child was taught. We compare the two curricula, outcome by outcome, so they can.
The year is rarely in question. The courses are.
Schools place a child by age and the records from the last school. Students complain when they are put into a course they have already had; a gap in maths compounds, and at secondary level the wrong course can close a university programme years later.
Fixed by age and records.
In Dubai and Abu Dhabi the transfer certificate fixes it by rule, and moving a child off their year needs the regulator’s approval. Elsewhere it is the school’s own policy.
Decided by the school, without seeing what the child was taught.
A placement test shows what the child can do on the day; the records show subject names and grades. This is the decision the comparison informs.
A file for the regulator or the Head.
Where a change of year or of curriculum needs approval, the comparison is evidence for the school’s file.
First-hand at the international school where this started, and colleagues at other schools report the same. How often it happens has not been counted yet: the pilot counts the decisions changed or confirmed.
One sending curriculum, compared with the school’s own, outcome by outcome.
Each outcome is read in its own language and kept in the words of its official document: an action, a concept and its limits. Each gets its counterpart in the school’s curriculum, or “no counterpart found”, with the evidence. Built once, it serves every child who joins from that curriculum.
решити · системи линеарних једначина · највише три непознате
επιλύειν · γραμμικά συστήματα · τριών εξισώσεων με τρεις αγνώστους
Two scripts, no shared word. Both independent runs paired these two outcomes, and our reviewer, marking blind, judged them the same outcome. A person checks every pair before a corridor is used, and a person at the school decides.
The placement itself; a claim that a child has not learned something; a single overlap score.
Built, measured, and checked by one reviewer so far.
The chain runs end to end, from the official documents to the compared outcomes, with ordinary code checking each step. Its model calls are pasted by hand today; automating them is the next build.
The reviewer so far is our cofounder. Next: two outside raters, one for maths and one for history, marking blind, and a test against a general AI model given the same curricula, scored by a rule fixed in advance. “No counterpart found” has not yet been checked by a person, so we never claim a child has not learned something.
Large international schools and school groups first.
Curriculum as source code A curriculum document is parsed into its real structure, then reduced to the smallest independently teachable claims it makes: each one an action applied to a concept under stated limits.
Original language throughout Nothing is translated during extraction. An English reading is kept beside the original for reviewers, never in place of it.
Checked at every step Each model call runs on its own, and ordinary code checks its output before the next step. Any step that can be code rather than a model call is code, and each output points back to the sentence it came from.
Matching Independent runs of a matching model compare the two lists of outcomes, and code combines their answers. A person reviews every proposed pair; where the runs disagree, it is flagged rather than quietly resolved.