A single-classroom pilot is the smallest honest way to find out whether this work belongs in your building. One teacher, one section, one term. That is the whole shape. This page tells you what a leader should actually expect, week by week, and what has to be true at the end for anyone to continue.
What a pilot is, and what it is not
A pilot is a supervised experiment inside one classroom, run by the teacher who owns that classroom, with a shared log the principal or department head can read at any time (Class C). It is not a vendor deployment. It is not a data pipeline. It is not a promise that scores go up. It is a structured attempt to see whether active inference (as a learning lens; Parr, Pezzulo, and Friston, 2022, Class E) and the teacher workbench (as a preparation tool for staff, not a student-facing chatbot) fit the way this particular room already works.
The teacher stays in charge of pedagogy. We stay in charge of honesty scaffolding and workbench operation. The leader stays in charge of scope, cadence, and the go, no-go call at the end.
The shape of one term
A term-length pilot is roughly twelve weeks, structured in four visible phases.
Weeks 1 to 2, framing. The teacher and one of us read the existing unit together. Nothing changes yet. We produce a one-page map: where an active-inference lens might help students name their own uncertainty, where it would be a distraction, and where the current plan is already doing the work. If the map says "leave it alone," the pilot ends there and the invoice reflects the review alone.
Weeks 3 to 5, honesty policy. Before any tool touches student work, the class agrees on the AI-authorship fence: what the workbench may draft (feedback templates, worked examples for the teacher, rubric prompts) and what it may not touch (student writing, student names in prompts, grades). The fence is posted in the room and signed by the teacher. See the pilot's honesty scaffold in Honesty in Classrooms, the AI-Authorship Fence.
Weeks 6 to 9, live use. The teacher uses the workbench two or three times per week for lesson preparation and feedback drafting. Every use lands in the shared log with a timestamp, a purpose line, and a note on whether the output was kept, edited, or discarded (Class C). Students never see the workbench directly. What they see is a teacher whose feedback got faster and more specific, and a classroom vocabulary that now includes evidence classes and gates.
Weeks 10 to 12, honest write-up. The teacher writes a two-page reflection. What worked, what did not, what would need to be true to continue, and what would kill it. We add a matching two pages of log summary. The leader reads both, together, in one sitting.
What to expect, honestly
Realistic outcomes at the end of one term look like this.
- Feedback turnaround shortens, usually by a meaningful margin the teacher can feel, though the effect is uneven across assignment types (Class C).
- Students start using two or three pieces of active-inference vocabulary correctly, most often "prior," "update," and "evidence class." The vocabulary sticks because it is useful, not because it is drilled.
- The teacher reports higher preparation quality and, often, lower preparation load. Sometimes the opposite: better prep costs more time in weeks 6 and 7 before it saves any.
- Test scores are not the target and are not claimed. Any correlation reported in the write-up is tagged Class C or Class U and framed as a signal to watch, not a result.
What you should not expect: a transformation, a headline number, a case study we can sell to the next school. Pilots that look like that in month three are usually pilots someone is spinning.
Gates for continuation
At the end of the term, four gates decide whether anyone continues. All four must clear.
- The teacher wants to continue and can say, in one paragraph, what specifically they want to try next.
- The shared log shows no fence violations, or shows violations that were caught, logged, and repaired within the same week (Class C).
- The leader, reading the write-up cold, can restate the pilot's honest limits in their own words.
- Nothing in the reflection points to student harm, staff burnout, or public credibility risk.
If any gate fails, the pilot ends cleanly. We document why. That documentation is part of what you paid for.
Where to read next
- Partnership With Schools, What SolutionWright Does and Does Not Do: the pillar page for scope and lines.
- Cohort Pricing and How It Actually Works: the money side, without a sales pitch.
- How We Measure a Pilot Honestly: the specific instruments and their failure modes.
- /workshop: if you would rather start with a teacher cohort before a classroom pilot.