Seedlabs

BreakPoint: a perturbation-task generator for breaking stable wrong models

A teacher picks a target concept and the wrong-but-stable model students hold; the tool generates a short task engineered so that model produces a confident, visibly wrong prediction — the moment the student's own construction breaks and can be rebuilt.

Social SciencesEducational theories and practices
Career Development and DiversityMath and science curriculum teams building conceptual-change activities targeted at documented misconceptions.

Concept

If, as [0] argues, knowledge is constructed rather than received, then the lever for change is viability: a student keeps their current construction until it stops working for them. Re-explaining cannot dislodge it, because the model was never built from explanation in the first place. What dislodges it is a lived contradiction — the student's own model making a confident prediction that their own check then refutes.

BreakPoint is a generator of 10-minute perturbation tasks. Input: a target concept and a named construction (chosen from a built-in library of documented models, or imported from a diagnostic). Output: a ready-to-run task with four parts — (1) a prediction prompt that forces the student's model to commit publicly ('will 0.5 x 8 be bigger or smaller than 8? write your guess'); (2) an execution step where the student's own method produces the answer; (3) the confrontation ('you predicted 40, you got 4 — what happened?'); (4) a rebuild prompt asking them to state the revised rule in their own words. Each task ships with the predicted wrong answer and the exact follow-up question, so the teacher knows in advance what to look for. The wedge is one thing only: engineered breakdown of a specific documented construction, not general 'engaging activities.'

Why now

[0] establishes that the student's construction — not the delivery — is the thing that changes, and that the criterion-referenced testing apparatus it criticizes measures outcomes while ignoring constructions, which is why existing materials contain so few engineered perturbations. [1] supplies the mechanism the tasks operationalize: Piagetian disequilibrium and accommodation — a construction reorganizes when it fails, not when it is contradicted by authority. [6] shows mathematics is ill-structured from the learner's perspective, so stable alternative models persist exactly where schools assume a single right answer, making math the natural first domain. [4] shows classrooms are already saturated with evaluation; BreakPoint converts existing evaluation moments into perturbation moments rather than adding new classroom time.

Written by GLM · 13 Oct 2026

AI assessment

Backed by 4 papers84

A sharp, pedagogically grounded tool that converts diagnostic data into actionable 'cognitive conflict' tasks to break stable misconceptions.

Evidence strength
4/5
The idea strongly operationalizes Piagetian disequilibrium and constructivist theory cited in the research, moving from theoretical 'knowledge construction' to a specific instructional mechanism.
Market pull
4/5
Curriculum developers (like PhET or Amplify) have a high incentive to build 'conceptual change' activities, as these are the gold standard for deep learning in STEM.
Novelty & moat
3/5
While it complements ModelDig (which diagnoses the model), BreakPoint focuses on the active 'breaking' and 'rebuilding' phase, filling a gap in the intervention cycle.
Feasibility
5/5
The MVP is a structured template generator; it does not require complex AI, only a library of documented misconceptions and a logic-based task generator.
Wedge clarity
5/5
The wedge is extremely specific: a 10-minute perturbation task for a single documented construction, avoiding the 'platform' trap.
Simplicity / focus
5/5
The product is focused on one single output (the perturbation task) with a clear four-step sequence.

Scored by Gemma 4 · 13 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • PhET Interactive Simulations, University of Colorado Boulderorganization

    Its simulations already create discrepant events; a generator that pairs each sim with the documented constructions it can break extends their classroom usability with ready-made scripts.

  • ExploreLearningcompany

    Its Gizmos simulations would gain prediction-confrontation-rebuild scripts per misconception, deepening how teachers actually use the sims beyond demonstration.

  • Dan Meyerindividual

    His three-act-task format is the hand-built version of this exact wedge; a generator scales the format across the hundreds of concepts he cannot hand-craft.

  • Amplifycompany

    Its Desmos-based math curriculum is built around task design; generated perturbation tasks feed that pipeline at lower production cost per concept.

Research it builds on

  1. Educational Psychology: Theory and Practice
    Robert E. Slavin · 2025 · 1348 citations
    All ideas from this paper →
  2. Treating Mathematics as an Ill-Structured Discipline
    Lauren Β. Resnick · 2025 · 120 citations
    All ideas from this paper →
  3. Social Comparison and Education
    John M. Levine · 2026 · 95 citations
    All ideas from this paper →
  4. Learning as a Constructive Activity
    Ernst von Glasersfeld · 2026 · 84 citations
    All ideas from this paper →
Leave feedback
feasibility