STEM PRACTICE AND EVIDENCE WORKBENCH
Audit structured prompt evidence for reasoning, strategy comparison, open response, and connection to peer contributions.
Review output
Load the worked example to inspect realistic inputs, or enter anonymous classroom or laboratory evidence. Results remain in this browser tab until you copy or download them.
Transparent method
Feature score equals the number of yes entries among open response, reasoning demand, multiple strategies, and peer connection. Strong means score is at least the entered minimum; otherwise the local model says review. Every feature remains visible, and prompt text is never silently interpreted by a hidden language model or keyword list. Supported means the entered strong-or-review claim matches the disclosed feature count. The result describes planned opportunity, not actual discourse quality. Review the mathematical goal, expected reasoning, teacher moves, language access, wait time, norms, representation options, response patterns, and whether all learners can enter and extend the conversation.
Confirm the question, source, units, notation, response coding, model assumptions, and review threshold. Keep an unchanged copy of the original evidence.
Run the worked example, inspect every intermediate row, and compare it with a hand calculation, diagram, representation, or approved subject method.
Record the method, assumptions, units, classifications, and flagged rows. Verify the editable report independently before educational or laboratory use.
How to interpret the result
Use while reviewing a set of planned mathematics prompts for varied opportunities to explain, compare, justify, and respond. The analyzer applies structured reviewer coding rather than a keyword score. It does not listen to classroom talk, assess mathematical correctness, measure participation equity, infer cognitive demand, or judge an individual learner.
The compare-methods prompt contains all four entered features and the explanation prompt contains three, so both strong claims meet a threshold of three. A calculate-only item has no listed discourse features and its review claim is supported. The final closed prompt contains only reasoning and strategy flags, so its strong claim is flagged.
Feature score equals the number of yes entries among open response, reasoning demand, multiple strategies, and peer connection. Strong means score is at least the entered minimum; otherwise the local model says review. Every feature remains visible, and prompt text is never silently interpreted by a hidden language model or keyword list.
Supported means the entered strong-or-review claim matches the disclosed feature count. The result describes planned opportunity, not actual discourse quality. Review the mathematical goal, expected reasoning, teacher moves, language access, wait time, norms, representation options, response patterns, and whether all learners can enter and extend the conversation.
Do not mark a prompt open merely because it is long, count a request to show work as genuine reasoning automatically, assume different answers imply different strategies, require public speech as the only participation route, or use a high feature count as proof of productive discourse. Inspect enactment and evidence from learner thinking.
Learning boundary
This browser-local workbench applies only the disclosed rules. It cannot diagnose a learner, certify an experiment, or replace teacher, laboratory, and safety judgment.
Review checklist
Practical questions
No. Processing occurs in the current browser tab. Avoid personal information and retain authoritative records only in the approved school system.
No. It describes entered evidence under one disclosed rule. Review reasoning, uncertainty, access, and several evidence sources.
Treat it as an editable calculation aid. Show your own reasoning, cite required sources, and follow course and academic-integrity rules.
YOUR NIRMION
