Boom AI
Evidence review · for instructors
One exit ticket in. One human decision out.
Boom reads the work a student already turned in, shows the evidence behind every interpretation of it, and suggests what to try next. Nothing changes for that student until the instructor decides.
Fictional students and college. No sign-in, no setup.
Item: 1/3 + 1/4, show how you know
What the student wrote
“1/3 + 1/4 = 2/7 because 1+1=2 and 3+4=7”Adds numerators and adds denominators as separate whole numbers.
Evidence: the answer is 2/7, and the student’s own explanation describes the add-across procedure.
How it works
Three moves, one class period.
Boom sits on work you already collect. Collect it, read the interpretation, make the call — then the class gets the reteach it actually needs.
01 · Collect
Work you already have
An exit ticket, a quiz item, a short constructed response. Boom records the setting, the accommodations, and the objective the item is meant to measure.
02 · Interpret
Claims with their evidence
Each response is broken into the features that matter, then matched against documented misconceptions and prerequisite gaps — with confidence, alternatives and what would reduce uncertainty.
03 · Decide
The instructor’s call
Agree, disagree or annotate. Approve a suggested next step, or write your own. The decision is logged, and the agreement rate stays visible to you.
Features
Built to be checked, not trusted blindly.
Every part of the review shows its work: what the student did, what Boom thinks it means, how sure it is, and what a person should do about it.
Objective-anchored reading
Each item is tied to one objective and its success criteria, so a look at the work answers a specific question instead of a vague one.
Interpretation, separated
The student’s response is stored as-is, then broken into evidence features. Interpretations point at those features — never at the answer alone.
Stated confidence
A confidence level, the reason for it, alternative explanations, and the uncertainty flags that apply: one data point, limited work shown, collection conditions.
Next steps that fit the period
A probe question, a small-group reteach, a contrasting example — each with grouping, timing and minutes, and a rationale tied to the evidence.
Agree or disagree
The instructor accepts or rejects each interpretation and suggestion. Disagreement is the interesting data, and it is kept.
Decision audit and record
Every decision lands in an audit trail, next to a JSON record of the whole review you can copy, share or archive.
The record
Plain data, not a black box.
Every review produces one open record: the source, the item, the response, the features, the interpretations and the suggestions. Nothing about a student is hidden inside a score.
- Source and conditions travel with the evidence.
- Model name and version are recorded on every interpretation.
- Decision history is kept with who made the call and when.
{ "objective": { "code": "MATH050-SLO3", "statement": "Add fractions with unlike denominators" }, "response": { "content": "1/3 + 1/4 = 2/7 because 1+1=2 and 3+4=7", "correctness": "incorrect", "features": [["f1", "Answer is 2/7"], ["f2", "Adds numerators and denominators"]] }, "interpretations": [{ "claim": "Adds numerators and denominators as whole numbers", "category": "misconception", "features": ["f1", "f2"], "level": "moderate", "p": 0.7, "uncertainty": ["single_data_point", "limited_work_shown"] }], "suggestions": [{ "action_type": "probe_question", "minutes": 3, "timing": "next_lesson" }], "decision": { "by": "instructor", "status": "pending" } }
Trust
Support for the instructor, not surveillance of the student.
Boom reviews work an instructor already assigned and already owns. It does not grade, does not discipline, and does not contact students on its own.
A person decides
No interpretation, suggestion or message reaches a student without an instructor decision.
Read-only
Boom does not write back to the gradebook or the student system. Records are inputs, not edits.
Evidence on the record
Each claim points at the features of the work it came from, with the uncertainty that applies.
Disagreement counts
How often the instructor agrees with Boom is measured openly — including the times it is wrong.
Try it on a fictional class.
One exit ticket, three students, six interpretations. Agree, disagree, and watch the audit trail fill in.