Thinking Types

The Problem

Most classroom analytics tell teachers how much a student knows. They don't tell teachers how a student thinks. A student who gathers evidence from five sources but never synthesizes them needs different support than one who builds rigorous arguments but never considers opposing views. Existing tools flatten these differences into a single score.

Three Dimensions

Thinking Types measures every student along three axes, each derived from signals already present in mCLASS reading assessments and Amplify ELA essays. No new assessments required.

The Eight Types

Three dimensions with two sides each produce eight thinking types. Each type has genuine strengths and growth areas — no type is better or worse. Students in the boundary zone (4.0–6.0 on any dimension) may show traits of two types.

Knowledge Graph

Beyond thinking style, the dashboard maps each student's mastery across a graph of 15 reading, writing, and higher-order thinking skills. Skills are connected by prerequisite relationships — so teachers can trace a weakness in inferential comprehension back to its root in vocabulary or literal comprehension.

A type-comparison view lets teachers see how different thinking types perform across skills. Architects and Detectives tend to lead on depth-related skills; Explorers and Wanderers show more breadth but less structure. These patterns guide grouping and differentiation decisions.

What the Dashboard Does

• Overview — class-wide distribution across eight thinking types
• Student profiles — per-student dimension scores, type history across units, and skill mastery bars
• Grouping — suggested peer review pairs based on complementary thinking styles
• Alerts — flags depth decreases, stable patterns, and positive growth across units
• Knowledge graph — skill tree with class-level and per-type mastery views

Technical Stack

Built with FastAPI (Python), SQLite, Jinja2, and vanilla JavaScript. No frontend framework — the entire dashboard is a single-page app rendered with template literals and direct DOM manipulation. Deployed on Render.

The data pipeline extracts signals from ORF (oral reading fluency), MAZE (reading comprehension), essay text analysis, and AWE (automated writing evaluation) scores. A weighted scoring engine computes the three dimension scores, and a classification algorithm assigns thinking types with confidence levels and boundary detection.