Inspiration
A physics lab can go wrong before the graph is drawn: a length copied in metres instead of centimetres makes otherwise sensible readings look impossible. Lab Detective addresses this school-life problem by helping students question a measurement without hiding or rewriting it.
What it does
Students enter pendulum lengths, elapsed times, units, and the number of oscillations. The app calculates period T = t/N, plots T or T² against length, fits T² against length, and estimates g from a positive fitted slope. Deterministic physics checks flag readings worth reviewing and explain the evidence.
The demo deliberately pairs an 80 m entry with timing consistent with 80 cm. Preview cm hypothesis redraws the graph in a separate scenario; the original table remains 80 m. Students should verify their notebook before manually changing a measurement. A timing or oscillation-count error can also explain a mismatch: a hypothesis is not proof.
How it was built and AI disclosure
Lab Detective is a browser-only application built with HTML, CSS, JavaScript ES modules, and SVG graphs, hosted on ChatGPT Sites.
The physics model follows the small-angle simple-pendulum relation T² = 4π²L/g. Reference: OpenStax, University Physics Volume 1, section 15.4, Pendulums. The sample readings are demonstration data, not measurements collected from students.
Challenges and technical learnings
Unit conversion must happen before comparison or fitting. An outlier can distort a regression dramatically, while silently correcting it would erase the evidence. Separating original measurements from an explicit hypothesis makes that tradeoff visible. Another challenge is communicating uncertainty: the rule thresholds are transparent heuristics, not statistical confidence intervals or a definitive diagnosis.
Validation and current limits
Agent-run validation passed 15 physics tests covering calculations, conversions, fitting, invalid input, and preservation of original readings. Hosted desktop checks verified the demo, separate hypothesis, both graph modes, manual correction, adding/removing readings, invalid oscillation counts, reset, and milliseconds conversion. Mobile rendering and optional WebMCP integration remain unverified.
This MVP supports the simple pendulum only. Readings stay in page memory and reset on reload. It does not yet export or save experiments, propagate measurement uncertainty, or account for every departure from the small-angle ideal model.
What is next
Useful next steps are CSV export, saved experiments, explicit measurement uncertainty, teacher-led classroom testing, and additional experiments. The goal is to make a second look at lab data understandable while keeping the student responsible for the evidence.
Log in or sign up for Devpost to join the conversation.