FOOD ANALYSIS LAB · W11 · SIX-SCENE INTERACTIVE LAB

Process first, compare second: what does spectral shape tell us?

After W10 calibration and repeat scans, clean spectra and build a defensible qualitative comparison. Allow 3–5 minutes per scene; navigate freely.

All curves are generated from Gaussian peaks and seeded noise for teaching, not measured food or chemical reference spectra. A, B, C and A′ are fictional library labels. Do not use them for authenticity or safety decisions.

1 · Predict

2 · Operate

Move a slider or select a scenario to recalculate; settings carry between scenes. Each scene shows its own processing step; the last uses the full shared pipeline. Scene 4 isolates SNV response to pure offset and gain.

3 · Observe

Same grid, same processing: top-3

Library candidate ranking (all labels are synthetic)
Rank	Candidate	Pearson r ↑	Cosine ↑	SAM ° ↓	Euclidean ↓

There is no universal similarity threshold. This lab does not perform quantitative NNLS. Even signal coefficients normalized to sum to 1 are not mass fractions: standards, response calibration and independent validation are required.

4 · Explain and record

Which setting changed? What changed in the numbers or curve, and why? (5+ characters; saved per scene)

Take your evidence with you

CSV contains all 281 points and processing stages. The JSON record contains settings, all candidate scores, predictions and explanations. Progress stays in this browser; nothing is uploaded.