Waveform QC on CloudSprite

Theory lives in the RF-domain pages: methodology, thresholds, alignment. This page is the product loop for “which of my S21 measurements in this project look like outliers?”

Filter one family

Pick the project, the trace type (usually S21), and the batch (lot=15, a day, or the whole project if it really is one family). Need ≥ 3 traces.

Score, then average

1. Load those S21 traces.
2. Align x (interpolate or trim).
3. Convert to dB.
4. Pairwise Pearson r; per-trace score = mean of the row excluding the diagonal.
5. Flag scores more than σ standard deviations below the mean (default σ = 2). Print every score.
6. Average the non-outlier traces.

Do not average first. Do not drop outliers from the project.

Publish

cs.set_project("cable-qual")
cs.track_script(__file__)
cs.publish(traces=[clean_avg], name="s21-lot15-average")

Tag outlier datasets outlier (preview → confirm). Parameter the average with qc_sigma=2 and the filter you used. Notebook: outliers + average, not every passing trace unless someone asks.

Assistant path

A typical QC request is: explain the Pearson method, generate a script parameterized for the current project, then Save as Script → Run. The chat links the outlier list and the published average. Tagging is preview-then-confirm; datasets are not deleted as part of QC.

If every trace flags, σ is too tight or the filter mixed families — an empty average is not published.