PIV Desk - notice

The app's agent prompt is derived from the agent skill "openpiv" (@k-dense-ai/openpiv, skill
version 1.1, skill author OpenPIV Team, tested against openpiv 0.25.4) - its single-pass pipeline
scripts/runner.py, its validation, units and multi-pass guidance, and scripts/analyze.py - in the
repository k-dense-ai/scientific-agent-skills.
  https://github.com/k-dense-ai/scientific-agent-skills (skills/openpiv)
The skill's front matter declares the BSD-3-Clause license; the repository is MIT-licensed. No
text of the skill is redistributed verbatim; the prompt was rewritten for this app.

The in-browser analysis (piv.js, pivkit.js) is an independent JavaScript implementation of the
behaviour of openpiv 0.25.4 (OpenPIV, https://github.com/OpenPIV/openpiv-python, GPLv3):
pyprocess.extended_search_area_piv with its intensity normalisation, masking, linear/circular FFT
correlation (including the one-column-short inverse transform of the linear path), Gaussian /
centroid / parabolic sub-pixel peaks and peak2peak / peak2mean ratios; get_coordinates;
validation.sig2noise_val, global_val and local_median_val; filters.replace_outliers and
lib.replace_nans; scaling.uniform; tools.transform_coordinates; tools.save's vectors.txt format;
tools.rgb2gray's weights; and the skill's analyze.py statistics, vorticity and strain. No OpenPIV
source code, test data or images are included: the page was written from the documented behaviour
and checked against the installed package's outputs. imgio.js reads PNG, TIFF, BMP and JPEG the way
imageio / Pillow present them to openpiv. The synthetic pairs are generated by synth.js (Gaussian
particle images, Lamb-Oseen vortex, uniform shift, shear layer) and are the page's own.

Verification (2026-09-27): 526 image pairs - 490 random synthetic pairs over window, overlap, search
area, correlation, sub-pixel, ratio, threshold, validation, replacement and drop_invalid settings,
and 36 runs over the six image pairs bundled with openpiv 0.25.4 - 175,637 vectors in total.
175,631 raw vectors agreed with openpiv within 0.001 px per frame and every signal-to-noise ratio
within 0.01%; 1 validation flag differed. The 6 remaining vectors lie in the normalised
(search area > window) path, where openpiv computes in single precision and a near-zero point next
to the correlation peak makes the sub-pixel fit sensitive to rounding. vectors.txt matches openpiv's
file line for line except for last-digit differences in about 9% of lines.