Two-state blinking, a known Gaussian point-spread function, and a Poisson noise model (a normal approximation for pixel means of 30 or more). Photon yield is per active molecule per frame. The single-emitter fit screens spot brightness and shape against this model. It assumes the configured mean brightness is the same for every molecule. Rejected spots reduce detection recall; dim or overlapping molecules can still give biased fits.
Fit error is RMS per axis for one-to-one matches to active emitters in the same frame, within one PSF sigma. Missed emitters and unmatched fits are counted separately. These measurements do not establish image resolution.
Drift correction subtracts the exact simulated motion. It demonstrates ideal correction; it does not estimate motion from the images.
Point markers are a fixed fraction of the view on every screen, not uncertainty kernels. Camera brightness uses one fixed scale for the acquisition. The camera's all-frame mean averages the raw acquired frames and is normalized separately. Both retain stage drift.
The conventional view sums the same Gaussian PSF over every emitter, with all emitters on. It shows the ideal fluorescence image before camera sampling, without noise, background, or drift. Its brightness is normalized independently. All four views show the same physical crop; colors indicate intensity or position, not emission wavelength.
The actual object is the known geometry or uploaded image. Molecules shows its labels; the reconstruction contains only accepted camera fits. Labels may blink repeatedly or remain unseen during a finite acquisition. The same seed, settings, and source image reproduce the same run. Shared links do not include uploaded images.