What it is — Single-particle and single-molecule tracking (SPT/SMT) images sparse, individually resolvable fluorophores at 5–50 ms exposure over thousands of frames, then recovers each emitter's position by fitting a PSF model to its spot. The fitted position carries a precision, typically 15–40 nm — an order of magnitude below the diffraction limit. Those positions are linked across frames into trajectories, and the trajectories are the measurement. Nothing here is a picture of a cell: the quantity being estimated is motion, and the answer is a diffusion coefficient, an anomalous-diffusion exponent, the fraction of molecules in each diffusive state, or the shift a compound produces in that distribution. In target-engagement work the headline number is the bound fraction — the share of molecules whose motion is arrested by binding — and its dose response.
Localisation precision is not resolution — It is the uncertainty every downstream diffusion number inherits, and it must be measured rather than assumed. Static localisation error adds a positive offset to the mean-squared displacement at zero lag and inflates D; motion blur over a finite exposure averages the emitter's position and deflates it. The two biases have opposite sign and do not cancel. Both have closed-form treatments in the literature, and quoting a precision at all requires the camera calibration — per-pixel offset, gain in e⁻/ADU, sCMOS read-noise variance — to have been carried through ingest with the pixels. This is Measure Where It Matters stated in nanometres.
Where it breaks — Linking. Emitter density, blinking gaps and the choice of algorithm (nearest-neighbour, LAP, multiple-hypothesis) change the trajectory set that comes out, and they change it non-randomly. Above a density threshold the assignment is simply not identifiable. Stage drift adds a coherent apparent motion that a diffusion fit reads as directed transport. The systemic distortion is track-length bias: fast molecules leave the axial detection volume sooner, so short trajectories are preferentially those of fast movers, and any per-track-weighted statistic under-represents the fast population unless that is explicitly corrected. A track-length distribution is therefore a convolution of biology and photophysics, not a measurement of either — one of the Uncertainty, Drift, and Failure Modes in Bioimage Analysis this modality generates routinely.
The standards gap is real — There is no ground truth for a diffusion coefficient in a live cell, and no community standard for SPT quality control: no minimum-information checklist, no agreed emitter-density threshold, no governed exchange format. Simulation benchmarks and immobilised-fiducial acquisitions are what stand in for truth, which makes this a working example of Validation Without a Ground Truth rather than an exception to it. We state the absence plainly instead of borrowing a standard that was written for a different measurement.
reproducible, wrong bound fraction. Whether the number of diffusive states is fixed by the analyst or inferred from the data is a reportable methodological choice, not a default.
How we run it — the Particle Tracking pipeline, stage by stage — the nine-stage imaging backbone reshaped for this modality, with the tools, sub-steps and deliverable at each stage.