The best representations have the highest resolution. Spatiotemporal resolution is measured in microns or micro-seconds, which are linearly linked by derivatives like velocity. Both of those depend on the information quality (i.e. temporal or phase resolution) of the inputs: Resolution ∝ input timing precision
To be useful a representation must be trustworthy. "Trust" in computer science is a categorical decision involving authentication protocols. Trust in continuous representations like flow crystals, on the other hand, must be statistical (in the sense of a model's error cloud), based on priors and data. So trust depends on information quantity and quality:
Trust ∝ input bandwidth
Bandwidth-based trust means, among other things, that one can trust one's senses and one's body (typically megabytes) far more than one can trust any form of language (typically bytes).
Trust depends on proximity. If information into or out of a representation travels through space, as with photons or phonons, bandwidth and hence trust will decay in parallel with signal amplitude:
Trust ∝ 1/r2
While stated above for a passive systems, this line of argument applies to active sensorimotor systems too, with two differences. First, the resolution of a sensorimotor representation is limited by round-trip timing-jitter (from internal as well as external causes), not just by the jitter of the inputs. Likewise, trust reflects the statistical reliability of the entire motor-feedback loop ("sensorimotor contingencies"), not just of the input part. Second, active systems may use their autonomy not just to initiate motor actions, but to seek out good calibration data, foraging for information as well as for physical resources.
Aphorisms grounded in mathematics