Whereas the way in which information is represented across neurons - eg with local tuning or not - is more interesting. And a locally tuned response looks like a continuous grandmother response.. But if that does not count as “local coding”, I’ll move on.

Well, it’s not realistic (to me) that a neural representation go from 0 to max after an infinitesimal increase in the grandmother-ness of any real stimulus (non-linear attractor-like responses, like Wills et al 2005, sure, absolute step function not so much). So maybe it’s just not the debate for me

By (this) definition all continuous variables are represented by distributed codes. That does not leave much to discuss or insight to be gained, even though “distributed” would normally mean that the firing of an individual neuron is not interpretable on its own.

Should have said (distributed) every unit counts for each value (obv each unit counts in local codes too, but different units for different values).

Yes, grandmother cells are discrete. But, if your simple cells do not have localist coding of angle, then no continuous representation can be localist. Distributed reps (eg in PDP books) were supposed to be like binary numbers (every unit counts), which doesn’t seem to fit here.

Yes, these two issues can be confounded, but discrete vs continuous (eg is grandmother-like valid) is different to local vs distributed (eg can I tell what is represented from a single neuron or must I see the whole population).

The head direction cells are a good example of a population code in which each neuron is tuned to a single direction (and nothing else).
Do HD cells count as localist coding? (If not then I suspect nothing will.)

True, but the referee-informed “arbitrary decision” is still better than the editorial board’s “arbitrary decision” (altho it does require more effort).

Yes! And thanks to @shanewriter.bsky.social for pointing us to your online dataset when I asked about the theta modulated head-direction in the original Tsanov et al 2011

Thanks! There cld be species diffs, but also:
Theta mod HDCs have broader tuning curves (partly due to l/r sweeps, see Vollan et al 2025) so criteria for classic HDCs might miss tmHDCs in AV?
We do see adaptation in classic HDCs in mouse AD (not PoS) in your online data (Duszkiewicz)
Talk offline?