Wishlist

flybody simulates no neurons and uses no connectome, so it cannot clear L1 — and L0 does not describe it either, because there is no connectome being used as wiring. It is the platform other projects attach a brain to. It sits unrated with that explanation, but the honest fix might be a separate mark for platforms rather than a gap in the scale.

Wanted: A decision: widen the scale, add a platform mark, or leave these unrated on purpose.

Every one of the 96 levels is one reader’s draft. A second person checking the evidence against the source moves it to reviewed; disagreeing moves it to disputed, which is just as useful. Five proposed L2s were already downgraded on a second reading — four for running a generic Hebbian or STDP rule rather than a fly one, one for gating "dopamine" on a keyword the user types — so the second pass is not a formality.

Wanted: A second reader on the 16 L2s. Those are the claims that would embarrass the site if they were wrong.

L1 says 'leaky integrate-and-fire or better'. Fly Dino uses signed normalised leaky tanh units — state carries over time, but it is a rate model, simpler than LIF. It is rated L1 with that caveat written into the note. Several other projects sit in the same place, so the answer should be a rule rather than a judgement call each time.

Wanted: A sentence in the rating rules that settles rate models one way or the other.

A good number of indexed projects are browser demos with a public URL. Watching a fly circuit fail at Doom next to its rating is a stronger argument than either on its own — and it makes an unrated project easier to rate, because you can see what it does before reading how.

Wanted: An embedded demo on detail pages where the project publishes one.

Right now the index sorts by the order the source lists carry. Sorting or filtering by evidence — rated, unrated, disputed, most recently read — would make the gaps visible and give anyone wanting to help an obvious place to start.

Wanted: A sort control on the index.

The index reads two community lists. A third would mostly duplicate them — the last sweep found 33 of 99 entries already indexed — but a list built around a different dataset, the larval connectome or BANC rather than MaleCNS and FlyWire, would not.

Wanted: A curated list the index does not already read.

Several projects ship a complete, properly grounded fly plasticity rule that is off by default. The site currently rates the default run — so they sit at L1 with the dormant arm described in the note — because a level that depends on which flag a reader sets is not checkable. It is a conservative call that feels unfair to a good implementation, and it deserves a rule stated on the scale page rather than only in the method.

Wanted: A decision: rate the default, rate the best supported path, or add a mark for "plasticity available".