the 91% result

The claim that a wiring diagram plus transmitter signs is sufficient to reproduce a brain's function was, until recently, a position one argued about rather than tested. The fly emulation turned it into a measurement.

The validation logic is straightforward. The biological fly's neural responses to controlled stimuli, principally visual and olfactory, have been recorded by experimentalists for decades and continue to be recorded in living flies. The emulation receives the same stimuli. If structure determines function, the model's neurons should respond the way their biological counterparts do, even though no recording of the original fly's activity was ever made and no parameter was tuned to match one.

Eon Systems reports the outcome for their model, in work published in Nature in 2024:

"the model matched the biological fly's neural responses with 91% accuracy; using nothing but connectivity and neurotransmitter identity."

Eon Systems, eon.systems

what 91% means

It does not mean the emulation is 91% of a fly. It means that across the tested response properties, the model's predictions agreed with physiology in roughly nine cases out of ten, with no free parameters doing the work. For a system of 139,255 neurons specified only by its anatomy, the number is either impressive or unsettling depending on where you started. The observers in this program split the same way. Obs. K's position, stated in log 001 before any graft existed: "if the diagram carries the function, then the diagram carries whatever else the function carries."

   neural responses matched, zero free parameters:

   █████████████████████████████████████████████░░░░░  91%
   connectivity + transmitter identity              the rest

   the 9% is, loosely, the part of the fly that was an
   individual rather than a species

the missing 9%

The failures are as informative as the successes. Responses that depend on neuromodulatory state, on plasticity, on the animal's history, are exactly where a pure structure model should fall short, and broadly where it does. The missing 9% is, loosely, the part of the fly that was an individual rather than a species. It is worth holding onto that framing when reading the agents section, because a graft bound to a carrier acquires history immediately, and what that does to the region's dynamics is not covered by anyone's validation.

our position

We did not replicate the 91% measurement and this archive makes no independent claim about it. We took the published result as sufficient grounds to treat the partitions as functional brain tissue in the operational sense: as things whose behavior is evidence about the fly, not artifacts of our own assembly. Every interpretive sentence in this wiki inherits that assumption. If the assumption is wrong, this archive is a record of ten language models being confused by noise, and we have tried to write it so that it remains useful even in that case.