CLI

dotnet run --project src/DotFly.Cli -c Release -- <command> (or the built dotfly executable). Neuron selections everywhere accept body IDs, set:NAME, type:NAME, class:NAME, superclass:NAME with globs, an optional @L / @R side suffix, and ; for unions — e.g. "type:LC4@L;type:ORN_DM1".

command	what it does	options
`info`	environment, SIMD width, versions
`build malecns <data-dir>`	MaleCNS Feather release → `.dfb` checkpoint	`-o`, `--filter superclass\\|traced`, `--weights <file>`
`build flywire <completeness.csv> <connectivity.parquet>`	FlyWire (Shiu et al. files) → `.dfb`	`-m 630\\|783`, `-o`
`inspect <checkpoint>`	provenance and counts; a body, a type; who a neuron drives or who drives it	`--body`, `--type`, `--upstream-of <ID>` (with `--hops`, `--min-weight`, `--class`), `--top`
`run <checkpoint>`	stimulate populations, optionally silence others, report who fires — during and after the stimulus	`--stimulate`, `--hz`, `--silence`, `--seconds`, `--stim-seconds`, `--seed`, `--backend cpu\\|reference`, `--threads`, `--gain`, `--top`, `--csv`
`bench <checkpoint>`	steps/s and real-time factor for the standard scenarios	`--scenario`, `--hz`, `--seconds`, `--threads 1,4,8`, `--seed`
`explore <checkpoint>`	live spike raster and statistics in the terminal; keys toggle recurrent transmission (`r`), external input (`e`) and the rate (`+`/`-`)	`--stimulate`, `--hz`, `--watch`, `--ratio`, `--threads`, `--seed`, `--duration`, `--rows`

Recipes that produced the findings

# Who drives the proboscis motor neuron? (found the GRN → MN9 pathway)
dotfly inspect .data/malecns-v1.0/malecns-v1.0-superclass.dfb --upstream-of <MN9 id> --hops 2 --min-weight 5

# Does a readout return to baseline after the stimulus stops? (the runaway attractors)
dotfly run <ckpt> --stimulate "type:ORN_DM1" --hz 60 --seconds 3 --stim-seconds 1 --silence "class:Kenyon_Cell;type:lLN*"

# Is a readout lateralized? (wind: AMMC012 answers one antenna only)
dotfly run <ckpt> --stimulate "type:JO-C*@L;type:JO-E*@L" --hz 60 --seconds 2 --silence "class:Kenyon_Cell;type:lLN*" --csv left.csv
dotfly run <ckpt> --stimulate "type:JO-C*@R;type:JO-E*@R" --hz 60 --seconds 2 --silence "class:Kenyon_Cell;type:lLN*" --csv right.csv

# A response curve (DM1_lPN vs ORN input rate)
foreach ($hz in 5,10,20,40,80,120) { dotfly run <ckpt> --stimulate "type:ORN_DM1" --hz $hz --seconds 2 --silence "class:Kenyon_Cell;type:lLN*" --top 400 }

# Performance
dotfly bench <ckpt> --scenario type:T4a --threads 1,4,8

--csv writes every neuron that fired with its type and count, for scripting. DOTFLY_PROFILE=1 prints the kernel profile at the end of a run; DOTFLY_PIN_CALLER=1 pins the calling thread (never inside a host).