Fly connectome (FAFB v783)

A complete, real-world MORK example: a unified, persisted, manifest-driven representation of the FAFB v783 adult Drosophila (fruit-fly) connectome on the MORK substrate, and the Fig-6 information-flow model (Dorkenwald, Schlegel et al., Nature 2024) run over it. Lives in examples/connectome/.

It is a real heterogeneous graph organized into world-model Spaces — a concrete instance of the world-modeling architecture ("World Modeling in Hyperon for PRIMUS", Goertzel 2025) — all co-resolving on one key, the neuron root_id, with the raw data kept immutable as evidence. It exercises the MORK substrate end-to-end: prefix-scoped multi-Space .act snapshots, cold-mmap reads, and the zipper-frontier traversal from the Zipper Queries guide.

The Spaces

Space	Atom	Source	Role
`S_ent`	`(neuron <id> <flow> <super_class> <class> <sub_class> <side>)`	`classification.csv.gz` (139,256 neurons)	identity / ontology. For afferent neurons `class` is the sensory modality — info-flow seeds are derived by querying this Space, not hand-listed.
`S_rule`	`(syn <pre> <post> <Σsyn_count>)`	`connections_princeton.csv.gz` (5.34 M rows → 3,732,460 unique edges)	the wiring.
`S_map`	`(skel-br <id> <bid> <st> <sx> <sy> <sz> <et> <ex> <ey> <ez> <len> <rad> <n>)`	139,273 SWC skeletons → 28 branch-decomposed `.act` shards	spatial morphology (branch endpoints, cable length, radius).
`S_evid`	the raw 31 GB SWC files	`sk_lod1_783_healed/<id>.swc`	immutable evidence. Provenance is `root_id → <id>.swc` — already implicit in every `skel-br` atom (no CID field, no re-import).

Every Space keys on the same root_id, so one id resolves its class (S_ent), morphology (S_map), synapses (S_rule), and source file (S_evid).

Design — MORK-canonical, no technical debt

One .act per Space (canonical per-Space tries), cold-mmap'd on use (act_open_mmap, ~0 RAM) — built once, instant thereafter. No /tmp staging, no re-ingest per run.
Manifest-driven paths (manifest.jl) — single source of truth, ENV-overridable (CONNECTOME_DATA_ROOT / CONNECTOME_STORE / CONNECTOME_SKEL_DIR / MORK_PROJECT); no hardcoded absolute paths.
Reads via the documented zipper algebra (read_zipper_at_path + prefix-narrowed walks over the .act snapshot — the sanctioned high-speed read path for relational reachability).
Derive, don't pre-bake: seeds come from an S_ent match, not a text file.

Files (`examples/connectome/`)

File	What
`manifest.jl`	resolved paths (data root, store, shards, MORK project)
`load_ent.jl`	`classification.csv.gz → S_ent.act`; `query_seeds(modality)`
`load_rule.jl`	`connections.csv.gz → S_rule.act` (Julia-side aggregation, no `awk`/TSV)
`load_map.jl`	`S_map` accessor over the skeleton shards; `query_skeleton`, `neuron_morphology`, `evid_swc_path`
`info_flow.jl`	the Fig-6 flow over the persisted store: `info_flow`, `validate`, `all_modalities`

Companion artifacts: InfoFlow.metta / InfoFlowMS.metta / InfoFlowFast.metta (pure-MeTTa exec-calculus models — a different, interpreter-based approach), import_fafb_skeletons.jl (the S_map importer), extract_{fafb,banc,flywire}_subset.sh (subset/oracle generators; BANC + FlyWire are future cross-species runs), STAGE2_RESULTS.md.

Usage

Run inside the MORK project's warm REPL (the substrate uses raw MORK/PathMap):

include("examples/connectome/info_flow.jl")
using .ConnectomeManifest
m = ConnectomeManifest.manifest()

# Build the store (idempotent; first S_rule build ~4 min, then instant cold-mmap):
ConnectomeManifest.ensure_store()
open_store(m)                       # builds/opens S_rule.act + S_ent.act

# Single-modality info-flow (seeds derived from S_ent):
seeds, ranks = info_flow(m, "thermosensory")

# Regression: zipper-frontier flow == independent in-Julia oracle, from the store:
validate(m, "thermosensory")        # → 0 rank-mismatches

# Fig-6 d/e across all 7 afferent modalities on one shared connectome:
all_modalities(m)                   # per-modality rank histograms + Jaccard overlap

# Cross-Space resolution by neuron id:
include("examples/connectome/load_map.jl")
neuron_morphology(m, "720575940611720362")   # branches, cable length, S_evid path

Validation

Build: S_rule.act reproduces the 5.34 M → 3,732,460 unique-edge aggregation.
Flow vs oracle: thermosensory reaches 367 neurons with rank-1 = 46, and the zipper-frontier flow matches an independent naive-rescan oracle node-for-node (0 mismatches) — both re-derived from the persisted store.
Fig-6 d/e: all 7 modalities run on one shared connectome; the cross-modality Jaccard matrix shows the expected biology (e.g. AN↔mechanosensory ≈ 0.41, thermo↔gustatory = 0.0).
Cross-Space join: all 29/29 thermosensory seeds resolve in S_ent, S_map (e.g. 214 branches / 601,722 nm), S_rule, and S_evid on the same root_id.

Data & citation

Real FlyWire/Codex FAFB v783 exports (CC-BY-NC). Cite Dorkenwald, Schlegel et al., Nature 2024 ("Neuronal wiring diagram of an adult brain") for any published result. Raw data is gitignored; not redistributed here.