braindough is a local-first monorepo for brain-modeling experiments. The
first target is Meta/Facebook TRIBE v2, with deterministic fake-backend runs in
CI and heavyweight model/cache/run data kept outside the repository.
mise install --locked
just bootstrap
just ci
just run-fakeLocal heavy data defaults to:
export BRAINDOUGH_HOME="/Volumes/Virtual Machine HD/Projects/braindough"TRIBE v2 runs are opt-in because they may download large model assets and use machine-specific acceleration:
just storage-init
just run-tribeIf TRIBE v2 cannot run on the current machine, run-tribe writes an explicit
skipped artifact with the blocker instead of failing silently.
just bootstrapinstalls pinned tools, syncs Python deps, and installs hooks.just checkruns formatting checks, lint, type checking, tests, and artifact validation.just run-fakeruns the CI-safe deterministic smoke experiment.just run-tribeattempts the first local TRIBE v2 suite.just dataset-bold5000-downloadstages BOLD5000 Release 1.0 stimulus-name metadata and processed ROI response archives under external storage.just run-bold5000-realruns a bounded real-data BOLD5000 ROI encoding benchmark using Release 1.0 ROI matrices and stimulus metadata. BOLD5000 Release 2.0 is author-recommended for new functional analyses and remains future adapter scope.just run-bold5000-preprint-tr34andjust run-bold5000-preprint-tr3run the larger seeded-random BOLD5000 metadata benchmarks used by the preprint workflow.just run-bold5000-preprint-tr34-grouped-sensitivityruns a lower-cost grouped-by-filename split sensitivity for repeated-stimulus leakage checks.just run-fake-focused-ultrasoundruns a CI-safe focused-ultrasound protocol bridge that stores synthetic target/sham metadata without claiming acoustic modeling or real sonication.just run-tribe-focused-ultrasoundruns the same bridge through a bounded local TRIBE v2 budget.just report RUN_DIR=...rebuilds human-readable reports for an existing run.
See docs/ for architecture, storage layout, artifact schema, TRIBE v2 notes,
and the experiment backlog.