You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This file documents the top-level repository layout outside docs/ and src/.
Those two directories are treated as stable documentation and source-code roots
and are intentionally out of scope for this cleanup pass.
Review mark format:
Reviewed 2026-06-18: role and cleanup policy checked in this pass.
External: managed outside this repo cleanup policy.
Ignored: local artifact/runtime path covered by .gitignore.
Stable Project Roots
Path
Role
Notes
Review
configs/
Runtime and evaluation configuration
Presets live in configs/presets/; mamba_whole_graph.yaml is the current frozen main-line preset. configs/shipping/ holds the resolved shipping config the native builders consume (decision_relevant since #465 PR-4b).
Reviewed 2026-10-03
include/
C++ / CUDA public headers
Native tracking, media, perception, and package-facing headers.
Reviewed 2026-06-18
scripts/
Reusable command-line tools
Evaluation, training, native build helpers, benchmarks, and model tooling. Every script carries a # status: label and a docstring; see the generated docs/ownership/scripts_inventory.generated.md (or each dir's README.md) for the function/status index, and scripts/README.md for the cleanup ledger.
Reviewed 2026-07-21
tests/
Test suites
Unit, integration, native, benchmark, golden, and experimental tests.
Reviewed 2026-06-18
shipping/
Python-free shipping runtime (native)
saccade_shipping_config: strict loader for configs/shipping/*.resolved.json (#465 PR-4a); saccade_shipping_native_config: CUDA-free JSON → tracker-parameter mapping and readback; saccade_shipping_native: builds the GPU tracker/GMC/pipeline from the resolved config and fails closed on readback (PR-4b). post_detector_*: the native post-detector replay host (NMS → detection filters → GMC → tracker, serial) and tools/saccade_replay (developer replay of a Python dump, PR-5 / U3a). decision_relevant since PR-4b; built standalone (CPU CI) and from the root build.
Reviewed 2026-10-03
tools/
Developer coordination tools (not pipeline code)
resctl.py: cross-worktree GPU / CPU / benchmark leases and handoffs; state lives in .git/worktree-runtime/, never tracked. See docs/WORKTREE_RESOURCES.md.
Reviewed 2026-09-18
third_party/
Vendored external code
Currently includes TrackEval. Avoid mixing local experiment output here.
Reviewed 2026-06-18
docker/, Dockerfile, docker-compose.yml
Container entry points
Keep deployment/build container changes here.
Reviewed 2026-06-18
infra/
Local infrastructure config
PostgreSQL/systemd and service-level support files.
Reviewed 2026-06-18
.github/, .githooks
Repository automation
CI and local git hook material.
Reviewed 2026-06-18
CMakeLists.txt
Native build entry point
Keep at repo root.
Reviewed 2026-06-18
pyproject.toml, uv.lock, .python-version
Python project metadata
Keep at repo root.
Reviewed 2026-06-18
main.py
Legacy/manual app entry point
Keep until confirmed unused; do not create new root scripts beside it.
Reviewed 2026-06-18
build_fpn_reid.py
Moved native helper
Moved to scripts/native/build_fpn_reid.py; root copy removed.
Reviewed 2026-06-18
Data, Models, And Runtime State
Path
Role
Git policy
Review
datasets/
Local datasets such as MOT17, DanceTrack, SportsMOT, CrowdHuman
Ignored. Large local data only.
Reviewed 2026-06-18
models/
Model metadata and local model binaries
Metadata and placeholders may be tracked; large weights/engines are ignored.
Reviewed 2026-06-18
storage/
Local service state
Runtime storage such as PostgreSQL and MLflow artifacts should stay ignored.
Reviewed 2026-06-18
.venv/
Local virtual environment
Ignored. Recreate with uv sync.
Reviewed 2026-06-18
.env, .env.example
Local/private env and template
.env ignored; .env.example tracked as template.
Reviewed 2026-06-18
Generated Outputs
These directories are local experiment/runtime output. They are not source of
truth unless a result is explicitly promoted into a tracked report or decision.
Path
Role
Cleanup rule
Review
out/
Short-lived evaluation outputs
Safe to delete after important results are summarized elsewhere.
Reviewed 2026-06-18
output/
Historical evaluation/sweep outputs
Safe to prune after preserving useful metrics.
Reviewed 2026-06-18
results/
Evaluation result directories
Large local artifact root; promote only summaries/tables.
Reviewed 2026-06-18
runs/
Training and experiment runs
Large local artifact root; keep only active checkpoints locally.
Ignored unless explicitly tracked. Promote reusable code into scripts/ or tests into tests/; root copies were moved to scratch/root_artifacts_20260618/.
Reviewed 2026-06-18
1
Removed tracked stray file
Contained a stale log/tail fragment, not a project entry point.
Reviewed 2026-06-18
Explicitly Out Of Scope
Path
Reason
Review
docs/
User requested no cleanup in this pass.
Out of scope 2026-06-18
src/
Directory map and language bridge: src/README.md. This layout file still does not inventory src/.
Out of scope 2026-06-18; map added 2026-09-11
.git/
Git internal state.
External
.agents/, .codex/, .gemini/
Agent/tool-local metadata.
External
Cleanup Order
When the workspace is noisy, clean in this order:
Delete stale build/cache artifacts: build/, build-native-coverage/,
.mypy_cache/, .pytest_cache/, .ruff_cache/, .cache/.
Prune old generated outputs: out/, output/, results/, reports/,
logs/.
Prune inactive training runs under runs/ after checkpoint provenance is
captured in report_data/ or the relevant research note.
Move reusable root-level debug scripts into scripts/ or tests/; leave
truly one-off probes in scratch/.
Do not move datasets/ or models/ as part of cleanup. They are intentionally
large local roots and often referenced by scripts.