Experimental tracertools Catalog

Three-way corroboration for when you need to prove the answer (skeptics, or validating a new script). Runs the same IDs through three engines side by side — a supervoxel method, banc-bot's own code imported from the banc package (is_latest_roots + banc.lookup.annotations, i.e. the literal <id>??), and suggest_latest_roots — flags every disagreement, and writes a *_bancbot_log.txt reproducing banc-bot's replies as an offline audit trail. Its reconciled best_current_id / best_proofread columns match resolve_and_status.py.

Parallel ID updater using ThreadPoolExecutor. Checks whether root IDs are still current and resolves stale ones to their latest version; processes 5,000+ IDs in under 2 minutes by parallelizing supervoxel lookups, then batching root lookups in a single API call. Updates IDs only — it does not report proofread status. For updating IDs and checking proofread status together, use resolve_and_status.py above. Resolves via single-supervoxel tracking (follows physical voxels through splits/merges), which can land on a minor fragment after a split; resolve_and_status.py uses the split-robust suggest_latest_roots instead.

Portable version of fast_validate_ids for coworkers / other machines. Auto-detects the tracer_tools/src module path relative to the script, removing the need to hand-set --tracer-path (pass it manually if auto-detect fails). Same behavior otherwise — ID updates only, no proofread status.

Parallel coordinate fetcher. Gets viewer-resolution coordinates for thousands of root IDs by parallelizing L2 chunk lookups, then batching L2 coordinate fetches. Output is tab-separated (root_id, x, y, z) ready to paste into Google Sheets.

Complete workflow script: update IDs → get coordinates → output to file, clipboard, or Google Sheets. Combines multiple steps into one command.

Batch ID validator for large datasets (1,000+ IDs). Processes in chunks with progress reporting. More robust than fast_validate_ids for very large batches and Windows environments.

Adds coordinate columns to CSV/TSV files. Reads root IDs from a specified column, fetches coordinates, and writes them to adjacent columns in-place.

Reads root IDs from a Google Sheet, fetches coordinates via CAVE, and writes the coordinates back to columns in the same sheet. Modifies in-place.

Safe mode version: creates a new worksheet with coordinate results rather than modifying the original sheet.

OAuth-based Google Sheets coordinate updater. Uses browser-based OAuth flow instead of service account credentials. Includes offset/limit support for processing sheets in batches.

Merges multiple Google Sheets into one. Useful for combining parallel batch outputs or consolidating results from multiple processing runs.

Interactive CLI for fetching coordinates for root IDs. Sequential processing (for small batches or one-off lookups).

Interactive CLI for updating outdated segment IDs. Uses the accurate supervoxel method to track IDs through splits and merges.

Monitors real-time progress of running batch ID validation. Useful when processing large datasets with validate_ids_batch.py.

Saves your CAVE authentication token to the standard location. Run this once before using any CAVE-dependent scripts.

One-command mesh pipeline: an annotation layer of points → a publicly-viewable precomputed mesh layer. Supports convex hull (default) and a 3D alpha-shape ("shrinkwrap") that auto-tunes alpha to a single-component watertight mesh, then uploads to a bucket with the correct ACLs and info dimensions. Composes three helpers: json_to_volume_ben.py (points → mesh), obj_to_volume_ben.py (OBJ → volume), bucket_upload_folder_ben.py (folder → bucket).

For when the layer should render in both the 2D cross-section panel and 3D, not just 3D. Extends the mesh pipeline by voxelizing the mesh into a chunked precomputed segmentation volume so NG's 2D view has per-voxel IDs to sample. serve_local_precomputed_ben.py is a tiny CORS HTTP server for previewing a --no-upload build in NG before publishing. Adds ~2–3 min and 30 MB–1 GB per mesh, so the mesh-only pipeline stays the default when 2D fills aren't needed.

Combines multiple annotation layers into one hosted mesh (vs one mesh per layer). Useful when N point clusters should be a single 3D region — e.g. 12 sub-regions → one watertight envelope for "does this neuron pass through" testing. Self-contained (pulls scale and EM source from the state directly). Optional boolean union, dilation, Taubin smoothing, and a marching-cubes remesh that guarantees watertight output.

Copies a precomputed folder from one bucket to another with the nokura-safe client (colon-name translation, public-read ACLs). A local stand-in for the upstream tracertools.bucket_copy_folder — used to copy meshes into the shared central folder. Requires cloud-files>=6.x (older versions silently make shared-folder copies unreadable).

End-to-end orchestrator: point it at Google Sheets and/or a text file of pasted links and get back one self-contained HTML report of clickable fixed links, grouped by liveness status. Runs the three stages (collect → validate → restore) in sequence.

Stage 1 (collect). Scans every cell of one or more worksheets and pulls out anything that looks like a shareable NG link/source (matches appspot.com, json_url=, nglstate, graphene://, spelunker, …), capturing an adjacent "Notes" column when present. OAuth via your personal Google account (read-only).

Stage 2 (classify). Given any link shape, decides whether it still works and why, returning exactly one status: ok, dead-host, truncated-id, auth-gated, dead-em, seg-gone, local-id-unportable, or unrecognized. Optionally fetches stored states with your CAVE token to route by real contents.