sparcd-exploration

A workspace for building small, focused, mostly-static tools that work alongside SPARC'd — each one tuned end-to-end for a single feature.

A shared landing page ties them together at the deploy root. The uploader and tagger share a connection gate and a saved-login session, so you authenticate once and move between them; the explorer has its own sign-in.

The tools

• apps/sparcd-explorer — a marimo notebook that signs in to the SPARC'd MinIO backend, loads a collection's Camtrap-DP CSVs, and renders a Field Notebook view: a hex-binned map, a species dashboard, stat cards, and drill-in tabs for images, detections, and locations. Query filters apply on Search; display options tune presentation live. Exports to a static Pyodide bundle that runs entirely in the browser (see Static deploy).
• apps/sparcd-uploader — a static, browser-based tool for preparing and uploading camera-trap image batches. Drop a folder; it scans JPEGs and MP4 videos, runs EXIF, SHA-256, and thumbnails in Web Workers, validates the batch, then writes the canonical Camtrap-DP layout through the s3-safe boundary. Dry-run by default.
• apps/sparcd-tagger — a static, browser-based tagging interface for camera-trap images. It reads the same buckets, renders an upload's images from presigned URLs, and writes back the canonical Camtrap-DP metadata the other readers already consume.
• apps/sparcd-home — the shared landing page and app switcher served at the deploy root.

Each app's README.md (and plan.md where present) carries its full design and phase breakdown.

Approach

• Alongside SPARC'd. SPARC'd is the system of record; the tools here read from it and add focused views on top.
• One tool, one job. Each app in apps/ solves a single concrete user problem (a specific report, a specific view, a specific export). When a new need shows up, we add a new app.
• Static where possible. Prefer designs that can ship as a static bundle (Pyodide / WASM, prebuilt data files, signed S3 URLs). Each tool stays cheap to host, easy to share, and free of server-side state.
• Bring your own S3. The browser tools have no backend and no server-side secret. Users supply an S3-compatible endpoint and credentials; IAM/provider policy and bucket CORS are the real access gates. Writes go through @sparcd/s3-safe, an append-only boundary with no delete or copy API and a single reviewed, ETag-gated conditional-replace path (replaceIfUnchanged).
• Optimize per feature. With a narrow scope per app, we pick the best primitives for that job — data model, layout, interactions — without compromise for anything else.

Layout

apps/
  sparcd-home/       # shared landing page + app switcher (static HTML)
  sparcd-explorer/   # marimo notebooks for data exploration (Python, uv)
  sparcd-uploader/   # batch prep + upload, BYO-S3 (TS, Vite)
  sparcd-tagger/     # tagging interface, BYO-S3 (TS, Vite)
packages/
  auth-ui/           # shared connection gate + saved-login session
  camtrap/           # Camtrap-DP data contract (readers, merge, time-shift)
  s3-safe/           # S3 client boundary (append-only + reviewed replace)
  types/             # shared TypeScript types

Toolchain

• Node ≥ 20 + pnpm 10 — workspace + task runner
• Turborepo — pipeline orchestration across apps/packages
• uv — Python env/deps for any Python-based app (e.g. marimo)

Quick start

pnpm install                                  # installs turbo and JS workspaces

pnpm --filter sparcd-uploader dev             # Vite dev server (uploader)
pnpm --filter sparcd-tagger dev               # Vite dev server (tagger)

pnpm --filter @sparcd/sparcd-explorer install:py   # uv sync for the marimo app
pnpm dev --filter @sparcd/sparcd-explorer     # marimo edit --watch

Or run every app's dev task at once:

pnpm dev

The Vite apps prefill the S3 endpoint from a gitignored apps/<name>/.env (VITE_SPARCD_S3_ENDPOINT). Credentials are never prefilled — they are entered at runtime.

Adding a new app

1. mkdir apps/<name>
2. Add a package.json with name, private: true, and at least dev / build scripts. Python apps wrap uv run … in their npm scripts.
3. pnpm install to pick it up via the workspace.
4. Tasks defined in turbo.json (dev, build, lint, start, check, test) will run across whichever apps implement them.

Static deploy

.github/workflows/pages.yml builds the landing page and each web tool and publishes them via GitHub Pages on every push that touches an app, a shared package, or the workflow itself. The landing page sits at the root, with each tool under its own path (/explorer, /uploader, /tagger). Live at:

https://culverlab.github.io/sparcd-exploration/

The deployed pages run entirely in the visitor's browser — the explorer runs Python via Pyodide, and the Vite tools talk to S3 directly. SPARC'd credentials are entered in the connection gate; there is no server-side secret. The S3/MinIO endpoint must permit CORS from the Pages origin for data fetches and uploads to succeed.

Docs

• Each app's README.md (and plan.md where present) — that tool's design, data contracts, and status
• docs/design-system-field-notebook.md — the shared visual design system
• docs/archive/ — superseded proposals and point-in-time reports, kept as decision history; nothing there describes the current system