qFoldIT · org.github.com/qfoldit · dashboard v1.1

One scientific problem.
Five runtimes, one canonical state.

A live snapshot of the qFoldIT workspace: engine toolbelts, the cross-engine capability matrix, evidence levels behind each unlocked gameplay pattern, and what's verified vs. still open — compiled from the reorganized workspace archive (STATUS.md, per-toolbelt manifests, the reference compiler, and the org's own architecture doc). Updated 2026‑08‑13 11:30 to include WEB-TOOLBELT (src/) and QFOLDIT-STANDALONE manifests and compile results.

Snapshot date: 2026‑08‑13 11:30 UTC Workspace: 637 files · ~68 MB, reorganized into 5 top-level folders github.com/qfoldit Updated: WEB-TOOLBELT (src/) + QFOLDIT-STANDALONE now have qfoldit.adapter.json and compile results
561+
Registered tools across 4 toolbelts
103+
Tool categories
5
Deployment targets (3 engines + web + desktop)
27
Public repos in the org
2 / 3
Engines at 5/5 pattern compile success
1
Pattern at evidence level E2

Control plane

scientific state → UAG → routing policy → engine toolbelts
Scientific state
Scientific experience model
UAG — Universal Assembly Graph
Capability registry + canonical actions
Routing policy
├── UEFN-TOOLBELT — 356 tools, 55+ categories (0/5 compile · adapter present, all capabilities partial)
├── UNITY-TOOLBELT — 109 tools, 24 categories (5/5 compile · supported)
├── UNIGINE-TOOLBELT — 90 tools, 23 categories (5/5 compile · supported)
├── WEB-TOOLBELT (src/) — 16 tools (0/5 compile · manifest present, no UAG bridge / interaction / vis)
├── WEB-TOOLBELT (static/) — 6 WebMCP commands, live sandbox (presentation layer, not in compiler)
└── QFOLDIT-STANDALONE — 37 tools, desktop (Electron) (0/5 compile · manifest present, blocked only by scientific.visualization on 2 patterns)

Engine toolbelts

tool counts measured directly from source (decorator / attribute grep), not from stale registries
UEFN-TOOLBELTPartial
Tools356 (org page: 358+)
Categories55+
LanguagePython
MCP modeexternal process
LicenseAGPL-3.0
Test suite117 / 117 passing
Adapter manifestpresent (all capabilities partial)
No live-engine verification yet; all four gate capabilities are partial. The closest path to success is implementing the Python equivalents of Unity/UNIGINE's UAGBridgeTools.
UNITY-TOOLBELTSupported
Tools109 (spec: 105 · org page: 102)
Categories24
LanguageC#
MCP modenative in-editor bridge
LicenseAGPL-3.0 (attribution)
CIworkflow added, unverified*
Compile result5/5 success
Real live callback (OnMouseDown → UnityEvent) confirmed in QFoldITInteractable.cs.
UNIGINE-TOOLBELTSupported
Tools90 (spec: 84)
Categories23
LanguageC#
MCP modeexternal process + official MCPBridge
Licensecustom
CI6 / 6 tests passing
Compile result5/5 success
Interaction is physics-selectability + persisted registry — honestly scoped as narrower than Unity's live callback.
WEB-TOOLBELT (src/)Partial
Tools16 (measured from src/)
StackTypeScript + Vite
LicenseAGPL-3.0
Buildtsc + vite — clean
Adapter manifestpresent (0/5 partial)
Compile result0/5 success, 5/5 partial
No interaction, geometry.procedural, camera, lighting, or scientific.visualization — the least built-out MCP codebase.
QFOLDIT-STANDALONEPartial (close)
Tools37 (13 categories)
LanguageTypeScript + Electron
MCP modelocal WebSocket bridge, 127.0.0.1
LicenseAGPL-3.0-only
Buildnode --check · electron-builder (win/mac/linux)
Adapter manifestpresent (0/5 partial)
Compile result0/5 success, 2/5 blocked only by scientific.visualization
Offline-first desktop packaging of the WEB-TOOLBELT runtime. The closest target to 5/5 success — implementing scientific_visualization_create would flip 2 patterns to success.
WEB-TOOLBELT (static/)Live sandbox
Tools6 WebMCP commands
StackThree.js + WebGPU + Rapier.js
Deployqfoldit.github.io/web-toolbelt/
LicenseAGPL-3.0 (core) + ARR (static site)
Buildplain JS, no Vite
Not in the compiler — this is a separate presentation layer with real PDB fetching and multiplayer, but no MCP tool registry. It may be folded into src/ in the future.

P0 / P1 / P2 capability matrix

from the framework's own roadmap — status per engine, on shipped code only
CapabilityPriorityUEFNUnityUNIGINEWEB-TOOLBELT (src/)QFOLDIT-STANDALONE
scientific.visualizationP0 partial supported supported unsupported unsupported
interactionP0 partial supported supported unsupported partial
geometry.proceduralP1 partial native native unsupported partial
physics.jointsP2 (backlog) partial supported partial unsupported supported

Cross-engine compile result

reference/compiler.py × gameplay-patterns.unlocked.json, re-run and confirmed with all manifests present
PatternUEFNUnityUNIGINEWEB-TOOLBELT (src/)QFOLDIT-STANDALONE
protein_folding_construction partial success success partial partial
quantum_ground_state_portal partial success success partial partial
hp_lattice_spatial_puzzle partial success success partial partial
admet_safety_strategy partial success success partial partial
l_systems_procedural_construction partial success success partial partial

Evidence level per unlocked pattern

E2 = cross-validated against an independent method or literature benchmark · E1 = computational prediction only
E2
quantum_ground_state_portal
VQE result cross-checked against exact diagonalization and the published H₂/STO‑3G benchmark (−1.1373 Hartree) — the only pattern with two independent evidence types at once.
qfoldit_quantum_lab
E1
protein_folding_construction
Backend verified end-to-end; folding prediction itself not yet cross-checked against an independent method.
protein_design_mcp
E1
hp_lattice_spatial_puzzle
QAOA circuit run on a local simulator; no cross-model or benchmark comparison recorded yet.
qfoldit_qfold_lab
E1
admet_safety_strategy
Live, authorized third-party API — connected, not yet independently benchmark-validated. The least confirmed of the five despite being a real commercial service.
boltz_api
E1
l_systems_procedural_construction
Output field mapping flagged as best-effort from public docs, not independently inspected — first candidate for re-check once the schema is confirmed.
qfoldit_plant_growth_lab

CI & build verification

what was actually re-run this pass, not just read
UEFN-TOOLBELT verified
pytest tests/ → 117/117 passing
UNIGINE-TOOLBELT verified
pytest tests/test_mcp_server.py → 6/6 passing
WEB-TOOLBELT (src/) verified
npm ci && npm run build → clean (vite/client types fix applied)
UNITY-TOOLBELT unverified*
game-ci/unity-test-runner * needs a Unity license secret — not yet run by hand

Org repository inventory

27 public repos on github.com/qfoldit · grouped by role · highlighted = in this workspace archive

Core stack — in this workspace

  • UEFN-TOOLBELT
  • UNITY-TOOLBELT
  • UNIGINE-TOOLBELT
  • WEB-TOOLBELT
  • QFOLDIT-STANDALONE

Core stack — referenced, not in archive

  • Scientific-Object-Schema
  • Protein-Design-MCP
  • Atomic-MCP
  • UEFN-VERSE-MCP
  • science-skills
  • qfoldit.github.io
  • .github

Additional MCP servers

  • Game-MCP
  • Protein-MCP
  • qFold-MCP

Pre-publication / shared-only

  • engine-adapter-spec
  • scientific-gameplay-framework

Suggested next steps

ordered by impact, none blocking the others
Highest impact
Implement scientific.visualization for QFOLDIT-STANDALONE. This is the cheapest path to a third 5/5 target — the manifest already exists, and 2 of 5 patterns are blocked only by this capability. Work is just a scientific_visualization_create tool + a scientific_subject entry in UAG validator.
Quick win
Run the Unity CI workflow once by hand (workflow_dispatch) with a licensed Unity install — it's the one workflow in the workspace that hasn't produced a real green check yet.
Low risk
Reconcile the three drifting tool-count figures (org page vs. spec vs. direct count) by wiring a CI step that regenerates the count from source on every push, so it can't drift again.
Documentation sync
Update STATUS.md and ARCHITECTURE.md to reflect that WEB-TOOLBELT (src/) and QFOLDIT-STANDALONE now have qfoldit.adapter.json and compile measurements — all current docs still refer to them as "no manifest".
Feature
Port QFOLDIT-STANDALONE's renderer/ modules back into WEB-TOOLBELT (src/) — Standalone already has UAG bridge, physics joints, camera, lighting, animation, audio, asset loading. Copying those would close Web's gap in one pass.
Registry
Merge the two pending entries in 05-drafts-pilot/drafts/science_mcp_registry.additions.json once a live reachability check confirms them, per the schema's own gate rule.