MoonkaleThe problems foreseen before building, ranked. Difficulty 1–5 (effort + unknowns); Risk = how much else breaks if it goes wrong. Problems actually hit while building are logged in the Problem Log as P-nnn; what is being worked on next is the Roadmap; the current state per area is Status.
Two id spaces
Until 2026-10-01 these rows were numbered
P-01 … P-35, which collided with the Problem Log’sP-032 … P-035(P-32 “git” here, P-032 “release-build observability” there). The ranking now usesR-nneverywhere in the vault; the log keepsP-nnn.
Status, as of Milestone 17 (2026-09-29): ✅ done · ◐ partly done (what is missing is said) · ○ open.
| # | Problem | Diff | Risk | State |
|---|---|---|---|---|
| R-01 | Core graph model + Source trait + query IR | 3 | very high | ◐ Node/Edge/Source/Query exist and every source uses them; not built: GraphView (query + watermark), node/edge properties, Source::subscribe (changes_since long-poll instead), the lifting rules in sources::lift — Graph-Native Model |
| R-02 | Folder source (native) + explorer panel | 2 | low | ✅ watched since Milestone 16 |
| R-03 | JS interop protocol (eval + channel) on all webviews | 3 | medium | ✅ web, WebKitGTK, Android WebView (P-047 rules) |
| R-04 | Code editor with CodeMirror behind CodeEditorBackend | 3 | medium | ✅ splices since spec 018 |
| R-05 | Extension API + static registry; built-ins as extensions | 3 | high | ◐ Extension trait (panels, commands, settings, flow libraries), tiers, permissions. Not built: moonkale.toml, activation events, when clauses, contribution points beyond panels/commands — and ui still hard-codes the catalogue (Extension Catalogue critique) |
| R-06 | tree-sitter index + wiki-link extraction | 3 | medium | ◐ native tree-sitter (arborium), wiki-links, Rust/Julia/Python symbols (Julia/Python only since the #16 fix, 2026-10-02); other languages, calls and references open |
| R-07 | Graph view v1: wgpu 2D, WebGL2/WebGPU, CPU layout, pick/popup | 4 | high | ✅ |
| R-08 | SQLite + DuckDB sources + table editor | 3 | low | ✅ read-only |
| R-09 | Markdown editor: source mode, links, backlinks, local graph | 2 | low | ✅ |
| R-10 | Remote source via api (web/mobile parity) | 3 | medium | ✅ token auth (M7), TLS (M11) |
| R-11 | Milkdown WYSIWYG behind RichTextBackend | 3 | medium | ✅ |
| R-12 | Typst preview | 2 | low | ✅ no packages |
| R-13 | Terminal: PTY + xterm view + links | 2 | low | ✅ plus a Rust twin (M12) |
| R-14 | LSP client + local spawning | 3 | medium | ✅ full-document didChange still (issue #19) |
| R-15 | Graph DB sources: Ladybug, Falkor | 2 | low | ◐ LadybugDB (Linux only since M17, P-144); FalkorDB not started |
| R-16 | Cross-source edges + consistency | 4 | high | ○ needs projects (Projects and Sources) |
| R-17 | LLM gateway: providers, tool surface, policy, audit | 3 | medium | ✅ Claude Code, Anthropic, OpenAI-compatible, Ollama, mock |
| R-18 | Embeddings + hybrid search | 3 | medium | ✅ BM25 + brute-force cosine, in memory |
| R-19 | Postgres/Supabase, Turso sources | 2 | low | ◐ Turso embedded (M17, read-only); Postgres not started |
| R-20 | Remote LSP + remote terminal on api (security) | 4 | high | ◐ one token per server; the audit found gaps (Audit 2026-09-23); no accounts |
| R-21 | Stack-trace / AST → graph | 2 | low | ◐ traces; ASTs not |
| R-22 | GPU compute layouts; 100k+; desktop surface (P-001 Graph surface in desktop webview) | 5 | very high | ◐ Barnes–Hut on the CPU reaches 100k; surface plan A in use; GPU compute and the coarse tier (Case Selector) open |
| R-23 | wasmtime extension runtime + WIT world + permissions UI | 4 | high | ◐ JSON ABI v1 over core modules (ADR-0013 JSON ABI before components); no WIT, no UI contributions, no fuel limits (issue #4) |
| R-24 | Flow editor + Lux.jl codegen | 3 | medium | ◐ editor + codegen (M6); running the model with errors on blocks open |
| R-25 | Mobile: file access, collapsed shell, touch | 4 | medium | ◐ Android build, phone shell, gestures, touch drag; Storage Access Framework open (spec 028) |
| R-26 | TypeDB, HelixDB sources | 3 | medium | ◐ HelixDB embedded (M17, read-only, P-145); TypeDB not started |
| R-27 | 3D graph | 3 | low | ✅ planes per kind, orbit camera |
| R-28 | Browser-side WASM extensions | 5 | very high | ✅ Worker + SharedArrayBuffer, JSON ABI |
| R-29 | Rust-native backends (terminal → code → rich text) | 5 | research | ◐ terminal (M12) and code (M14) as opt-in twins; rich text none |
| R-30 | Collaborative editing (CRDT over the op stream) | 5 | research | ○ |
| R-31 | Structured Typst / Excalidraw-in-flow | 4 | research | ○ |
| R-32 | Git integration: status/diff, commit/log, history as a graph | 3 | medium | ✅ through the git CLI (not gix); no push/pull |
| R-33 | Entity log: events, fold, snapshots, checkpoints ↔ commits | 4 | high | ✅ JSONL per folder, compaction (M9); issue #17 open |
| R-34 | Presence: awareness, hub, desktop + web | 3 | medium | ✅ rooms, cursor lines, desktop hub client (M9) |
| R-35 | Windows / macOS / iOS builds | 3 | medium | ◐ CI builds Windows and macOS packages; a friend tests macOS; no iOS |
| R-36 | One store for internal state (settings, layout, history, sessions, index) instead of a dozen files, ready to sync | 4 | very high | ○ candidates in Milestone 17; comparison pending — Internal State, ADR-0014 One store for internal state |
| R-37 | The core as a library: ui without extension or driver dependencies, Workspace split into services, a server contribution point, reusable crates (graph renderer, core) | 4 | high | ○ planned — Milestone 18 - Library Refactor |
| R-38 | Security findings of the external audit (issues #1–#5, #7–#10, #18, #20) | 3 | high | ○ — Audit 2026-09-23, Security |
| R-39 | CI that runs the tests (none did until 2026-10-01) | 2 | high | ◐ ci.yml: fmt, layering rules, tests, clippy, wasm check, six browser suites; macOS/Windows/Android not tested in CI — Milestone 18 - Implementation Log |
| R-40 | Projects: several sources saved as a project, selector, sync | 4 | medium | ○ desired behaviour in Projects and Sources |
| R-41 | Writes to databases (OLTP) — every database source is read-only | 3 | medium | ○ announced after Milestone 17 |
Reading the table
- What is left of the original top risks: R-01 (the model’s missing half), R-05 (the extension API’s missing half), R-16, R-20 and R-22. R-36 and R-37 are new and have the widest blast radius — every crate touches the state and the
Workspace. - R-37 and R-39 come before new features: they are what makes the rest cheap to change. R-38 comes before any server is exposed beyond a trusted network.
- Anything marked research has no committed date.