Skip to main content

​
ARCLUX — Roadmap Blueprint

Long-term direction. Not decisions.md (why past choices were made), not status-*.md (current implementation state). This is where ARCLUX is headed, and why.

​
Core Principle

ARCLUX stays a deterministic structural-truth engine. No AI/ML inside core packages — ever. Every fact ARCLUX reports must be traceable to real parsed code: an import statement, an export declaration, a resolved path. Not a guess. Not a probability. A fact. Intelligence/semantic layers are welcome — as a separate consumer package (packages/intelligence/), never woven into engine/, parser/, graph/, or detectors/. See ARCHITECTURE_MAP.md for the enforced boundary.

​
Phase 0 — Reliability

Status: In Progress The foundation has to be trustworthy before anything gets built on it.
  • Python parsing/edges silent-failure bugs (wasm path root cause found and fixed — follow-up verification may still be needed)
  • Hardening: does the pipeline hold up against large real repos, not just fixtures
Status: Not Started Search today matches filenames (fuzzyScore.ts). That’s not enough. A match should surface structural context ARCLUX already computes: imports, exports, consumers, dependencies, affected routes — not just “here’s a file with that name in it.”

​
Phase 2 — Architecture Health Score

Status: Not Started 20 detectors already exist. Right now they produce a flat list: “17 problems found.” That’s a report, not a diagnosis. Turn detector findings into a health view: Structural integrity 94% Dependency hygiene 88% Layer consistency 91% Dead code 97% Risk areas packages/auth — 4 high fan-out modules packages/api — circular dependency chain Every number here comes from real detector output. No AI-invented scoring, no subjective weighting from nowhere.

​
Phase 3 — Optional Intelligence

Status: Not Started, no urgency A new packages/intelligence/ package for anyone who wants to build semantic search, embeddings, or RAG on top of ARCLUX’s structural output. Open to contribution (e.g. from collaborators building their own intelligence layers elsewhere) — not a priority for the core team to build proactively.

​
Phase 4 — External Integrations

Status: Later, if ever LSP bridge, MCP/agent interface, IDE integration. Deliberately last. Bringing this in early risks blurring who owns the source of truth for parse → index → graph → impact → detect.

​
Vision (no phase assigned)

  • TUI — a third consumer of engine/, alongside CLI and web. Terminal-native, keyboard-driven. See decisions.md for the original discussion.

​
License

This roadmap, like the rest of ARCLUX, is released under the Apache License 2.0. See LICENSE at the repo root. Nothing here is a promise or a commitment — it’s direction, not a contract. Phases can reorder, merge, or drop entirely as real usage teaches us something the plan didn’t anticipate.