> ## Documentation Index
> Fetch the complete documentation index at: https://arclux-os.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# roadmap

> Detail progres: roadmap

# 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

***

## Phase 1 — Structural Search

**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**

19 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`](../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.