# Roadmap
This roadmap is the public, contributor-facing view of where rvt-rs is headed.
The support boundary lives in [`docs/status.md`](docs/status.md). The full task
decomposition lives in [`TODO.md`](TODO.md) and the matching GitHub issues.
## Product Goal
rvt-rs should become a first-class open-source utility for BIM and AEC users who
need to inspect, validate, and exchange Revit files without installing Revit or
uploading private models to a third-party service.
The project is not there yet. It is currently a strong Revit inspection and
reverse-engineering toolkit with a partial IFC/viewer path. **Generic
real-project typed element extraction remains mostly unsolved**, with
narrow exceptions: production `walker::iter_elements` prefers typed MVP
decoders on `Global/Latest`, merges version-gated **ArcWall** partition
recovers (Revit 2023 standard), and merges fail-closed partition MVP
recovers for **Level** / **Material**, 2024
**ArcWallRectOpening** index rows (ElemTable-confirmed related ids;
never inventing typed `Door`/`Window` success), and 2024 partition
element-record **Wall** / **Door** / **Window** / **Column** /
**Floor** / **BuildingPad** instances (#204 / #211 / #212), plus, on 2024
and 2025, furniture, casework, plumbing fixtures, specialty equipment,
ceilings, curtain-wall mullions and panels, railings, wall sweeps, ducts and
pipes (RE-33), structural framing, structural columns, structural
foundations and generic models (RE-36), and lighting fixtures, air
terminals, food-service equipment, planting, parking, entourage, hardscape,
elevators and ramps (RE-37). IFC export
wires partition **Level** → storeys, **Floor** / **BuildingPad** →
`IFCSLAB` (or `IFCSHADINGDEVICE` under a per-element Revit export
override), **Room** → `IFCSPACE`, and **Material** → `IfcMaterial`: on 2024 and
2025 the file's own material records, named by ElementId with their
colours (RE-58, #34), and display-name strings before that.
**RE-19 (2026-08-29) negative on magnetar corpora:** no reliable Door vs
Window discriminator in the 2024 opening index / nearby partition strings /
ElemTable payloads, and no schema-field `Wall` / fail-closed 2024 ArcWall
envelope — see
[`reports/element-framing/RE-19-door-window-wall-negative.md`](reports/element-framing/RE-19-door-window-wall-negative.md).
Host IFC voids/fills, Floor↔ElemTable id binding, and slab extrusion
thickness remain open. Eighty-one per-class decoder structs exist in
`elements::all_decoders()`; MVP classes are consulted by `iter_elements`,
while the broader registry remains a library building block (see
[`docs/status.md`](docs/status.md) and
[`docs/compatibility.md`](docs/compatibility.md)).
## Current Position
| Area | Current state | Next decision point |
|---|---|---|
| Container, compression, metadata | Shipped | Maintain compatibility and bounds checks. |
| `Formats/Latest` schema | Shipped | The whole schema is read since #410; classify the 9 to 12 residual field encodings per release with byte evidence, and take class tags from the definition ordinal (#154). |
| ADocument/document-level walker | Partial | Expand confidence across project releases and older files. |
| Typed project elements | **Partial** | MVP typed path + ArcWall + partition Level/Material + 2024 opening index (ElemTable-confirmed ids) in `iter_elements` (fail closed). Every `DecodedElement` carries M3-07 provenance/confidence (CLI/Python/viewer + default IFC hide below 0.55). RE-19: no Door/Window discriminator / no schema-field Wall. RE-20: no Level ElementId map in the opening-index bytes; its note that `Level` is absent from Formats was an artifact of the 64 KB schema scan (#410: `Level` is class 2422 of Revit 2024's schema). Storeys come from element records instead (RE-24, RE-27, RE-51). |
| IFC writer | Partial | Levels/Floors/Rooms/Materials from partition MVP emit honestly; ArcWall geometry on 2023; Door/Window host IFC + slab extrusion still open (blocked on RE-19); Floor/Room storey bind idle (RE-20). |
| Browser viewer | Partial | File Status lists recovered storey names + material display-name samples + honest Parameters row (empty until AProperty host joins — RE-20: no AProperty carriers on magnetar Global/Latest); scene tree groups under `IFCBUILDINGSTOREY` (ArcWalls by elevation; Floors/Rooms stay Unassigned — RE-20 negative on Level ElementIds). |
| Python/CLI surface | Partial | Stabilize JSON schemas and one-shot inspect workflow. |
| Write path | Partial | Keep stream-level writes honest; defer semantic writes until openability can be proven. |
## Milestones
### 0.2.0: Audit-Clean Alpha (inspection-focused)
Purpose: make the repository easy to trust and easy to contribute to before
deep decoder work accelerates. Positioning detail:
[`docs/release-0.2.0-plan.md`](docs/release-0.2.0-plan.md) — **inspection
alpha with experimental export**, not converter-grade.
- One-command local quality gate.
- Explicit cargo-audit/cargo-deny expectations.
- README, roadmap, compatibility, and status docs aligned.
- GitHub issue forms for decoder work and corpus submissions.
- Contribution map for non-maintainers.
- Release artifact verification documented (honest crates.io / docs.rs /
PyPI status).
### 0.3.0: Real-Project Wall/Floor MVP
Purpose: prove that rvt-rs can recover meaningful typed building elements from
real project files, not only synthesized fixtures.
- Redistributable project corpus with license metadata.
- Known-count fixtures for levels, walls, floors, doors, and windows.
- Generic partition record scanner.
- `ElemTable` id to partition-record offset linkage.
- Typed MVP decoders + ArcWall + partition Level/Material / 2024 opening-index (ElemTable-confirmed) / 2024 element-record Wall/Door/Window/Column/Floor/BuildingPad merge wired into `iter_elements` without false positives; IFC Level/Floor/Room/Material emission (slab plan profiles recovered from OST_SketchLines records, #31/RE-25; wall bodies join-trimmed and column sections joined to their family type, #215/RE-26, with the joins themselves read from the `+0x88` reference list and column bodies cut by the walls they name, #238/#239/RE-29; schema-field Wall and typed Door/Window host binding still open).
- Decode confidence and provenance attached to every element.
### 0.4.0: IFC Geometry Beta
Purpose: export useful IFC only when decoded evidence is strong enough.
- Explicit export modes: strict, proxy, and diagnostic.
- No misleading generic proxies in default export.
- IFC diagnostics sidecar describing decoded/skipped elements.
- IfcOpenShell validation for generated outputs.
- Comparison tooling against Revit-exported IFC when fixtures allow it.
### 0.5.0: Viewer Beta
Purpose: make unsupported states clear to non-technical users.
- Decode/export confidence surfaced in the viewer.
- Supported-file guidance before export.
- Demo gallery using redistributable files.
- Browser regression tests across desktop and mobile viewports.
- Accessibility and responsive layout pass.
- Desktop distribution investigation.
### 1.0.0: First-Class Utility
Purpose: ship a complete, honest workflow that gives meaningful value to users
without requiring them to be Rust, Revit API, or reverse-engineering experts.
- Supported input profile documented in user language.
- End-to-end open -> inspect -> diagnose -> export workflow.
- Actionable failure modes for unsupported files.
- Non-technical documentation and screenshots.
- Release artifacts verified and reproducible.
## Contribution Priorities
Start with [`docs/contribution-map.md`](docs/contribution-map.md). The highest
leverage work is:
1. Redistributable corpus files with known counts.
2. Partition-stream probes that turn byte observations into falsifiable decoder
hypotheses.
3. Tests that prevent `iter_elements` from claiming false positives.
4. Documentation that keeps user-facing support boundaries honest.
5. Viewer diagnostics that explain what the tool could and could not decode.
## Out of Scope
rvt-rs will not:
- Use Autodesk proprietary SDK internals, leaked documents, or decompiled
proprietary implementation code. See [`CLEANROOM.md`](CLEANROOM.md).
- Claim production RVT-to-IFC conversion before real project typed elements and
geometry are corpus-proven.
- Provide a Revit API-compatible surface.
- Resolve cloud-worksharing, licensing, or external linked-model semantics in
the near-term product.