Roadmap¶
WorldForge is pre-1.0. The roadmap is intentionally capability-driven rather than feature-count driven: every public surface should stay truthful, typed, reproducible, and host-owned where upstream runtimes or robot controllers are involved.
Current Focus¶
- Treat the provider and platform production track as complete baseline infrastructure; use fresh selection records for new provider batches instead of extending the original tracker.
- Keep the provider catalog honest: adapters advertise only capabilities implemented end to end.
- Make checkout validation boring: one command should exercise lint, docs, tests, coverage, package build, package install, and dependency audit.
- Preserve reproducible artifacts for evaluation and benchmark claims.
- Keep optional model runtimes out of the base package while making their wrapper commands clear enough to run on a prepared host.
- Keep the robotics showcase TUI focused on prepared-host policy+score evidence while core world, provider, eval, and benchmark workflows remain CLI-first.
Near-Term Milestones¶
| Area | Milestone |
|---|---|
| Provider adapters | Promote scaffold adapters only after upstream-runtime contracts, fixtures, and failure modes are validated. |
| Benchmarking | Attach provenance and preserved input fixtures to any published benchmark number. |
| Evaluation | Expand suite coverage while keeping scores framed as deterministic contract signals. |
| Harness | Continue world editing, run inspection, report export, and provider diagnostics through optional Textual screens. |
| Release engineering | Prefer signed, attested, tag-verified releases and trusted publishing. |
For historical issue planning, use the detailed Provider And Platform Roadmap. That first production track is complete and should be treated as baseline infrastructure.
For the next implementation batch, use the Roadmap Continuation. It narrows the project to three current streams: provider evidence and runtime cohorts, evaluation evidence and claim integrity, and operator workflow plus adapter authoring.
For the expanded next roadmap batch, use the Roadmap Expansion. It defines 30 structured issues across production grade quality/DevX/docs, demos and end-to-end showcases, and new features. Nano World Model work is excluded because that issue is already assigned.
For the second expansion batch, use Roadmap Expansion 2. It adds 30 follow-on issues that deepen artifact schema governance, docs/test quality gates, external-provider demos, capability preflight demos, scenario reuse, regression reporting, runtime asset manifests, and composed workflow traces.
For the third expansion batch, use Roadmap Expansion 3. It adds 30 more issues that push toward pre-1.0 maturation: static typing and public-API guardrails, stable doctor and error-code contracts, anchor/CHANGELOG integrity, adoption-oriented showcases (notebooks, recorded tours, narrative walkthroughs, an adopt-your-robot demo), and feature deepening (cost accounting, named snapshots, run retention, capability-aware deltas, audit logs, scenario inheritance, a Prometheus exporter, trace redaction policy, a preview probe, and a pluggable persistence backend interface).
For the active provider evidence cohort, use the Provider Cohort Selection Record. It selects the next provider work items, records explicit deferrals, and keeps public provider capability claims unchanged until implementation evidence exists.
Non-Goals¶
- WorldForge will not bundle LeWorldModel, LeRobot, GR00T, torch, CUDA, checkpoints, datasets, or robot controllers into the base package.
- Scaffold adapters will not be presented as real integrations.
- Deterministic evaluation or benchmark outputs will not be used as physical-fidelity claims.
- Local JSON persistence will not be treated as a service-grade concurrent datastore without a separate design.