Skip to content

roadmap.md

Latest commit

 

History

History
58 lines (43 loc) · 2.83 KB

roadmap.md

File metadata and controls

58 lines (43 loc) · 2.83 KB
Raw

Roadmap

The source of truth for planned / in-progress work is the Beads (bd) issue tracker:

bd list
bd ready

This docs/ folder keeps longer-form planning notes for context:

  • yolozu/data/manifest/adaptive_vision_roadmap.json — dated, packaged public projection of the adaptive local-vision scope and guardrails
  • reports/adaptive_vision_roadmap.md — human-readable report generated from that projection
  • docs/oss_support_scope.md — mandatory OSS boundary and explicitly optional or not-planned evidence work
  • docs/roadmaps/pytorch_trt.md — PyTorch/ONNX/TensorRT implementation notes (historical)
  • docs/roadmaps/yolo26_competition.md — YOLO26 toolchain goals (historical; see bd epic)
  • docs/roadmaps/symmetry_commonsense_realtime.md — Symmetry/commonsense constraint plan (historical)
  • docs/yolo26_size_buckets.md — YOLO26 n/s/m/l/x size bucket envelopes (params/FLOPs)

If bd list looks stale, refresh the local database from the exported beads-sync snapshot:

bash refresh_beads_sync.sh

For RunPod and two-machine details, see:

  • docs/beads_github_workflow.md
  • deploy/runpod/README.md (“Beads (bd) refresh on RunPod”)
  • deploy/runpod/refresh_beads_sync.sh

Adaptive local vision projection

The live task graph for environment-qualified local vision starts at Bead YOLOZU-ll2.81. Beads remains authoritative for issue status, dependencies, ownership, and completion.

The packaged adaptive_vision_roadmap.json is a dated public projection of product scope, target maturity, and safety boundaries. It intentionally omits live status and personal ownership data. The generated adaptive_vision_roadmap.md is a report, not qualification evidence.

The OSS support scope defines completion around truthful interface contracts, fail-closed behavior, privacy/provenance controls, and public maturity labels. Official multi-platform qualification, design-partner/on-prem validation, commercial reviews, and high-cost research benchmarks are optional evidence rather than mandatory OSS completion. Streaming, tracking, OCR runtime, and third-party isolated-runner implementation are not planned without renewed community demand and new Beads.

The current open-vocabulary and low-latency candidate screenings are holds, so their adapters and exact-host qualification are also not planned without a new managed pass and concrete demand. The enabled weekly scout and freshness jobs have successful manual artifacts. Their first cron events remain operational observations and must not be reported as verified before they occur, but they do not block OSS roadmap completion.

Regenerate or verify the report with:

python3 tools/generate_adaptive_vision_roadmap.py --check --json