Storage-aware workspaces

Parked as competition candidate; final shortlist unsigned
Provisional idea; no final shortlist has been approved.Status as of 2026-10-15 09:27 UTC.
FIG. A16 — THE PILE IS MOSTLY DEPENDENCY SHEETS, NOT SOURCE
Storage-aware workspaces: conceptual diagram of the hypothesis, not a measured result

01 · PROBLEM

Alexey’s linked worktrees grew quickly on disk. A scan of one host found 111.7 GiB across 472 worktrees, with dependencies and builds accounting for 62.1% of the physical union.

Who: Alexey, and operators constrained by local disk.

02 · HYPOTHESIS

Account for source, immutable dependencies, mutable outputs, caches, and logs separately. Compare ordinary package sharing, sparse or copy-on-write worktrees, and remote execution without hiding allocation costs.

03 · EVIDENCE SO FAR

LATEST3 October: the principals proposed parking the local competition product after repeated whole-footprint gate failures. The latest fixture measured 47.76% savings against a greater-than-50% gate, with incumbent clone 47.38% near-equal. These tiny fixtures do not establish host-wide savings or safe removal of existing worktrees; remote mode remains unvalidated.website/projects.json

Earlier research, from before 3 October 2026, 02:24 UTC. Kept for context; it is not the current status.

SUPPORTSBEFORE 3 OCTRead-only scan of the real host: 111.7 GiB physical union; Python .venv 57.3 GiB across 215 dirs, sharing nothing.research/claude/u7-real-worktree-measurement.md
AGAINSTBEFORE 3 OCT264 worktrees sit on commits already in origin/main. Cleanup, not a platform, may be the bigger lever.research/claude/u7-real-worktree-measurement.md
AGAINSTBEFORE 3 OCTAn ordinary pnpm union is already ~49.97% below summed individual trees. Strong incumbent baseline.research/codex/package-storage-validation.md

04 · NEXT FALSIFICATION TEST

Keep useful storage advice and read-only diagnostics. Reopen a product only with new user/task evidence beyond ordinary sharing and lifecycle tools; do not move the gate or extrapolate a passing third-workspace result.

StatusParked as competition candidate; final shortlist unsigned
We would drop or park it ifIf ordinary package and workspace lifecycle tools solve the pain with less complexity, an Artifacts-specific local product may not be needed. A useful workspace diagnostic could still survive.

05 · PUBLIC SOURCES

research/claude/u7-real-worktree-measurement.mdresearch/orchestrator/worktree-storage-pain.mdresearch/codex/storage-validation.mdresearch/codex/package-storage-validation.mdresearch/codex/retained-lanes-review-2324.mdresearch/antigravity/r8-physical-storage-isolation.md