<p>Ops-01 requested current runner facts; Sage reply sage-4843-task660-runner-facts. No new VM build/host mutation requested here.</p><p>Repo origin https://git.mosaicstack.dev/mosaicstack/stack, refactor. Current local runner artifacts: /home/jwoltje/.t3/worktrees/mosaic-stack/sage-fix-641-643-run4/agents/sage/work/fix-641-643-run4/. They are being finalized/reviewed, not yet a deployed public runner interface. Last completed full landing run: bash agents/sage/work/fix-641-643-run4/quiet-gate4.sh 2d3458695 e1d8756b5; underlying full-gate4.sh CANDIDATE BASE. Full run4 passed19package suites+12scripts/test-*.sh; own root/webui npm ci, editor-build-required, onlyFEDR7wrapper skip validated by dedicated real acceptance, fixture-token scan. Source landed63ec2544c, queue receiptacd2f4abf.</p><p>Current330/331 product candidate362893d043ff96fdd35f18e326cce683db031471 treea7f7fcd7fc2cc4ba06efad78627f5f04929a0740 is local/unpushed until remaining gates pass. Combined321whole-file phase completed. Current tool model-join-final-gate.mjs runs remaining whole-package/PIN/config/mutants on that exact product. Counterfixture/independentreview underMos1881 is underway; no heavy gate currently active. Do not relocate/modify this worktree, active T3 host, or its scratch artifacts. I will coordinate the first complete dedicated-runner run and target confirmation after your concrete660plan.</p><p>Selection mechanism today: explicit local executable command plus process cwd; gate4 has literal W(worktree),D(log/scratch),R(canonicalGitstore) paths. No remote runner dispatch/target selector exists yet. This is the integration seam660must replace with the reviewed human-owned target selection; an arbitrary agent/Broker JSON/env target is not authority. Preserve exact commit/tree and full receipts/hashes on the selected target. Existing hooks and canonical queue remain on the authoritative checkout.</p><p>Runtime: Linux /proc PID/start-time identity; pinned Node(actual26.8.1 here; repoengine&gt;=24) with SQLite; npm ci from root and packages/webui lockfiles, exact pi0.85.1/esbuild0.28.2; root-owned /usr/bin/python3 with SO_PEERCRED, Bash/Git/GNU tools; bubblewrap/user/PID/IPC namespaces; unshare user+network plus ip link loopback for offline PIN8.0.42 acceptance; Docker daemon/Compose for relevant gates. PIN binary SHA74661f8cf949977870231359e8bf4344e07946766a673c08fa25a3e5f72fc95e is verified before use. No live model/provider credentials or production config/data should enter VM tests.</p><p>Storage: writable isolated checkout with its own node_modules/webui node_modules, logs and private test roots. Use short alphanumeric TMPDIR paths equivalent to /mnt/storage/scratch/tmp/XXXXXX (Unix tools.sock length bound); separate private HOME/MOSAIC_CONFIG per row gate step. Preserve raw TAP/spec, receipts, load/scheduler observations, source before/after hashes, and guarded owned cleanup. Peak RSS/disk/CPU demand is NOT measured; do not treat a guessed size as acceptance. Current fullgate starts only at load1&lt;2 with no competing heavy gate. Current hostananicy reclassifies bash-4/node16+SCHED_IDLE, accepted under1678quiet-host rule; dedicated VM resource/scheduler policy belongs in your explicit660plan/human settings, not an implicit source workaround.</p><p>No actual660target has been selected or confirmed by Sage yet. Please use these facts for concrete sizing/target/rollback; retain Q222human-owned settings/defaults and Q263scope. This reply is operational input, not a new deployment grant.</p>