Historical repository record migrated on 2026-10-10. Original repository path: docs/experiments/benchmarks/performance-frontier-admission-01.md Original SHA-256: 5b3fcb4e6ca1d3941371e8f06db5979e77b3a07ebb953502cd579e2714cd316b Stated outcomes, measured revisions, failures and limits remain historical. Local raw evidence is private and is not uploaded. Relative links and the data-drive prefix were normalized.

Performance frontier: admission attempt 01

Date: 2026-10-05. Outcome: environment unavailable; workloads not run. Plan: next labs. HEAD: 57aad2972d5238556ed33d3d1d9fff58fa0dc542, with existing working-tree changes.

The execution preflight used the required launcher:

python3 -B tools/resource_group.py -- /usr/bin/true

Exit status: 2. Exact output:

resource group: NOT RUN: [Errno 30] Read-only file system: '$FABRIC_DATA/scratch/work-g4enc25l'

Failure occurred creating owned scratch, before starting a service or the child command. No build, experiment or validator ran. No workload CPU/RSS/acceptance measurements exist for this attempt. No scratch was created by this failed creation; historical scratch and failure evidence were left untouched. There was no fallback to another drive or an uncapped process.

Read-only prerequisite review

• The spill protocol requires the same binary as the original C2 runs. completion/spill_measure.py currently records the selected binary's hash but checks only input and manifest equality against the prior run. Admission must compare the original frozen binary hash before this observer runs. Rebuilding into the shared example path first risks invalidating the required comparison. This is an identified harness prerequisite, not a measured performance defect.
• Original storage evidence covers five tests. Current completion_storage.rs additionally has missing-query-table, span write/sync fault, and full-chain pre-publication-pause cases; they remain uncompiled and unrun in this attempt.
• completion/run_job.py is tied to the earlier completion campaign's receipt directory and budgets. Existing registered prerequisite runs can use it with new IDs. New frontier comparisons need their own frozen protocol and explicit four-hour campaign accounting before admission; the older 24-hour allocation must not silently replace the new plan's smaller allowance.

Resume order: restore permitted data-drive/cgroup execution; check original binary provenance; complete registered spill and storage prerequisites; run relevant repository checks; then register/admit the new comparative cells. Protocol registration still requires writable Git metadata. No protocol or historical acceptance gate was changed during this attempt.

Subsequent access retry

After the session filesystem permissions were restored on 2026-10-05, the same command exited 0 under service fabric-work-803b3980d6ee44fe81b2368f4b40f538. The launcher created data-drive temporary storage, verified actual memory.max=21474836480 and memory.swap.max=0, and completed its child. Systemd reported 40 ms service runtime, 29 ms CPU and 7.1 MiB peak memory. The launcher removed its owned temporary directory on success. The receipt is under target/resource-containment/runs/ with the service name above. This clears the storage/containment access blocker; it is a preflight result, not execution of any lab experiment. The original failed attempt remains recorded.