Wirewalk public lab toolkit — version 1.0 (2026-11-21) https://www.wirewalk.com/lab/ Python 3.9+; standard library only. MIT license; see LICENSE.txt. Download, inspect source, then run locally. No email signup or automatic upload. 1. inventory.py python3 inventory.py --out baseline.json Read-only Linux baseline; optional local nvidia-smi query with an 8-second timeout. No privilege escalation, commands through a shell, network requests or config changes. A new output file is created with mode0600. Existing files are not overwritten. Addresses, hostnames, serials and GPU UUIDs are omitted; review all output before sharing. Missing tool/query errors are distinct from an observed device list. This is not a complete health check or performance test. 2. counter-delta.py python3 counter-delta.py counter-samples.csv --unit-bytes 4 Offline CSV processing. Header: timestamp_seconds,device,counter,value Each device/counter requires at least two samples. Units must be verified explicitly. Rejects bad samples, duplicate timestamps and decreasing counters (possible reset/wrap). Raw counter values are integers; rates use Decimal arithmetic. Included values are synthetic, not hardware measurements. Result: rail-a 1 GiB/s, rail-b 0.5 GiB/s. 3. validate-manifest.py python3 validate-manifest.py benchmark-manifest.json Checks required structure, selected field types/ranges and result-state consistency. Does not run benchmarks, inspect referenced evidence files, verify placeholder replacement, attest scientific correctness, approve an acceptance criterion or certify a result. The sample is intentionally not_run/not_measured. Fill in actual evidence before use. 4. acceptance-checklist.csv Local worksheet with ownership, evidence and explicit not_run state. Adapt to the accepted scope. It does not replace a written acceptance plan. Read the guides and each script's --help before use. The site includes complete explanations, version-sensitive command caveats and related primary documentation.