vxedge result archives: raw per-run archives behind the published results

Raw per-run measurement archives for the vxedge experiments (isolation, slicing, packaging, OTA and topology). Each campaign is one tarball with a sha256 checksum. Every run archive records its input manifest digest in metadata.json. The summaries in vxedge/vxperiments record the digests of the frozen campaigns, and vxedge/docs FACTS.md registers the four behind the hypothesis verdicts. The record holds 24 tarballs and SHA256SUMS. The tarballs contain 435 run archives collected between 2026-07-12 and 2026-10-30, of which 375 are distinct. Campaigns exp-001: isolation, VLAN and tc reservations under infotainment load (H1) exp-002 and exp-002b: virtual-ECU packaging and its decomposition (H3) exp-003: co-located OTA and the ADAS slice (H4) exp-005: cold-start first-packet latency, one dated rerun archive exp-010: S-NSSAI slicing against static VLAN mapping (H2) exp-014: star against daisy-chained zones, one dated rerun archive Naming Pattern Meaning exp-NNN.tgz The frozen campaign: the published numbers come from this exp-NNN-repro-<date>.tgz A rerun made on that date exp-NNN-quick.tgz The ~1 minute pipeline check: not reportable, kept for completeness exp-010-no-qos-*, exp-010-qos-fix-unmitigated-* Earlier arms of the slicing experiment, kept as history Three rerun tarballs hold run archives byte-identical to a frozen campaign: exp-001-repro-2026-08-03 to exp-001, exp-003-repro-2026-08-03 to exp-003, and exp-010-repro-2026-08-05 to exp-010. They are not independent repeats of those campaigns. New in this version exp-003-repro-2026-10-30 and exp-010-repro-2026-10-30 are new full campaigns repeating H4 and H2. Both reached the same verdict as the frozen campaigns. H4: ADAS p50 was 56.2 µs at baseline and 54.6 µs during the OTA burst, a change of -1.6 µs against a threshold of 1.9 µs, with no packet loss. 3 of 6 metric and arm medians fall inside the frozen campaign's IQR. H2: p99 growth from idle to maximum load was 1.00× for S-NSSAI and 0.91× for VLAN, inside the 1.1× tolerance by a small margin. p99 was 339 to 356 µs against 317 to 337 µs in the frozen campaign, and median loss was 0.0033 % at 400 Mbit and 0 at the other loads. 3 of 15 metric and load medians fall inside the frozen campaign's IQR. Both reruns used a newer kernel than the frozen campaigns, and other virtual machines were running on the host. The input digests differ from the frozen campaigns by comment edits and, for exp-003, a resource sampler added to the runner. Verify and use sha256sum -c SHA256SUMS # every line must say OK tar xzf exp-001.tgz # one directory of per-run zip archives Analyse the archives with the notebooks in vxedge/vxperiments (notebooks/analyze_exp001.py and its siblings). They read the archives only, never live measurements.