{
  "frame": "CPI-50K v1",
  "series_start": "2026-08-23",
  "frame_adopted": "2026-08-21",
  "reason": "Editions before 2026-08-23 were measured on the default Tranco list, which includes Cloudflare Radar (CC BY-NC) and Chrome UX Report (CC BY-SA). The non-commercial clause restricts use, not only redistribution, so those editions are excluded from everything published or sold.",
  "bridge": {
    "note": "The frame change did not move the measurement. Measured on the domains readable in BOTH frames, the underlying figures were unchanged.",
    "common_domains": 17179,
    "any_block_before_pct": 20.5,
    "any_block_after_pct": 20.57,
    "asymmetry_ratio_before": 18.9,
    "asymmetry_ratio_after": 18.9
  },
  "unscheduled": [
    "2026-08-24"
  ],
  "unscheduled_reason": "Real measurements that are not scheduled census editions. 2026-08-24 was the Raspberry Pi cutover validation run (21:29 UTC on a Monday; 31,562 domains parsed). Every census starts at 01:00 Europe/Luxembourg (23:00 UTC the evening before) on Sunday and Wednesday and finishes around 03:00 UTC. The file and its anchor are retained and the data is sound, but publishing it as an edition would imply a Monday census that never happened, so it is excluded from the trend, the comparison baselines, the change sets and the public series.",
  "withdrawn": [
    "2026-09-05"
  ],
  "withdrawn_reason": "Editions that were published and are NOT sound. Kept separate from the unscheduled list on purpose: an unscheduled run is a real measurement taken at the wrong time and its data stands (2026-08-24); a withdrawn one is a measurement that should never have been published at all, and nothing may be derived from it. 2026-09-05 was produced by a retired daily job on a laptop that had been advancing a single sweep in slices since 11 August; when the last slice completed on a Saturday the job ran the full publish chain. The result is a sweep smeared across twenty-five days rather than a snapshot of one night: 29,672 domains with a readable robots.txt against the census's 31,488, every one of the eighteen crawler block rates about a third of its true level, a crawler_states map holding one key instead of eighteen, and both wire legs absent. It is excluded from history-index.json entirely rather than filtered downstream, so no generator can reach it. The raw artefacts are in _quarantine/2026-09-05/ on the census machine and in git at commit 5924150. Corrected publicly in corrections.json as 2026-09-05-08."
}