q6/q8 need the cumulative A(H5N1) human-case source (HAI summary links it, doesn't state it)

[smodee]

Summary

BFG q6 (human H5 cases reported to WHO in the Jul–Dec 2026 window) and q8
(predominant H5 subtype) need the cumulative confirmed-human-case count and the
per-subtype breakdown. The who_h5_hai scraper added in #53 correctly resolves the
WHO "Influenza at the human-animal interface" monthly summary PDF — but that
document reports the reporting period's events (e.g. the current issue is an
A(H9N2) cluster in China) and only links to the cumulative A(H5N1) total; it
does not state it. So q6/q8 still lack their anchor.

The cumulative figure lives on a separate WHO page the summary points at:
https://www.who.int/teams/global-influenza-programme/avian-influenza/avian-a-h5n1-virus
("Cumulative number of confirmed human cases of avian influenza A(H5N1) reported to
WHO") — a ~7 KB page whose count sits in a table.

Evidence (evidence-coverage audit, live 2026-07-05)

• q6: 4 dashboards survive → 1 record (the H9N2 China event, not an H5 total).
• q8: 4 dashboards survive → 0 records.
• The HAI PDF prose explicitly reads: "…Cumulative Number of Confirmed Human Cases
  of Avian Influenza A(H5N1) Reported to WHO https://www.who.int/.../avian-a-h5n1-virus"
  — i.e. it defers the number to that page.

What's needed

1. Add the cumulative A(H5N1) page as a source (new sources.yaml id under
   respiratory.h5n1 / animal_spillover.h5n1), and, if its table doesn't extract
   cleanly, a small custom_scrapers/<id>.py that renders the cumulative count +
   per-subtype rows to prose (same pattern as who_cholera / cdc_measles in BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53).
2. Keep who_h5_hai for recent-event context.

Caveats

• The count is table-bound; may need the Docling refiner (now wired for
  scraper-resolved PDFs by BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53) or targeted parsing.
• Window timing: q6/q8 resolve over Jul–Dec 2026, which had barely opened at run
  time (2026-07-05), so there is little in-window data yet regardless of source —
  a base-rate/trend framing matters, and near-term the "current value" is ~0 new
  in-window cases. The cumulative page is still the right anchor for the trend.

References

• Scraper added in BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53: bioscancast/stages/extraction/custom_scrapers/who_h5_hai.py
• Diagnosis + write-up: data/investigations/bfg-evidence-quality-2026-07-05.md