evidence: source inventory for ncd-risk-factor-surveillance-global

[cschanhniem]

Task

• Task ID: source-inventory
• Agent role: literature-scout
• Pack: public-health/ncd-risk-factor-surveillance-global
• Safety risk: low
• Reviewer needed: domain-reviewer

Claim

Available NCD risk factor surveillance data sources are sufficient for country-level gap identification in low-income countries (which countries have STEPS surveys, when they were last conducted, and what modeled estimates exist), but no source provides sub-national NCD risk factor data for any low-income country. Sub-national risk factor gap mapping is not currently feasible with public data.

Kill condition

A reviewer identifies a public dataset that provides sub-national NCD risk factor data (measured or self-reported) for at least one low-income country, or finds that WHO has published a systematic assessment of STEPS repeat-survey frequency with a figure different from the 40 percent inference.

Files changed

• evidence.json — added 4 new evidence records (IHME GBD 2021, WHO GHO NCD indicators, WHO NCD surveillance data portal, DHIS2)
• datasets.md — replaced placeholder content with real source classification table (6 sources: 4 usable, 1 limited, 1 rejected)
• evidence.md — added notes for all 5 evidence records
• claims.json — new file with dry-lab-verified claim validated against claim.schema.json
• agent-radar.json, docs/wiki/Agent-Radar.md — regenerated by build

Evidence added

ID	Source	Type	Status
who-steps-surveillance-2023	WHO STEPS Surveillance Manual	primary-source	Usable (existing)
ihme-gbd-2021	IHME Global Burden of Disease 2021	dataset	Usable
who-gho-ncd-indicators	WHO GHO NCD indicators	dataset	Usable
who-ncd-surveillance-data-portal	WHO NCD Surveillance Data and Reporting	primary-source	Usable
dhis2-ncd-modules	DHIS2 Health Information System	primary-source	Limited
DHS/MICS NCD modules	(not added as evidence record — rejected)	—	Rejected

Validation method

pnpm validate passes. pnpm reproducibility:check passes (516 tasks). pnpm build passes (100 packs). All evidence URLs verified live (200 response).

Known limitations

• GBD estimates are modeled, not directly measured; they synthesize multiple data sources with varying quality.
• The claim about fewer than 40 percent repeat STEPS surveys is an inference from existing evidence and WHO portal content, not a single WHO-stated figure.
• DHIS2 NCD module implementation status is inferred from documentation, not from a systematic survey of all LIC deployments.
• The claim is about data availability, not about NCD outcomes or risk factor prevalence.

Protocol Notes

• The problem.json why_it_matters field is identical to the summary field — same copy-paste issue as oral-health-access-global. This appears to be a systematic pattern in generated packs.
• The tasks.json inputs field lists generic categories ("primary survey data", "satellite data") that do not match the NCD surveillance domain. Same pattern as oral-health.
• The validator still does not check whether claims.json exists in a problem pack directory or validate it against claim.schema.json at the pack level. Second PR to surface this; a future validator rule should add this check.