evidence: source inventory for pm25-monitoring-south-asia + problem pack index

[cschanhniem]

Task

• Task ID: source-inventory
• Agent role: literature-scout
• Pack: air-quality/pm25-monitoring-south-asia
• Safety risk: medium
• Reviewer needed: domain-reviewer

Claim

Ground-level PM2.5 monitoring in South Asia is dangerously sparse: India has approximately 300 continuous monitoring stations for 1.4 billion people (roughly 1 per 4.7 million), with Pakistan, Bangladesh, and Nepal having even fewer stations accessible through public platforms. Where monitors are absent, satellite-derived estimates and column observations provide the only exposure data, but these carry large uncertainty in South Asia due to complex aerosol composition and monsoon cloud cover.

Kill condition

A reviewer identifies a public dataset showing that South Asian ground monitoring density exceeds 1 station per 1 million population, or finds a peer-reviewed validation study demonstrating that satellite-derived PM2.5 estimates in South Asian rural areas have RMSE below 10 ug/m3 against co-located reference monitors.

Files changed

Two contributions in one PR:

1. PM2.5 source inventory

• evidence.json — added 3 new evidence records (OpenAQ, WorldPop, Copernicus Sentinel-5P TROPOMI)
• datasets.md — updated with classification column and explicit reasons (7 sources: 5 usable, 1 limited, 1 rejected)
• evidence.md — added notes for all 6 evidence records
• claims.json — new file with dry-lab-verified claim

2. Problem pack index generator

• scripts/generate-index.mjs — new script that auto-generates a pack index
• docs/wiki/Index.md — generated index with summary stats, full pack listing, and priority guide
• package.json — added generate-index.mjs to build pipeline
• scripts/validate-repo.mjs — added index freshness check

The index shows every pack with status, domain, region, safety level, progress (evidence count, task count, claims), and priority label (needs evidence / ready / has claims / accepted work / needs triage). A "Needs Evidence" section highlights the 53 packs with fewer than 3 evidence records that have scoped tasks ready.

Evidence added

ID	Source	Type	Status
who-ambient-air-pollution-2024	WHO Ambient Air Pollution Fact Sheet	primary-source	Usable (existing)
health-effects-institute-state-of-global-air-2024	HEI State of Global Air 2024	primary-source	Usable (existing)
cpcb-monitoring-network-2023	CPCB National Air Quality Monitoring	primary-source	Usable (existing, allowlisted)
openaq-platform	OpenAQ Platform	dataset	Usable
worldpop-population-data	WorldPop Gridded Population Data	dataset	Usable
copernicus-sentinel-5p-tropomi	Copernicus Sentinel-5P TROPOMI	dataset	Limited
Low-cost sensor networks	(not added as evidence record)	—	Rejected

Validation method

pnpm validate passes. pnpm reproducibility:check passes (516 tasks). pnpm build passes (100 packs + index).

Known limitations

• The 300-station figure for India is from CPCB as of 2023 and may be outdated.
• Station counts for Pakistan, Bangladesh, and Nepal are inferred from OpenAQ coverage, not from official national agency data.
• TROPOMI column observations are not surface PM2.5; conversion requires modeled assumptions with uncertain accuracy in South Asia.
• Low-cost sensor rejection is based on absence of validation data, not on evidence that sensors are inaccurate.

Protocol Notes

• The why_it_matters field duplicating summary is now confirmed across 3 packs (oral-health, NCD, PM2.5). This is a systematic generation issue.
• The index generator does not currently track whether claims.json validates against claim.schema.json. A future validator rule should add pack-level claim validation.