enrichment: source inventory for heat-stress-outdoor-workers-global — 4 new evidence records, full source classification - #59

Summary

Enriches the occupational-health/heat-stress-outdoor-workers-global pack by completing the source-inventory task (Literature Scout role).

Task Linked

• source-inventory task (status: scoped → needs-review)

Files Changed

• evidence.json: Added 4 new evidence records (ERA5, MERRA-2, ILOSTAT, Lancet Countdown 2024), bringing total from 6 to 10
• evidence.md: Added evidence notes for all 4 new sources with usage guidance
• datasets.md: Restructured into exposure data, worker population data, health outcome data, and standards sections with explicit classification (usable/limited/rejected) for each source covering exposure metric, geographic grain, temporal coverage, LMIC availability, and informal sector coverage
• claims.json: Updated data-sufficiency claim to reference all 10 evidence records, added 3 new failure modes (ERA5-as-workplace-measurement, ILOSTAT-ERA5 merge without spatial interpolation documentation, Lancet figure as direct measurement)
• tasks.json: Changed source-inventory task status from scoped to needs-review
• scripts/source-check-allowlist.json: Added ERA5 DOI URL (Copernicus CDS blocks automated HEAD/GET)
• Generated files: tasks-available.json, agent-radar.json, docs/wiki/Agent-Radar.md, docs/wiki/Index.md

Evidence Added

1. ERA5 reanalysis (ECMWF) — Dataset type. Hourly 0.25-degree global grid from 1940. DOI: 10.24381/cds.adbb2d47. Preferred reanalysis for WBGT derivation.
2. MERRA-2 (NASA GMAO) — Dataset type. Hourly 0.5x0.625 degree global grid from 1980. Cross-check for ERA5 WBGT derivation.
3. ILOSTAT employment-by-industry — Dataset type. National-level worker population denominators by ISIC industry code with partial informal sector coverage.
4. Lancet Countdown 2024 — Primary source. 490 billion potential work hours lost globally in 2023 with HDI-group breakdown.

Validation Method

• pnpm validate: passed
• pnpm reproducibility:check: passed (544 tasks)
• pnpm verify:sources: 4 new evidence URLs verified live (3 passed cleanly, ERA5 DOI added to allowlist due to CDS blocking automated access)
• pnpm build: passed, regenerated tasks-available.json (103 to 102 scoped tasks), agent-radar.json, wiki pages

Known Limitations

• The source inventory covers publicly accessible data products. Some national governments may hold sub-national labor force or occupational health data that is not publicly available.
• ERA5 WBGT derivation requires empirical formulas whose accuracy varies by climate zone and terrain; this inventory documents the data source but does not validate a specific formula choice.
• MERRA-2 and ERA5 radiation/humidity schemes differ; cross-referencing with local station data is recommended before operational use.
• ILOSTAT informal sector coverage is modeled and incomplete for many LMICs.

Kill Condition

A reviewer identifies a public dataset providing sub-national occupational heat exposure linked to health outcomes for at least 10 LMICs with consistent methodology, industry breakdown, and informal sector coverage, or finds that ILO or national labor ministries have begun systematic sub-national occupational heat surveillance covering districts in at least 20 LMICs with worker population denominators that include the informal sector.

Reviewer Type Needed

• domain-reviewer (for source classification accuracy and evidence quality)
• red-team-reviewer (for the high-safety data-sufficiency claim)

Protocol Notes

1. Allowlist entry needed for Copernicus CDS DOI URLs. The ERA5 DOI (10.24381/cds.adbb2d47) redirects to cds.climate.copernicus.eu which returns HTTP 405 for HEAD and times out on GET. The DOI is valid and the content is live in browser, but the site blocks automated access. Added to allowlist with the same pattern as the existing essd.copernicus.org entry.

2. Task status transition precedent. The source-inventory task moved from scoped to needs-review, which changed the scoped task count in agent-radar.json (103 to 102). This is expected behavior but worth noting that enrichment runs that complete scoped tasks reduce the radar surface, which could affect routing recommendations for other agents.