[new-pack] stillbirth-measurement-quality-global — separate care-quality failure from counting failure before ranking burden

[cschanhniem]

Mode

Discovery

Daily run checklist

• Read AGENTS.md, SKILL.md, SAFETY.md, all relevant schemas
• Chose mode: discovery or enrichment
• If discovery: ranked 3–5 candidates, picked the strongest, drafted full pack, included ranking table in PR
• If enrichment: picked a scoped task from agent-radar or tasks-available, produced expected artifact, satisfied done condition
• Every claim has a kill condition
• Every evidence record has dated source, stable URL, method, limitations
• pnpm validate passes
• pnpm reproducibility:check passes (if task maps changed)
• pnpm verify:sources passes (if evidence URLs changed)
• pnpm build passes, regenerated files committed
• PR opened with required body fields
• Protocol Notes included if any friction was observed

Mode chosen: discovery

Pack targeted: public-health/stillbirth-measurement-quality-global

One-line summary of what this run added to the world's verified knowledge: a new problem pack that turns stillbirth from a vague burden claim into a checkable decision problem about when observed stillbirth burden should trigger care-quality verification versus counting-system strengthening.

Ranking Table

Problem	Scale (source)	Tractability	Neglectedness	Evidence gap	Why not the others
Stillbirth measurement gaps and intrapartum care quality	~1.9 million stillbirths globally in 2019, highest rates in Sub-Saharan Africa and South Asia (Lancet 2021)	Stronger than it looks: a verified distinction between intrapartum-quality failure and counting failure changes whether ministries and donors fund audits, labor ward verification, or CRVS strengthening first (Lancet 2016)	Under-attended relative to adjacent maternal and neonatal agendas because stillbirth sits between systems and is often absent from routine accountability surfaces. Evidence strength here is moderate, but the coordination failure is clear.	Which places have enough evidence to justify quality-of-care verification rather than treating observed burden mainly as a counting artifact? Named decision-maker: maternal-newborn unit, district obstetric QI team, donor funding EmONC or perinatal death audit work.	Picked because the wedge is narrower and more decision-relevant than generic burden packs, and it complements rather than duplicates existing maternal and neonatal packs.
Postpartum hemorrhage blood availability and referral reliability	PPH remains the leading direct cause of maternal death globally; maternal deaths were ~287,000 in 2020 (WHO maternal mortality estimates, 2023)	Potentially high, but tractability here quickly becomes facility-operational and country-specific. Harder to keep GitHub-native and evidence-first without drifting into protocol recommendation.	Important but less neglected inside maternal-health operations than stillbirth measurement; more institutions already frame blood and referral as known priorities.	Which districts should fund blood availability verification versus transport reliability first?	Rejected for now because the repo already has nearby maternal mortality and obstetric transport packs; this would risk overlap before adding a distinct verification wedge.
Lead exposure in pregnancy and early childhood in LMIC cities	Lead exposure contributes to major cardiovascular and neurodevelopmental burden; WHO states no safe blood lead level in children (WHO fact sheet)	Tractable for mapping, but the repo already has public-health/lead-exposure-urban-global, so discovery value is low.	Not a missing portfolio problem here.	Which neighborhood signals justify blood lead surveillance first?	Rejected because the portfolio already covers it.
Oral health access and workforce gaps	Oral diseases affect an estimated 3.5 billion people globally (WHO oral health fact sheet)	Real tractability on workforce and service access, but the repo already has public-health/oral-health-access-global.	Not a missing portfolio problem here.	Which countries have enough data for sub-national oral-health access work?	Rejected because it already exists in the portfolio.
Antenatal syphilis screening and stillbirth prevention linkage	Congenital syphilis and untreated maternal syphilis remain major causes of preventable stillbirth and neonatal death (WHO congenital syphilis framing, general reference)	Could be strong, but the stillbirth pack is a more general parent wedge. Syphilis is likely better handled as a follow-on scoped task or sub-pack once stillbirth measurement discipline exists.	Less neglected as a disease program than the cross-system stillbirth measurement failure itself.	Which countries should prioritize screening-quality verification versus broader stillbirth audit?	Rejected because it is downstream of the broader stillbirth measurement problem and narrower than the current discovery need.

Files changed

• Added problem-packs/public-health/stillbirth-measurement-quality-global/
• Updated root guides: AGENTS.md, SKILL.md
• Regenerated tasks-available.json, agent-radar.json, and wiki pages

Evidence added or changed

Added three DOI-backed evidence records:

• lancet-stillbirth-estimates-2021
• lancet-ending-preventable-stillbirths-2016
• lancet-stillbirth-rates-risk-factors-2016

Validation method

• pnpm build
• pnpm validate
• pnpm reproducibility:check
• Direct URL resolution check for the three newly added DOI URLs

Known limitations

• The stillbirth burden anchors are sufficient for discovery framing, but the pack does not yet contain a source inventory across survey, CRVS, and facility systems. That is intentionally the first scoped task.
• Neglectedness here is argued primarily from institutional fragmentation and under-ownership relative to adjacent maternal and neonatal agendas, not from a hard funding denominator assembled in this PR.
• Repo-wide pnpm verify:sources remains blocked by pre-existing unrelated 403/fetch failures in other packs, so this PR uses direct checks on the newly added URLs instead of claiming a fully clean repo-wide source pass.

Kill condition

The seeded claim fails if a reviewer shows a current public source set for at least one high-burden country where survey, CRVS, and facility stillbirth measures already align on threshold, denominator, time reference, and sub-national grain well enough that a separate reconciliation step adds no material decision value.

Reviewer type needed

• domain-reviewer
• Later follow-on work will also need replicator, red-team-reviewer, and field-reality-reviewer

Protocol Notes

• The repo-level validator enforces a fixed canonical ten-file pack set. Pack-local agent guides are allowed, but they cannot appear in canonical_files. That is stricter than the broad reading of the problem schema alone and worth keeping in mind for future discovery runs.
• Repo-wide source verification currently mixes genuine dead links, allowlisted CI-unreachable sources, and third-party 403s. That makes pnpm verify:sources noisy as a merge gate for otherwise clean discovery PRs that only add DOI-backed records.