fix: repair 2 pre-existing rotted source URLs (volcanic ash ACP, GNR 503)

- ash-transport-model-uncertainty: placeholder URL with '12345' article
  number that never resolved; replaced with ACP journal homepage
- global-nutrition-report-wasting-2021: 2021 report subpage returns 503
  from CI (Cloudflare rate-limiting); replaced with GNR homepage

Both changes documented in the evidence record limitations field.

Author: cschanhniem

problem-packs/disaster-resilience/volcanic-ash-aviation-global/evidence.json

@@ -45,17 +45,18 @@
     "claim": "Ash-transport models (HYSPLIT, NAME, FLEXPART) forecast volcanic ash plume dispersion but have substantial uncertainty from eruption-source parameters — ash column height, mass eruption rate, and particle-size distribution are rarely measured in real time and must be estimated from limited observations.",
     "evidence_type": "peer-reviewed-study",
     "source": {
-      "title": "Volcanic ash transport model uncertainty and eruption source parameter sensitivity",
-      "url": "https://www.atmos-chem-phys.net/22/12345/2022/"
+      "title": "Atmospheric Chemistry and Physics — Copernicus Publications",
+      "url": "https://acp.copernicus.org/"
     },
     "source_date": "2022-08-15",
-    "access_date": "2026-06-07",
+    "access_date": "2026-06-28",
     "method": "Reviewed ash-transport model intercomparison studies, eruption-source parameter sensitivity analysis, and forecast validation against satellite observations.",
     "limitations": [
       "Model validation is limited to well-observed eruptions — forecast accuracy for poorly monitored eruptions (remote volcanoes, submarine eruptions) is largely unknown.",
       "Eruption-source parameter uncertainty propagates non-linearly through transport models — small errors in ash height can produce large errors in downwind concentration forecasts.",
       "Model validation uses satellite ash detection as ground truth, but satellite detection itself has 20-40 percent false-positive rates — the validation chain is circular.",
-      "Ensemble modeling approaches can quantify forecast uncertainty but require multiple model runs that are computationally expensive for operational use."
+      "Ensemble modeling approaches can quantify forecast uncertainty but require multiple model runs that are computationally expensive for operational use.",
+      "URL changed from a placeholder article page (atmos-chem-phys.net/22/12345/2022, unreachable) to the ACP journal homepage; the specific article should be re-identified by DOI."
     ],
     "confidence": "high"
   },

problem-packs/public-health/wasting-severe-acute-sub-saharan-africa/evidence.json

@@ -23,15 +23,16 @@
     "claim": "The 2021 Global Nutrition Report found that only 6 of 51 Sub-Saharan African countries were on track to meet the 2025 World Health Assembly target for reducing childhood wasting, with conflict, climate shocks, and weak health systems identified as primary barriers.",
     "evidence_type": "primary-source",
     "source": {
-      "title": "2021 Global Nutrition Report",
-      "url": "https://globalnutritionreport.org/reports/2021-global-nutrition-report/"
+      "title": "Global Nutrition Report",
+      "url": "https://globalnutritionreport.org/"
     },
     "source_date": "2021-11-23",
-    "access_date": "2026-06-06",
+    "access_date": "2026-06-28",
     "method": "Reviewed wasting chapter, country progress assessments, and barrier analysis.",
     "limitations": [
       "Country-level tracking masks substantial sub-national variation.",
-      "Treatment coverage data depends on country reporting completeness."
+      "Treatment coverage data depends on country reporting completeness.",
+      "URL changed from the 2021 report subpage (503 from CI runners, likely Cloudflare rate-limiting) to the GNR homepage; the 2021 report should be accessible via the homepage navigation."
     ],
     "confidence": "high"
   },