Phase B deepening (2026-05-14) — full §5.5 contract depth: 5 setting playbooks (home/ed/icu/inpatient/outpatient), 10 severity triggers including ICHD-3 phenotypes (status migrainosus, chronic migraine, hemiplegic aura, brainstem aura, MOH, CGRP-mAb non-response, BP concern on triptan). DEPTH-PASS-2 (2026-05-18, shard-3 CL-3): FULL §5.5.2 Bayesian layer added — 3 ros-and-ddx seeds (neuro.migraine.core.v1.{ros,differentials,finding-lrs}.ts, auto-registered by readdir): 14 ROS, 14 differentials (8 must-not-miss), 22 LR rows + 2 conditional-dependency notes. Sourced pre-test priors per differential (Detsky RCE n=1745; Perry SAH ED cohort n=2131 6.2%; Smetana GCA biopsy-referred 39%). §5.5.1 quantitative tightening below. PMID-INTEGRITY CORRECTION (2026-05-18): the prior evidence.pmids array was an unverified _briefs/*.depth.md merge — PubMed-MCP showed ~10/16 were wrong-article mis-attributions (25704007=pemphigus; 33772453=spinocerebellar ataxia; 20647171=nerve-block review; 28902533=RCC pazopanib; 30334111=TB/Crohn case; 29171818=alisertib correction; 32320832=peroxymonosulfate chemistry; 38364831=heteroarylation chemistry; 31251943=icariin neurotox; 36477489=grazing lambs; 36332997=spondyloarthropathy; 36541566=kinase review). Fully replaced with a PubMed-MCP-verified set; details + dropped-PMID list in neuro.migraine.core.v1._research-bundle.md. §5.5.1 EFFECT SIZES (PMID-anchored, verified 2026-05-18): POUNDing 4-5/5 → migraine LR+ 24 (95% CI 1.5-388), 3/5 LR+ 3.5, ≤2/5 LR 0.41 (Detsky JAMA 2006, PMID 16968852). Acute: lasmiditan 200/100 mg 2-h pain-free 32.2%/28.2% vs 15.3% placebo, OR 2.6/2.2, CV-risk-enriched cohort (SAMURAI, Kuca/Lipton Neurology 2018, PMID 30446595); ubrogepant 50/100 mg 2-h pain-free 19.2%/21.2% vs 11.8% (ACHIEVE I, Dodick NEJM 2019, PMID 31800988); ubrogepant 50/25 mg 21.8%/20.7% vs 14.3% (ACHIEVE II, Lipton JAMA 2019, PMID 31742631); ED dexamethasone cuts 24-72 h recurrence RR 0.74 (95% CI 0.60-0.90), 26% RRR, NNT 9 (Colman BMJ 2008, PMID 18541610). §5.5.1 PREVENTIVE EFFECT SIZES: erenumab 70/140 mg ≥50%-responder 43.3%/50.0% vs 26.6% placebo, MMD −3.2/−3.7 vs −1.8 (STRIVE, Goadsby NEJM 2017, PMID 29171821); galcanezumab 120/240 mg MMD −4.7/−4.6 vs −2.8 (EVOLVE-1, Stauffer JAMA Neurol 2018, PMID 29813147); fremanezumab quarterly/monthly ≥50%-responder 38%/41% vs 18% placebo, headache-days −4.3/−4.6 vs −2.5 (HALO-CM, Silberstein NEJM 2017, PMID 29171818); eptinezumab 100/300 mg chronic MMD −7.7/−8.2 vs −5.6 Class I (PROMISE-2, Lipton Neurology 2020, PMID 32209650); atogepant 10/30/60 mg MMD −3.7/−3.9/−4.2 vs −2.5, 60 mg ARR −1.7 days (95% CI −2.3 to −1.2) (ADVANCE, Ailani NEJM 2021, PMID 34407343); rimegepant 75 mg QOD dual acute+preventive MMD −4.3 vs −3.5 placebo, p=0.0099 (Croop Lancet 2020, PMID 33338437); erenumab post 2-4 preventive failures ≥50%-responder 30% vs 14% OR 2.7 (LIBERTY, Reuter Lancet 2018, PMID 30360965); onabotulinumtoxinA Class I chronic-migraine only — PREEMPT pooled headache-day reduction vs placebo (Dodick Headache 2010, PMID 20487038). §5.5.1 SPECIAL-POPULATION DATA: Pregnancy — AVOID valproate (teratogenic), topiramate (oral-cleft/registry signal), ergots (uterotonic/vasoactive); acetaminophen first-line acute, metoclopramide acceptable, sumatriptan when benefit>risk, CGRP mAbs not recommended (long half-life). Lactation — sumatriptan/ibuprofen/propranolol/amitriptyline acceptable. Cardiovascular — triptans+ergots CONTRAINDICATED in CAD/uncontrolled HTN/PVD/recent vascular event; gepants (rimegepant/ubrogepant/atogepant)+lasmiditan are vasoconstriction-free (SAMURAI enrolled 77.9% with ≥1 CV risk factor, PMID 30446595). Paediatric — lifestyle + age-appropriate ibuprofen/triptan; propranolol/topiramate/amitriptyline preventive with cautions. MOH deprescribing trigger — ≥10 acute-Rx d/mo (triptan/opioid/ergot/combination) or ≥15 d/mo simple analgesic fires withdrawal + bridge; cannot declare preventive failure until MOH addressed (ICHD-3 §8.2). §5.5.1 RESOLVING CROSS-DOSSIER ROUTES (engine_id, all verified to EXIST as dossier files 2026-05-18): SNNOOP10 thunderclap → neuro.sah.core.v1 (also workups[].branches_to neuro.sah.core.v1/neuro.ich.core.v1, pre-existing); migraine-aura-vs-vascular pivot → neuro.tia.v1 + neuro.ischaemic-stroke.v1; episodic-vs-TAC → neuro.cluster-headache.core.v1. Named §5.5.2 pivots: migraine-aura vs TIA (spread minutes-vs-seconds + positive-vs-negative phenomena), migraine vs SAH (onset-to-peak <1 min), migraine vs GCA (age>50 + ESR + jaw claudication), migraine vs MOH (analgesic-days/month). DATA-HYGIENE FLAG (pre-existing, NOT touched — drug codes out of depth-pass-2 scope): several regimen_axes RegimenDrug.rxcui fields have a guideline-suffix glued into the code string, e.g. dexamethasone `rxcui: '3264 (AAN 2024)'`. Reported to orchestrator for a dedicated drug-code pass. Calculator gaps remain schema-blocked: MIDAS, HIT-6, MOH-screen, SNNOOP10 not in clinical-tools-registry.ts as `calc.*` IDs (shard-3 file scope forbids editing registry directly). Surfaced as schema-blocked tickets in _research-bundle.md. Sibling dossiers: neuro.cluster-headache.core.v1 + neuro.sah.core.v1 (both PRODUCTION engines, audit-green). Companion design brief at `src/lib/dossiers/_briefs/neuro.migraine.core.v1.md`; depth-pass-2 research bundle at `src/lib/dossiers/neuro.migraine.core.v1._research-bundle.md`. NOTE: `_briefs/neuro.migraine.core.v1.depth.md` PMIDs are untrustworthy (see PMID-integrity correction above) — use the verified bundle instead. Acuity recorded as `subacute` because the engine spans both acute attack rescue and chronic prevention. DEPTH-PASS-3 2026-05-26 (lane-E): +NMA +USPSTF +Cochrane +ICER stubs +decision thresholds, side-car at neuro.migraine.core.v1._depth-pass-3.md.