Phase B deepening (2026-05-14): Hunt-Hess phenotypes (HH I–V) and 6 aneurysm-location phenotypes (AComm, PComm, MCA bifurcation, basilar tip, vertebral/PICA, perimesencephalic angio-negative) encoded as severity_triggers — each carries grade definition + immediate management branch + prognosis anchor. 5 setting playbooks span the full SAH journey: home (prehospital triage — STAT CT centre, no NSAIDs/nitroglycerin in field, comprehensive aneurysm centre bypass for known prior SAH) → ed (door-to-CT + Hunt-Hess/WFNS/mFisher scoring + transfer) → icu (secure <24 h + DCI surveillance days 3–14 + nimodipine + induced HTN per HIMALAIA) → inpatient (rehab + cognitive eval + family screen) → outpatient (cerebrovascular clinic 6-wk/3/6/12-mo; repeat MRA 6 mo; smoking cessation; BP <130/80; statin per ASCVD). depth-pass-2 (2026-05-18, shard-3 CL-3): EVIDENCE-PMID CORRECTION — the Phase B 2026-05-14 anchor set contained FOUR wrong-article mis-attributions (33357465 was a fish-oil paper, NOT ULTRA; 2868172 was Bland-Altman statistics, NOT BRANT; 24290000 was an anti-Müllerian-hormone paper, NOT HIMALAIA; 24852458 was a spreading-depolarization paper, NOT STASH). All corrected this pass to the PubMed-MCP-VERIFIED set: ULTRA Post Lancet 2021 (33357465; mRS 0-3 60% vs 64%, adj OR 0.86), BRANT Pickard BMJ 1989 (2496789; infarction 22% vs 33% RRR 34%, poor outcome 20% vs 33% RRR 40%), HIMALAIA Gathier Stroke 2017 (29158449; RR poor outcome 1.0, SAE RR 2.1; design paper 23692645), STASH Kirkpatrick Lancet Neurol 2014 (24837690; common OR 0.97 [0.75-1.25] p=0.803). Added Perry CT-6h (21768192; CT sens 100% <6h, 92.9% overall), Perry Ottawa SAH derivation (24065011) + validation (29133539; sens 100%), Frontera mFisher (16823296; mFS 4 vs 0-1 OR 2.2), Ducros RCVS (18025032). All 12 PMIDs PubMed-MCP-verified title/journal/year/pages 2026-05-18; the shard NEEDS_SOURCE_REVIEW flag is retained as a process marker. Full audit + per-PMID one-liners in src/lib/dossiers/neuro.sah.core.v1._research-bundle.md §9. depth-pass-2 §5.5.2: 3 ros-and-ddx seeds created (prisma/seed/ros-and-ddx/neuro.sah.core.v1.{ros,differentials,finding-lrs}.ts; auto-registered by seed-ros-and-ddx readdir) — 12 ROS items, 11 differentials (aSAH lead + perimesencephalic / np-angio-negative / RCVS / cervical-vertebral dissection / CVST / pituitary apoplexy / primary thunderclap / GCA / migraine / traumatic SAH; pre-test priors on the symptomatic thunderclap-cohort scale w/ Perry-cohort PMIDs), 30 finding×diagnosis LR rows. ACUTE convention: aSAH cardinal pivots LR+ ≥8 (thunderclap onset 9.0, CT+ blood 60, LP xanthochromia post-window 40) with distinct strongly-discriminating rule-out LR- (Ottawa SAH negative 0.02, CT-negative <6h 0.01). 2 conditional-dependency notes (#A CT-sensitivity | hours-since-ictus — 3 mutually-exclusive time strata; #B LP-xanthochromia | hours-since-ictus). Named pivots encoded: SAH-vs-migraine (onset-to-peak <1min + CT), SAH-vs-RCVS (string-of-beads + recurrence), SAH-vs-dissection (vessel-wall MRI). Cross-dossier routing edges (all verified on disk): neuro.sah-grade1-3.v1, neuro.sah-grade4-5.v1, neuro.sah-perimesencephalic.v1, neuro.ich.core.v1. NEEDS_SOURCE_REVIEW (no PMID fabricated): dissection (Debette/Leys) + GCA (Smetana JAMA 2002) per-finding operating characteristics — anchored to verified AHA/ASA 2023 (37212182), magnitudes flagged. depth-pass-2 §5.5.1: quantitative effect sizes wired into regimen/contraindication rationales with units + verified PMID — nimodipine RRR 34% infarction / 40% poor outcome (BRANT 2496789); coil-vs-clip ARR 7.4% (ISAT 16139655); ULTRA TXA null adj OR 0.86 (33357465); euvolaemic-HTN vs none RR 1.0 / SAE RR 2.1 (HIMALAIA 29158449); STASH common OR 0.97 (24837690); CT sens 100% <6h / 92.9% overall (Perry 21768192); Ottawa SAH sens 100% (24065011/29133539); mFisher OR 2.2 (Frontera 16823296). 4 special-population severity_trigger rows added as data: pregnancy aSAH, elderly aSAH, ADPKD/familial-aneurysm screen, neurogenic stunned myocardium (induced-HTN interaction). 3 RESOLVING cross-dossier routes added to sibling_differentiation (neuro.sah-grade1-3.v1 / neuro.sah-grade4-5.v1 / neuro.sah-perimesencephalic.v1). No drug/RxCUI added or changed; no new calc.* ids. Trial paradigm (PubMed-MCP-verified): 2023 AHA/ASA + NCS abandon triple-H in favour of euvolemic induced HTN (HIMALAIA negative); ULTRA shows no functional benefit at long-course TXA; STASH shows simvastatin does not prevent DCI; BRANT remains the nimodipine Class I anchor; ISAT supports coiling-preferred (ARR 7.4% death/dependence, esp. posterior circulation). Calculators added: calc.hunt_hess (I–V → mortality 0/2/5/15/55%), calc.wfns (I–V → mortality 5/10/20/35/55%), calc.modified_fisher (0–IV → vasospasm risk 0/12/21/35/40%) — all three in clinical-tools-registry.ts with stroke/neurology specialty. Hunt-Hess + WFNS + mFisher all band-mapped in guideline_basis. Schema-blocked calculators (surfaced as tickets in docs/framework-audit/shard-3-neuro-sym-state.md; shard-3 file scope forbids registry edits): calc.ottawa_sah (Ottawa SAH clinical decision rule for ED rule-out of thunderclap headache), calc.sahit (SAHIT prognostic model — 12-mo unfavourable outcome prediction). Both surfaced as plain-English required_assessments + severity_trigger rationales until registry entries land. Nimodipine PO/NG only — NEVER IV (FDA boxed warning; fatal hypotension). Routine seizure prophylaxis NOT recommended (treat documented seizure only); long-course antifibrinolytic NOT recommended (ULTRA); statin NOT recommended for DCI prevention (STASH). No `protocol.sah` exists in the registry yet — only protocol.stroke / protocol.ich are registered. Schema-blocked queue item.