Phase E deep ARDS (2026-05-16): all recommendations re-anchored to trial-verified PMIDs with effect sizes — ARMA 31% vs 39.8% NNT~11 (10793162), PROSEVA HR 0.39 (23688302), DEXA-ARDS 21% vs 36% (32043986), RECOVERY IMV RR 0.64 (32678530), ACURASYS HR 0.68 (20843245), ROSE null (31112383), EOLIA 35% vs 46% (29791822) + Goligher Bayesian posterior 88-99% (30347031), ART harm HR 1.20 (28973363), Amato ΔP RR 1.41/SD (25693014), LUNG SAFE severity mortality (26903337). Primary guideline upgraded to 2024 Global Definition (37487152) + 2024 ATS Management Update (38032683); legacy Berlin retained as severity-band fallback; non-intubated HFNO/NIV ARDS + S/F ≤315 + lung-US encoded in flow. Differential wired as data: cardio.acute-hf.core.v1 (cardiogenic edema — NT-proBNP/echo/PCWP/diuresis pivots), id.sepsis.core.v1 (sepsis+ARDS+AKI multi-organ), pulm.aspiration-pneumonia.core.v1 (aspiration/pancreatitis → ARDS); workups[].branches_to routes to id.sepsis.core.v1 / pulm.cap.core.v1 / pulm.aspiration-pneumonia.core.v1 / cardio.acute-hf.core.v1 (5 cross-dossier engine_id references). Special populations encoded in contraindication_rules + playbooks: pregnancy (pH ≥7.30 limit, left-lateral tilt, specialised proning), obesity (PBW not actual-weight Vt, esophageal-pressure-guided PEEP), COVID-ARDS (RECOVERY O2-gated steroid), immunocompromised (early BAL for DAH/PJP/drug-induced). Pediatric = PALICC-2 pointer only (not authored, per scope). New severity_trigger driving_pressure_over_15 (Amato) added; steroid trigger O2-gated per RECOVERY; dexamethasone renal/GI/glucose gating in contraindication_rules. Schema-gap notes: (a) ARDS-specific calculators (calc.pf_ratio, calc.sf_ratio, calc.driving_pressure, calc.berlin_severity, calc.murray_lung_injury) NOT in clinical-tools-registry — encoded as plain-English assessments/monitoring; cross-shard registry ticket. (b) No calc.berlin_global_definition tool — definition logic lives in flow phase text. (c) SettingTransition objects not authored (forward-looking schema field). (d) No engine-specific test file. (e) RxCUI re-validation queued (npm run research:rxnav:validate) — no rxcui hand-authored this pass. DEPTH-PASS-2 2026-05-18 (shard-07-cardio-chronic terminal CL-4, golden-template mirror cardio.htn.core.v1; acute-critical floors §5.5.2.h): (1) co-located src/lib/dossiers/pulm.ards.core.v1._design-brief.md + ._research-bundle.md authored (18 verified PMIDs — acute floor ≥8 exceeded — each with effect size + 95% CI, retrieval 2026-05-18, PubMed MCP live; Consensus UNAVAILABLE → WebSearch substitute logged; dose/intervention-effect anchors with mortality HR/ARR + time; RxCUI validation log; pre-test priors with cohort sources; T_test≈5%/T_treat≈25% + escalation-ladder T_treat for prone/NMB/ECMO; cross-dossier routing to id.sepsis.core.v1 / pulm.cap.core.v1 / pulm.aspiration-pneumonia.core.v1 / cardio.acute-hf.core.v1); design_brief: repointed to the co-located dossier-layer path. DEPTH-PASS-2 Bayesian seed rebuilt to the htn golden-template Prisma shapes exactly (ENGINE_ID pulm.ards.core.v1): prisma/seed/ros-and-ddx/pulm.ards.core.v1.{differentials,ros,finding-lrs}.ts — 12 differentials with cohort-anchored priors incl. MECE ARDS severity strata (mild P/F 200–300 / moderate 100–200 / severe <100 / non-intubated) + the acute-hypoxemic-bilateral-infiltrates partition (cardiogenic edema / DAH / AEP / AIP / severe-CAP-sepsis / TRALI / massive-PE / acute-HP); 14 ROS items; 27 LR rows (14 LR+ AND 27 LR− — acute floor ≥8 each exceeded) incl. bilateral-opacities-not-fully-explained, nested P/F bands, lung-US B-line pattern, echo/PAWP/NT-proBNP cardiogenic pivot + no-LA-hypertension; 3 conditional-dependency rules (P/F precondition-gated on PEEP≥5/HFNO≥30; nested P/F bands MECE; cardiogenic-pivot mutually-informative — do not multiply). STALE-PMID FIX: the 2024 Global Definition was wrongly cited as PMID 38241520 (an unrelated MXene/microfluidics paper) across all three seed files — corrected to the verified 37487152 (Matthay AJRCCM 2024;209:37-47, DOI 10.1164/rccm.202303-0558WS). DEPTH-PASS-2 2nd phenotypic axis intent: regimen axis 1 ards_lung_protective_ladder is the severity-staged ventilation/escalation ladder (mild HFNO/NIV → invasive low-Vt+ΔP+conservative-fluids → severe P/F<150 prone≥16h+selective-NMB → refractory VV-ECMO) encoding intervention × severity gating as DATA via step applies_when/step_up_when; the phenotype matrix (focal-vs-diffuse · hyper-vs-hypo-inflammatory Calfee · COVID-vs-non-COVID O2-gated steroid · cause-directed sepsis/pneumonia/aspiration/pancreatitis/TRALI · RV-failure-protective) is encoded as data across contraindication_rules + severity_triggers + sibling_differentiation + setting playbooks (ARDS has no classic drug ladder; the phenotype gating lives in those data structures rather than a parallel near-duplicate regimen_axes entry, to avoid a misleading second pharmacologic ladder for a non-pharmacologic-primary disease). DEPTH-PASS-2 content refresh re-verified live (PubMed MCP 2026-05-18) to the 2023 ESICM CPG (37326646) + 2024 Global Definition (37487152) + 2024 ATS Update (38032683): Global definition incl. HFNO ≥30 L/min + SpO2/FiO2 ≤315 + lung-US + resource-variable pathway; low-Vt 6 mL/kg PBW + ΔP ≤15 + Pplat ≤30 (ARMA 10793162 31% vs 39.8%, Amato 25693014 RR 1.41/SD); early prone P/F<150 (PROSEVA 23688302 HR 0.39, 0.25–0.63); selective NMB tempered ROSE-vs-ACURASYS (31112383 null / 20843245 HR 0.68); conservative fluids shock-gated (FACTT 16714767); ECMO refractory EOLIA + Bayesian (29791822 RR 0.76 / 30347031 posterior 88–99%); recruitment-maneuver HARM (ART 28973363 HR 1.20) + 2024 ATS strong-rec-against; HFNO/helmet-NIV (FLORALI 25981908 HR 2.01 / Patel 27179847 intubation 18.2% vs 61.5%); subphenotypes (Calfee 24853585). Existing evidence.pmids[] confirmed correct (no stale/wrong PMID in the dossier .ts — only the seed files needed the 38241520→37487152 fix). status: PRODUCTION kept. DEPTH-PASS-3 2026-05-26 (lane-E): +NMA +USPSTF +Cochrane +ICER stubs +decision thresholds, side-car at pulm.ards.core.v1._depth-pass-3.md.