shard-3 neuro-sym autonomous authoring (2026-05-16). Authored at INTEGRATED — no panel/router entry (matches in-shard neuro.bell-palsy.v1 precedent: new neuro INTEGRATED dossiers ship without a router entry). Three load-bearing pivots: (1) is-it-a-seizure (history is the test; Sheldon point score PMID 12103268 — 94% sens/94% spec; lateral tongue-bite + prolonged post-ictal confusion favour epileptic; prodrome/upright/rapid-recovery favour convulsive syncope; eyes-closed/long/fluctuating favour PNES); (2) provoked/acute-symptomatic vs unprovoked (within 7 d of stroke/TBI/CNS infection or active metabolic/toxic/withdrawal/eclampsia → treat cause, generally NO chronic ASM — ILAE 2014 Fisher PMID 24730690); (3) recurrence-risk → ASM shared decision (AAN/AES 2015 Krumholz PMID 25901057: 21–45% at 2 yr, higher with remote insult/epileptiform EEG/significant MRI/nocturnal; immediate ASM cuts 2-yr recurrence but not long-term remission/QoL — FIRST PMID 9339678, MESS Marson PMID 15950714; epilepsy diagnosable after ONE seizure if ≥60% 2-yr recurrence — ILAE 2014). regimen_axis first_seizure_aed: Step 1 whether-to-treat shared decision; Step 2 focal-onset (lamotrigine/levetiracetam first-line; lacosamide/carbamazepine/oxcarbazepine/brivaracetam/zonisamide/phenytoin); Step 3 genetic generalised (levetiracetam/lamotrigine first-line; valproate only if childbearing potential excluded; AVOID Na-channel blockers; zonisamide/perampanel/clobazam add-on); Step 4 acute-symptomatic/provoked cause-directed + PNES anti-pattern guard. All 12 ASM RxCUIs RxNav-validated (forward + reverse) 2026-05-16. Contraindication rules encode: valproate hard-contraindication in childbearing potential; Na-channel-blocker-in-GGE avoidance; HLA-B*15:02 carbamazepine/oxcarbazepine; lamotrigine slow titration for rash; do-NOT-start-chronic-ASM-for-acute-symptomatic/single-provoked; do-NOT-treat-PNES-with-ASM; alcohol-withdrawal = benzodiazepine not chronic ASM; renal dose adjustment via CKD-EPI 2021; levetiracetam/perampanel behavioural signal. 3 setting playbooks (ED first-seizure triage / outpatient first-seizure clinic / ED→clinic transition); 7 severity_triggers (SE-in-evolution → route neuro.status-epilepticus.core.v1; focal deficit/persistent AMS → urgent imaging; reversible metabolic/eclampsia; high-recurrence unprovoked → diagnose epilepsy + ASM; suspected PNES → video-EEG/psychiatry, no ASM; alcohol-withdrawal seizure; childbearing-potential ASM decision); 5 sibling-differentiation rows (neuro.status-epilepticus.core.v1, symptom.syncope.ed.v1, symptom.altered_mental_status.ed.v1, neuro.ischaemic-stroke.v1, neuro.delirium.v1). Schema-blocked downstream (NOT invented as calc ids): no validated first-seizure recurrence-risk calculator in clinical-tools-registry — MESS prognostic model (Kim Lancet Neurol 2006 PMID 16545748; low/medium/high recurrence) and an ILAE-2014 ≥60% recurrence-threshold helper surfaced as schema-blocked tickets in the research bundle; recurrence factors encoded narratively in RISK_STRATIFICATION + severity_triggers. Registry ids used (all confirmed-resolving): workups workup.first_seizure (required), workup.status_epilepticus, workup.acute_stroke, workup.bacterial_meningitis, workup.encephalopathy, workup.toxic_alcohols, workup.hypoglycemia, workup.hyponatremia; panels panel.glucose_a1c, panel.metabolic, panel.cmp, panel.renal, panel.lft, panel.tox_screen, panel.cbc, panel.csf; calculators calc.phq9, calc.gad7, calc.ckd_epi_2021; cascades cascade.labs_command, cascade.electrolyte. No protocol.status_epilepticus (does not resolve) — workup.status_epilepticus used instead. DEPTH-PASS-2 (2026-05-18, shard-3 neuro-sym CL-3; last_reconciled bumped 2026-05-16 → 2026-05-18). §5.5.2 Bayesian differential layer added via 3 new auto-registered ros-ddx seed files (prisma/seed/ros-and-ddx/neuro.first-seizure-eval.v1.{ros,differentials,finding-lrs}.ts): 12 ROS items, 8 differentials with sourced clinic-population pre-test priors, 22 finding×diagnosis LR rows. ddx/finding namespace is first-seizure-evaluation-specific (ddx.neuro.fse.*; *fse* finding_display; FSE-specific diagnosis SNOMEDs) so upserts do NOT clobber the shared neuro.status-epilepticus.core.v1 ros-ddx rows. §5.5.1 quantitative tightening — effect sizes with units + inline trial/year/PMID (numeric provenance for the previously-directional recurrence drivers): (1) 2-yr recurrence after first unprovoked seizure 21–45% (AAN/AES 2015 Krumholz, Neurology 2015, PMID 25901057); (2) epileptiform-EEG recurrence relative rate RR 2.16 (95% CI 1.07–4.38) at 1–5 yr (AAN/AES 2015 PMID 25901057), pooled Cochrane RR 1.90 (95% CI 1.60–2.25; Adan CD013848.pub2 2025); (3) significant-imaging-abnormality recurrence HR 2.44 (95% CI 1.09–5.44) (AAN/AES 2015 PMID 25901057), Cochrane RR 2.19 (95% CI 1.74–2.76); (4) nocturnal-seizure recurrence OR 2.1 (95% CI 1.0–4.3) at 1–4 yr (AAN/AES 2015 PMID 25901057), Cochrane HR 1.41 (95% CI 1.13–1.75); (5) routine-EEG epileptiform yield ~23% of adults, Level B predictor (AAN/AES 2007 Krumholz, Neurology 2007, PMID 18025394); (6) 10-yr prospective cohort recurrence by factor — epileptiform VEEG 71.9%, prior brain insult 75.0%, focal-onset 69.7%, abnormal MRI 61.5%, nocturnal 61.4% (Seizure/Eur J Epilepsy 2024, PMID 39388981); (7) Sheldon seizure-vs-syncope point score 94% sensitivity / 94% specificity, n=671 (seizure 102 / syncope 437) (Sheldon, JACC 2002, PMID 12103268); (8) MESS — immediate ASM reduces recurrence over 1–2 yr, no long-term-remission change, n=1847 (Marson, Lancet 2005, PMID 15950714); (9) FIRST — 2-yr remission 84%→79% at 3 yr, 85%→86% at 10 yr, p=NS (FIRST Seizure Trial Group, Neurology 1997, PMID 9339678); (10) ILAE-2014 epilepsy-after-one-seizure post-test threshold ≥60% 10-yr recurrence, ≥2 unprovoked >24 h apart ≈60–90% recurrence (Fisher, Epilepsia 2014, PMID 24730690). RR/OR→LR conversions for the recurrence drivers are shown with arithmetic inline in the finding-lrs file (base p0≈0.35). §5.5.1 special-population branches (data-encoded): (a) pregnancy/eclampsia — pregnancy ≥20 wk or postpartum ≤6 wk + seizure → eclampsia bucket (LR+ 8.0 in finding-lrs C4), MgSO4 + OB route, valproate hard-contraindicated; (b) pediatric febrile-vs-afebrile — explicitly OUT OF SCOPE for this adult engine (pediatric febrile seizures / neonatal seizures excluded in the header) but the afebrile-pediatric and febrile-vs-afebrile distinction is recorded narratively as a routing boundary (refer pediatric neurology — no in-shard pediatric first-seizure engine to route to); (c) geriatric late-onset structural — elderly have a lower recurrence threshold from occult structural disease (remote-symptomatic LR+ 5.6, finding-lrs B3); lower ASM start doses, fall/interaction caution, levetiracetam behavioural caution, lamotrigine well tolerated; (d) driving-restriction by jurisdiction encoded as a counselling deliverable (jurisdiction-dependent seizure-free interval + clinician-notification duties — counsel AND document; carried in FOLLOWUP + outpatient/transition playbooks rather than as a numeric calc). §5.5.2 resolving cross-dossier engine_id routes (each appears as a linked_disease_engine on a differential and/or a ROUTE-* edge in finding-lrs): symptom.syncope.ed.v1 (convulsive/cardiac syncope mimic — ddx.neuro.fse.convulsive_syncope, ROUTE-1); neuro.transient-ischaemic-attack.v1 (focal-deficit mimic — ddx.neuro.fse.tia, ROUTE-2); neuro.status-epilepticus.core.v1 (seizure ≥5 min / recurrent without recovery — absolute precedence, ROUTE-3); neuro.ischaemic-stroke.v1 (acute structural acute-symptomatic cause — finding-lrs C2 route). All four resolve to in-shard dossiers already named in the dossier body. No new calc.* ids introduced — MESS prognostic model and the ILAE-2014 ≥60% threshold remain schema-blocked and are encoded narratively + as Block-B recurrence LRs feeding the high_recurrence_unprovoked severity_trigger. PMIDs introduced this pass: 39388981 (10-yr prospective recurrence cohort — WebSearch-verified against pubmed.ncbi.nlm.nih.gov 2026-05-18, added to evidence.pmids). Cochrane CD013848.pub2 (Adan G et al. 2025; 23 studies, n=5918) cited by DOI 10.1002/14651858.CD013848.pub2 — numeric PMID NEEDS_SOURCE_REVIEW (not added to evidence.pmids until a numeric PMID is confirmed). All other PMIDs reused from the existing WebSearch-verified _research-bundle.md. No drug codes added or changed (depth-pass-2 is differential/LR + §5.5.1 only). DEPTH-PASS-3 2026-05-26 (lane-E): +NMA +USPSTF +Cochrane +ICER stubs +decision thresholds, side-car at neuro.first-seizure-eval.v1._depth-pass-3.md.