PJP = host-defined opportunistic Bayesian rule-in engine. §5.5.2 wired: pretest set by host class (HIV CD4 band vs non-HIV iatrogenic substrate) in FRAME/CONTEXT; sequential test LRs (serum β-D-glucan sens ~91%/spec ~79% PMID 32479781, BAL immunofluorescence specific vs quantitative PCR sensitive, LDH sensitive/non-specific, HRCT ground-glass with high NPV if normal); confirm → empiric TMP-SMX → adjunctive-steroid threshold (PaO2 <70 / A-a >35 / SpO2 <92%) → immune reconstitution. HIV-vs-non-HIV is the dominant axis (tempo, burden, β-D-glucan NPV, steroid-evidence strength, mortality). Two conditional dependencies explicitly modeled and flagged do-not-read-in-isolation: (1) β-D-glucan post-test probability | host pretest band — a negative test rules out only at low-intermediate prior (non-HIV ≤20%, HIV ≤50%) (severity_trigger beta_d_glucan_negative_conditional_on_host_band; INITIAL_WORKUP phase logic; Del Corpo CMI 2020 PMID 32479781, Li 2015 PMID 26793343); (2) BAL-PCR positivity | colonization vs invasive disease — a low-burden positive PCR may reflect colonization, interpreted jointly with IF stain + HRCT + host (severity_trigger bal_pcr_positive_low_burden_colonization_vs_disease). Effect sizes wired with PMIDs (≥5): Bozzette/California Collaborative (PMID 2233917) 31-d respiratory failure 0.14 vs 0.30 (P=0.004), death 0.11 vs 0.23 (P=0.009), 84-d death 0.16 vs 0.26 (P=0.026); Gagnon/Consensus (PMID 2233916) survival 75% vs 18% (P<0.008); Del Corpo β-D-glucan (PMID 32479781) sens 91%/spec 79%, HIV 94% vs non-HIV 86%; Li (PMID 26793343) sens 0.91/spec 0.75; Cilloniz (PMID 31550942) HIV ~15% vs non-HIV ~50% in-hospital mortality; Feng (PMID 35313548) non-HIV nomogram AUC 0.865 / HIV C-index 0.904; Zolopa ACTG A5164 (PMID 19440326) early-ART AIDS progression/death OR 0.51 (95% CI 0.27-0.94), HR 0.53; Safrin ACTG 108 (PMID 8610948) TMP-SMX ≈ dapsone-TMP ≈ clinda-primaquine mild-moderate. Cross-dossier routing via workups[].branches_to + sibling_differentiation[].sibling_engine_id using real engine_ids confirmed by grep of src/lib/dossiers/: pulm.cap.core.v1, pulm.tuberculosis.v1, id.sepsis.core.v1, pulm.ards.core.v1, id.cmv-immunocompromised.core.v1, id.hiv-initial.chronic.v1, id.opportunistic-infection.hiv-transplant.v1. Differential pivots authored as sibling feature rows + DIFFERENTIAL/BRANCHING phase logic: PJP vs bacterial CAP vs CMV vs viral (flu/COVID) vs TB vs invasive fungal vs lymphangitic carcinomatosis vs drug-induced pneumonitis vs IRIS. Special populations: HIV (CD4 band prior, strong steroid RCT evidence, early ART per Zolopa ACTG A5164, secondary prophylaxis stop at CD4 >200 ×3 mo) vs non-HIV (iatrogenic substrate, fulminant/higher-mortality, lower diagnostic yield → earlier BAL, weaker steroid evidence at same threshold, immunosuppressant minimization) as DISTINCT branches; transplant/biologic/chemo (route id.opportunistic-infection.hiv-transplant.v1); pregnancy (TMP-SMX still first-line + folate + delivery-timing for neonatal kernicterus, avoid primaquine); G6PD (screen before dapsone/primaquine — absolute contraindication); severe sulfa allergy (alternatives vs desensitization for non-severe only); renal/hepatic dose-adjust per drug in contraindication_rules. Regimen axis: TMP-SMX 15-20 mg/kg/day TMP weight-based (mild PO / moderate-severe IV) + Bozzette adjunctive steroid taper (40 BID×5d → 40 daily×5d → 20 daily×11d) + alternatives by sulfa-allergy/G6PD/severity (clindamycin+primaquine, atovaquone mild, IV pentamidine severe, dapsone-TMP, sulfa desensitization for non-severe reactions) + duration HIV 21 d / non-HIV 14-21 d + secondary prophylaxis with explicit stop criteria (CD4 >200 ×3 mo HIV; immunosuppression resolved non-HIV) + deprescribing + explicit leucovorin-AVOID rule. Matches pulm.cap.core.v1 / pulm.pe.core.v1 regimen depth. RxCUI sanity-check vs RxNav REST /rxcui/{cui}/properties.json 2026-05-16 — THREE SAFETY-CRITICAL CORRECTIONS: (a) TMP-SMX combination 10180 (=sulfamethoxazole single-ingredient IN) / 10829 (=trimethoprim single-ingredient IN) → 10831 = sulfamethoxazole/trimethoprim (MIN); (b) the prior dossier notes mapped dapsone→3008 (=cyclosporine), primaquine→8745 (=promethazine), pentamidine→8167 (=phenylhydrazine) — all WRONG; corrected to dapsone 3108 (IN), primaquine 8687 (IN), pentamidine 7994 (IN), clindamycin 2582 (IN), all RxNav-verified. Confirmed-correct unchanged: atovaquone 60212 (IN), prednisone 8640 (IN), methylprednisolone 6902 (IN), oxygen 7806 (IN); added caspofungin 140108 (IN), leucovorin 6313 (IN). No hand-authored CUIs — every CUI WebSearch/RxNav-confirmed. Run npm run research:rxnav:validate for dose-specific SCD/SBD resolution. SCHEMA-GAP NOTES: (1) _types.ts has no first-class field for Bayesian pretest/LR/post-test/decision-threshold or a host-conditional graph — encoded in severity_triggers, phase purpose/advance_when, calculator guideline_basis, regimen rationale, and the .depth.md LR table; (2) RequiredCalculator.drives lacks diagnostic_gate — A-a/host bands reuse severity_classification; (3) no pretest-prevalence field — captured in calc.aa_gradient guideline_basis + .depth.md; (4) no co-infection/overlap field — CMV/TB co-infection encoded via severity_triggers + workups branches_to + BRANCHING phase logic; (5) no host-class enum (HIV vs non-HIV) — modeled in FRAME phase + non_hiv_pjp_fulminant_high_mortality trigger + sibling feature rows. PRODUCTION blockers: (1) all RxCUIs RxNav-verified this pass but run npm run research:rxnav:validate before dosing automation (SCD/SBD); (2) calc.aa_gradient / calc.ckd_epi_2021 not confirmed individually in clinical-tools-registry this scope (reused existing pcp_pneumonia workup adapter id only — no registry edits); (3) no engine-specific test file under tests/ beyond the shared contract test; (4) no _design-brief.md inside the package path (separate from the in-scope _briefs/ deliverables authored this pass); (5) package path pneumocystis-jirovecii-pneumonia/ unchanged (manifest/package pointers out of scope, not touched).