§0 — Fresh Case State
James Malik term-life case seeded with its team assigned (Maria Chen, Dr. Raj Patel, Tom Boyd, Priya Shah); the records-api-mock holds his profile, the cascade will pull it via MCP
James Malik term life case seeded with team assignments. Case ref LI-ITEST-JAMES-MALIK-2026-001. No agents have run yet — initial state. Team: Maria Chen (Case Manager), Tom Boyd (Underwriter), Dr. Raj Patel (Medical Director), Priya Shah (Policy Admin). Seed ran in beforeAll BEFORE auth sign-in, so role sessions are fresh and not invalidated by password resets. Each role-holder authenticated via real /auth form login (not REST token injection). Each role-holder will complete their assigned nodes via the real UI.
§1 — Webhook Fired
insurance-intake (D-15 service token) returns 202 and activates n1 — no platform logic between activation and the agent
Webhook POST /api/webhooks/insurance-intake → HTTP 202. n1 (Application Received) activated by the intake event. D-15 bearer token verified. D-32 cascade queuing n2 (Application Intake) via pg-boss.
§2 — Application Intake
n2 intake agent pulls James Malik live from the records-api-mock MCP — first_name/last_name/dob/occupation, not pre-seeded
li-application-intake-agent called get_application_by_id on records-api-mock (port 39801) and pulled James Malik real profile: first_name="James", last_name="Malik", dob=1988-03-14, occupation="Software Engineer", tobacco_use=false. Real MCP pull — not pre-seeded.
§4 — Evidence — Case Manager
Maria Chen (Case Manager) completes the evidence-ordering nodes through the real UI; the cascade fans out into the parallel Lab / APS / Background-MVR branches
Maria Chen (Case Manager) completed n3 → n4 → n10 → n11 via UI. Three parallel evidence branches now active: Lab (n4→n5→n6), APS (n7→n8→n9), Background/MVR (n10→n11→n12). Each completion recorded actor_type="user" attributed to case_mgr@meridian.demo.
§6 — Medical Review — Medical Director
Dr. Raj Patel (Medical Director) reviews the attending-physician statement and approves the lab-analysis agent's output — every action recorded as the actual reviewer
APS evidence branch complete. n7 (APS Request Sent — agent), n8 (APS Document Received — Maria Chen), n9 (APS Medical Review — Dr. Raj Patel). Dr. Patel reviewed the attending physician statement and confirmed clean medical history consistent with Preferred Plus. UI completion = actor_type="user" attributed to medical@meridian.demo in graph_events.
§8 — Risk Classification → Gate
the risk-classification agent writes risk_class=preferred_plus; Tom Boyd (Underwriter) approves the agent output and routes the n13c gate to Preferred Plus → n17, skipping Standard Review
li-risk-classification-agent ran after all 3 evidence branches converged. Profile: clean APS (Dr. Patel reviewed), clean MVR/background, non-tobacco, age 38, Software Engineer. Agent wrote case.underwriting.risk_class="preferred_plus". n13c condition gate will route to n17 (Policy Document Generation).
Tom Boyd (Underwriter) completed n13c (Risk Classification Routing) via UI branch picker. "Preferred Plus" selected → cascade routes to n17 (Policy Document Generation). n14 (Medical Director Review), n15, n16 (Standard Review branch) skipped. actor_type="user" attributed to underwriter@meridian.demo in graph_events.
§9 — Policy Document Generation
n17 policy-document agent runs on the Preferred Plus path; n14/n15/n16 (Standard Review) are skipped
n17 (Policy Document Generation — agent) complete. Preferred Plus path: n1→n2→n3→...→n13→n13c→n17. n14/n15/n16 (Standard Review) are skipped. n18 (Policy Delivery — Priya Shah) is now eligible.
§5 — Generated Documents — Review Surface
5 real PDFs generated in the natural cascade surface in the node Artifacts tab for human review (D-44 Phase 1 + 2), shown in cascade order. n2: Meridian Term Life Application. n7: APS Request Letter (D-44 Phase 2 — TERM_LIFE_APS_REQUEST_PIPELINE; physician sourced from MCP pull) — the request goes out first. n6: Lab Results Report + Lab Analysis Summary (evidence + judgment together for Dr. Patel — HIPAA gate). n17: Policy Schedule. Each document shows status=generated with Download and Preview. This is Step 5: the agent-generated docs are reviewable before the human approves the node.
n2 (Application Intake) Artifacts tab — agent generated the Meridian Term Life Application PDF. Shows the filled AcroForm: James Malik, DOB 1988-03-14, Software Engineer, $750k face value coverage. Populated by the li-application-intake-agent via fill_artifact (D-44). Maria Chen (Case Manager) and the team can download and review this document here.
n7 (APS Request Sent) Artifacts tab — agent generated the APS Request Letter PDF (D-44 Phase 2, TERM_LIFE_APS_REQUEST_PIPELINE). The letter is addressed to the attending physician sourced authentically from the n2 MCP pull (case.physician.*). It requests a full attending physician statement for James Malik, itemising the specific medical history, diagnosis, treatment, and prognosis information Meridian Life requires for underwriting. Generated at n7 before Maria receives the incoming APS at n8 — the request goes out first, then the labs come back, then the policy issues. This is the headline new document shipped in D-44 Phase 2.
n6 (Lab Analysis) Artifacts tab — TWO documents generated by the lab-analysis agent (D-44 multi-output, E2): (1) Lab Results Report: normalized biomarkers parsed from Maria's uploaded vendor PDF — BMI 22.4, BP 118/76, chol 175/95/62, glucose 88, HbA1c 5.2, nicotine Negative. (2) Lab Analysis Summary: the agent's clinical judgment — preferred_plus classification, confidence 0.99. HIPAA gate: n6 entered awaiting_review so Dr. Patel must review both documents together before approving. This is the human-review surface — evidence (what the labs said) + judgment (what the agent concluded).
n17 (Policy Document Generation) Artifacts tab — agent generated the Policy Schedule PDF. Shows the issued policy: Policy No. ML-{id}, James Malik, $750k face value, $337.50/month premium, Preferred Plus risk class, 20-year term. Generated by the li-policy-issuance-agent via fill_artifact (D-44). Priya Shah (Policy Admin) reviews this document before completing policy delivery (n18).
§10 — Policy Issued — Policy Admin
Priya Shah (Policy Admin) completes delivery, the applicant-signature gate, receipt, and first premium; the policy is issued
Priya Shah (Policy Admin) completed n18 (Policy Delivery), n19 (Applicant Signature gate — Approved), n20 (Policy Receipt Confirmed), n21 (First Premium Collection), n22 (Policy Issued milestone). Full Preferred Plus journey complete: webhook → MCP pull → parallel risk evidence → classification → condition gate → policy issuance. All human steps by assigned role-holders, all recorded as actor_type="user" in graph_events with correct email attribution (not admin).
§11 — Per-Role Attribution Verified
hard assertions (L11): James / Malik / preferred_plus / n17 complete / n14 skipped — and every human event in graph_events is attributed to the actual role-holder, not the admin
HARD ASSERTIONS PASS (L11): first_name="James" ✓ last_name="Malik" ✓ risk_class="preferred_plus" ✓ n17=complete ✓ n14=skipped ✓ n22=complete ✓. graph_events actor_type='user' count=30. Per-role attribution: 13/13 correct. Case Record shows interleaved human completions (Maria Chen, Tom Boyd, Dr. Raj Patel, Priya Shah) and agent actions (intake MCP pull, lab analysis, risk classification, policy gen). All human steps completed through real UI — each role-holder's name visible in the Case Record.
§12 — Case Record + Audit
the case-wide event log shows human and agent actions interleaved, with real role-holder names — the answer to 'the record only showed agent actions'
Case Record filtered to show only human actions. Each completed human node shows the actual role-holder who completed it: Maria Chen (case_mgr@meridian.demo): n3, n4, n5, n8, n10, n11. Tom Boyd (underwriter@meridian.demo): n13c (+ n13 review approval). Dr. Raj Patel (medical@meridian.demo): n9 (+ n6 review approval). Priya Shah (policy@meridian.demo): n18, n19, n20, n21, n22. This is the Plan D proof: actor_type="user" events attributed to the actual role-holders, NOT to meridian@demo.com (admin). The prior run had 32 user events all attributed to admin.
Complete Preferred Plus journey. n1-n17 complete. n14/n15/n16 skipped. n18-n22 completed by Priya Shah. Policy issued. All human decisions by actual role-holders via real UI — authenticated via /auth form login. D-31: no fake completions, no admin surrogates for human steps.
Append-only event log showing interleaved human and agent actions. Human events (actor_type="user") show real role-holders — not admin. Agent events show the LLM pipelines. This is the core Plan D deliverable: the audit trail proves real human participation with correct per-role attribution.