The FNOL summary (Lesson 7) captured loss facts. The coverage analysis (Lesson 8) produced the reasoned position. The next four artifacts close the loop: the Reservation-of-Rights letter that preserves carrier coverage defenses while the investigation continues; the Examination Under Oath outline for the suspect first-party theft claim where SIU has flagged red-flag indicators; the SIU referral letter naming human-reviewable facts separately from the Shift Technology score; and the adverse-determination letter on any AI-influenced denial with FCRA-style reason language under FCRA §615 where applicable. Each artifact carries §4 reason chain; each survives state-specific bad-faith review under Texas Insurance Code §541, Florida §624.155, California Cumis/Brandt; each captures the AI-touched evidentiary chain for §4.4 documentation and discovery defense. By the end, you can take any coverage-analysis output + SIU red-flag pattern + adverse-determination context and produce all four artifacts with reason-code discipline intact and bad-faith exposure structurally managed in three different state jurisdictions.

The Reservation-of-Rights Letter - Preserving Coverage Defenses

The Reservation-of-Rights letter (ROR) is the carrier's preservation-of-defense document during the coverage investigation. The ROR acknowledges the claim, names the open coverage questions, and reserves the carrier's right to deny coverage or limit coverage based on the investigation's outcome - without binding the carrier prematurely. Bad-faith exposure rises when the ROR is too aggressive (effectively a denial in disguise), too soft (waives defenses), or fails to name specific coverage questions clearly.

System message. "You are a licensed P&C claims adjuster drafting a Reservation-of-Rights letter on a [LOB] claim in [state]. NAIC §4 reason-chain discipline applies. Cite policy forms by exact edition with Westlaw/Lexis identifier on every case cited. Do not invent case-law citations or paraphrase holdings. The ROR preserves coverage defenses; it does not deny. Language must enumerate open coverage questions; reserve specific rights; commit to good-faith investigation; provide insured with policy form reference for review. State UCSPA regimes apply to ROR timing and content. Bad-faith exposure under Texas §541, Florida §624.155, California Cumis/Brandt (depending on venue) shapes the language tone."

User message (Atlanta water-damage example). "Draft the ROR letter on the Atlanta water-damage claim for insured Joseph Marquez. Context: FNOL summary (Lesson 7) + coverage analysis (Lesson 8) attached. Policy HO 00 05 04 11 + water-backup endorsement HO 04 95 04 11 at $10K sublimit + ordinance HO 04 77 04 11 at $25K sublimit + Georgia state-amendatory HO 01 11 GA. Claimed damages $15K-$26K. Open coverage questions: damage scope above $10K water-backup sublimit; ordinance-or-law application to specific code-required upgrades; allocation between covered water-backup and any potentially-excluded damage components; mitigation-duty fulfillment regarding day-4 contractor engagement. Artifact: ROR letter (1) acknowledging the claim and prompt notice; (2) naming the policy form and applicable endorsements with sublimits; (3) enumerating open coverage questions specifically; (4) reserving rights to deny, limit, or modify coverage based on investigation; (5) committing to good-faith investigation per Georgia O.C.G.A. §33-6-30; (6) providing insured contact for questions; (7) preserving any time-limited responses required of insured. Constraint: no premature denial language; no waiver language; no case-law fabrication; mark uncertain [verify against retrieval]. Format: business letter format, 600-900 words, dense but accessible to non-attorney insured."

The EUO Outline for the Suspect First-Party Theft Claim

Switch context. The Atlanta adjuster's File 8 in the diary is a HO first-party theft claim where Shift Technology's fraud score flagged red-flag indicators: (1) prior loss at same address with different carrier ($28K paid 2023); (2) jewelry valuation $42K with sparse documentation; (3) reported loss occurred 11 days after policy effective date; (4) insured's stated value of items inconsistent with ACORD application valuations. SIU has approved an Examination Under Oath. The L2 task: produce the EUO outline with topic areas, specific questions, document requests, and §4 reason chain naming the human-reviewable facts separately from the Shift score.

EUO outline prompt (excerpt). "Produce the EUO outline for the suspect first-party theft claim on insured John Hardy. Context: FNOL (file produced 2026-05-10); coverage analysis (HO 00 05 04 11 personal property Section I-C named perils; theft covered subject to exclusions and policy conditions); Shift Technology fraud score 0.81 with red-flag indicators (prior loss + jewelry valuation + 11-day claim-after-effective + value inconsistency); SIU approval letter attached; defense counsel engaged. Artifact: EUO outline with (1) preliminary topic areas - identity, policy procurement, prior insurance history; (2) loss-specific topics - date/time/location of theft, alleged stolen items with valuations, sources of valuations; (3) documentation topics - ownership documentation for high-value items, photos, receipts, appraisals; (4) prior-loss topics - 2023 claim circumstances; (5) Shift-flagged inconsistency topics - application valuations vs. claim valuations; (6) post-loss conduct topics - police report, recovery efforts, insurer notification timing; (7) §4 reason-code memo naming human-reviewable facts separately from Shift score with protected-class proxy assessment. Format: outline with topic headings + specific questions under each + document-request list + EUO scheduling logistics, length 1,200-1,800 words."

The §4 reason chain on the EUO. The EUO is investigative, not adverse-action. The §4 reason chain captures: human-reviewable facts (prior $28K loss, $42K jewelry with sparse documentation, 11-day claim-after-effective, valuation inconsistency); Shift score cited as signal that triggered SIU review; AI signal separated from facts. Protected-class proxy assessment per insured's demographic surface reviewed against Georgia DOI fraud-investigation guidance + carrier bias-test exhibit; no proxy identified in the variable set. The L4 algorithm-inventory entry on Shift Technology deployment documents the bias-test on the fraud-score distribution.

The SIU Referral Letter Naming Human-Reviewable Facts Separately

The SIU referral letter formally routes the suspect claim from the adjuster's diary to SIU for full investigation. The letter enumerates human-reviewable facts that justify SIU involvement; the Shift score is cited as the signal that triggered the review, not as the reason for the referral. The structural insight: §4 reason-chain discipline requires the facts that justify action to be enumerated separately from the AI signal that triggered the review.

The SIU referral memo structure. (1) Identification - claim ID, insured, policy, loss date. (2) Predictive signal - Shift Technology fraud score 0.81 with cluster cohort (prior-loss + valuation + timing). (3) Human-reviewable facts - enumerated separately: 2023 prior loss at same address ($28K paid); $42K jewelry valuation with sparse documentation (no receipts for $32K of claimed items, no appraisals); 11-day claim-after-effective on a 12-month new HO policy (effective 2026-04-29, loss reported 2026-05-10); valuation inconsistency between ACORD 4 application (jewelry rider $15K) and claim (jewelry alleged stolen $42K - almost 3x application). (4) Recommended actions - EUO scheduling, records subpoena (jewelry appraisals from claimed sources, prior-carrier 2023 file, bank records of jewelry purchases), independent appraisal, police-report follow-up. (5) Bias-test cross-check - cluster composition reviewed for protected-class proxy under Georgia DOI fraud-investigation guidance + carrier bias-test exhibit 2026-Q1; no proxy identified. (6) Governance trail - Shift model version logged, override decision logged if adjuster initially disagreed, SIU manager review checkpoint.

The structural failure mode the format prevents. A SIU referral that reads "Shift flagged this at 0.81 - referring to SIU" fails §4 reason-chain discipline. The carrier owns the §4 obligation; the §4 examiner walks the referral and asks "what facts justify the action?" If the answer is the score, the carrier fails. Naming the four human-reviewable facts (prior loss + sparse documentation + timing + valuation inconsistency) separately from the score produces the audit-defensible referral.

The Adverse-Determination Letter with FCRA §615 Reason Codes

The adverse-determination letter is the carrier's denial or limited-coverage notice to the insured. If the determination rests on AI-influenced analysis that includes consumer-report data (ISO ClaimSearch is consumer-report-adjacent on personal-lines claims; LexisNexis MVR + Rx data on L&H claims; FICO-style credit-based-insurance-score data on auto/HO underwriting), FCRA §615 adverse-action notice requirements apply. The letter must provide: the action taken, the consumer-report agency(ies), the consumer's right to dispute, the consumer's right to free annual report, and the reasons for adverse action.

FCRA §615 requirements summary. Section 615(a) of the FCRA requires that any person who takes adverse action against a consumer based on a consumer report provide: (1) oral, written, or electronic notice of the adverse action; (2) the name, address, and telephone number of the consumer reporting agency; (3) a statement that the consumer reporting agency did not make the decision; (4) notice of the consumer's right to obtain a free copy of the report; (5) notice of the consumer's right to dispute the accuracy of the report.

Adverse-determination letter prompt. "Produce the adverse-determination letter on the Atlanta first-party theft claim. Context: SIU investigation completed; EUO conducted; documentary evidence reviewed; finding - material misrepresentation regarding pre-loss ownership of $32K of claimed items; carrier determination - claim denied based on material misrepresentation provision in HO 00 05 04 11 + coverage limitation on items without pre-loss ownership documentation. AI involvement - Shift Technology fraud score 0.81 triggered SIU review; ISO ClaimSearch returned 2023 prior loss as one factor in the cluster analysis. Artifact: adverse-determination letter with (1) salutation and acknowledgment of claim; (2) statement of action taken (denial with limited-coverage option); (3) reasons for action - variables enumerated, AI signal separated from human-reviewable facts; (4) FCRA §615 disclosure where ISO ClaimSearch consumer-report-adjacent data drove adverse action: consumer reporting agency identification (ISO ClaimSearch), consumer's right to dispute, consumer's right to free annual report; (5) §4 reason-code section enumerating variables + protected-class proxy assessment; (6) appeal rights under Georgia O.C.G.A. §33-6-30 + carrier's complaint-handling process; (7) contact information for questions. Constraint: no premature denial of items that may have been pre-loss-owned (limit denial to documented-misrepresentation items); preserve appeal rights; cite policy form by edition; mark uncertain [verify]. Format: formal adverse-determination letter, 800-1,200 words, FCRA-compliant disclosure language."

The Bad-Faith Exposure Assessment Across Three States

The carrier's bad-faith exposure differs by state. The L2 adjuster operating across multi-state book runs a per-state bad-faith assessment on every adverse-determination decision before the letter issues. Three states matter most for 2026 personal-lines and small-commercial: Texas, Florida, California.

Texas Insurance Code §541 - Unfair Settlement Practices. §541 codifies a list of unfair claims-settlement practices and provides for statutory damages on practice-level violations. Bad-faith exposure on AI-touched adverse determinations: the practice-level pattern matters more than individual file outcomes. If the carrier systematically issues adverse determinations on AI-flagged files without §4 reason-chain discipline (variables enumerated + protected-class proxy reviewed + AI signal separated from facts), the carrier exposes to practice-level §541 claims. Mitigation: per-artifact §4 reason chain + L4 algorithm-inventory entry + bias-test exhibit on the AI signal source + override discipline.

Florida §624.155 - Civil Remedy Notice. §624.155 creates a private right of action with a 60-day Civil Remedy Notice (CRN) cure window. After CRN, if the carrier has not cured, the insured may file a §624.155 action; statutory damages + attorney's fees + extracontractual damages possible. Bad-faith exposure on AI-touched adverse determinations: the 60-day cure window requires rapid retrieval of the prompt log, model version, RAG sources, override discipline applied. The carrier reviews the AI-influenced disputed action within 60 days and either cures (revised decision with documented reason) or accepts the bad-faith exposure. The retention discipline turns from documentation hygiene into operational necessity.

California Cumis / Brandt. California's Cumis doctrine pierces defense-counsel control where the insurer's coverage position creates a conflict of interest with the insured's defense; the insured may select independent counsel at the carrier's expense. Brandt fees compensate the insured for attorney costs incurred to recover policy benefits. Bad-faith exposure on AI-touched adverse determinations: AI-driven coverage positions that arguably conflict with the insured's defense (e.g., the AI surfaces a coverage defense not yet raised by the insured's defense counsel) may trigger Cumis. Brandt fees apply where insured incurred attorney costs to recover wrongfully-denied benefits.

The §4 reason-chain operates across all three states. Variable enumeration + protected-class proxy assessment + AI signal separated from facts + verification chain + retention discipline + override pattern - these survive Texas §541 practice-level review, Florida §624.155 60-day cure cycle, California Cumis conflict review. The architecture is durable across jurisdictions because §4 is durable.

The L4 Algorithm Inventory on the Adverse-Determination Chain

The L4 algorithm-inventory entries covering the L2 Chapter 3 artifact set name: Shift Technology fraud-scoring deployment (model card + bias-test on fraud-score distribution + drift-monitoring + SIU-manager checkpoint); ISO ClaimSearch deterministic-reference-layer entry (match-validation discipline + FCRA-adjacent disclosure pattern); Five Sigma claims-core (FNOL template + coverage-analysis template + ROR template + EUO template + SIU referral template + adverse-determination template); carrier enterprise LLM (ROR drafting + adverse-determination drafting + EUO outline generation with RAG to ISO Mercury + Westlaw/Lexis + state-amendatory + carrier coverage-precedent + bound endorsement schedule + state UCSPA publications); cession-broker workflow (not applicable on these artifacts).

Each L4 entry names accountable executive (chief claims officer / VP of SIU / coverage counsel per scope); model card or system prompt with version control; bias-test exhibit with cadence (typically quarterly for high-risk); drift-monitoring runbook with PSI/KS/AUC thresholds; override discipline (adjuster review + SIU manager checkpoint + coverage counsel sign-off on adverse determinations); §4.2 contract addendum compliance status; §4.3 testing-validation cadence; §4.4 documentation completeness; incident-response runbook with category-level pattern monitoring; FCRA/UCSPA/MHPAEA regulatory chain attachment; state-overlay alignment.

The Full Chain from Bind Decision to Discovery Defense

Looking back across L2 Chapters 1-3, the Acme Warehousing UW workflow + the Atlanta water-damage claims workflow + the Atlanta first-party theft SIU workflow produce parallel six-to-nine artifact chains that survive §4 examination, state UCSPA enforcement, bad-faith litigation, and FCRA-adjacent consumer-rights review.

UW chain (Acme): triage memo → submission summary → appetite memo → reason-code memo → quote-with-restriction memo → treaty-cession recommendation. Six artifacts; six signatures; one §4 audit trail.

Claims chain (Atlanta water-damage): FNOL summary → coverage analysis → ROR letter. Three artifacts at coverage-determination stage; reserve + settlement chain continues separately.

SIU chain (Atlanta first-party theft): FNOL summary → coverage analysis → SIU referral → EUO outline → adverse determination. Five artifacts at adverse-determination stage; appeal + complaint-handling chain continues separately.

Each chain operates under the same architecture: L2 prompt template + L3 RAG retrieval + L4 retention + L4 algorithm-inventory entry + L5 board narrative. Each artifact carries §4 reason chain; every adverse recommendation enumerates variables; every AI signal cited as signal with human-reviewable facts separately; every protected-class proxy assessment documented; every verification chain logged. The discovery-defense posture survives because the architecture is durable; the regulatory-defensibility posture survives because §4 is durable; the operational posture survives because per-artifact verification habit catches residual hallucination at 60-second scan.

Key Takeaways

• Four artifacts close the L2 Chapter 3 claims workflow. ROR letter preserves coverage defenses; EUO outline structures the suspect-first-party-theft investigation; SIU referral letter routes to fraud team with human-reviewable facts separately from Shift score; adverse-determination letter with FCRA §615 reason codes where consumer-report data drove adverse action.
• The ROR enumerates open coverage questions and reserves rights specifically. Acknowledges claim; names policy form + endorsements with sublimits; reserves specific rights to deny, limit, or modify based on investigation; commits to good-faith investigation under state UCSPA. No premature denial language; no waiver language; no case-law fabrication.
• The EUO outline structures the suspect-claim investigation. Topic areas (identity, policy procurement, prior insurance, loss-specific, documentation, prior loss, Shift-flagged inconsistency, post-loss conduct); specific questions under each; document-request list; EUO scheduling logistics. §4 reason chain names human-reviewable facts separately from Shift score.
• The SIU referral letter names four human-reviewable facts separately from the Shift Technology score. Prior $28K loss at same address; $42K jewelry valuation with sparse documentation; 11-day claim-after-effective on new policy; ACORD 4 application valuation inconsistency. Score cited as signal that triggered review, not reason for referral.
• The adverse-determination letter provides FCRA §615 disclosure where ISO ClaimSearch consumer-report-adjacent data drove adverse action. Action taken, consumer reporting agency identification, consumer's right to dispute, consumer's right to free annual report, reasons for adverse action. §4 reason chain enumerates variables with protected-class proxy assessment.
• Bad-faith exposure assessment differs across Texas, Florida, California. Texas §541 statutory damages on practice-level violations; Florida §624.155 with 60-day Civil Remedy Notice cure window; California Cumis pierces defense-counsel control + Brandt fees compensate attorney costs to recover wrongfully-denied benefits.
• The §4 reason chain operates across all three states. Variable enumeration + protected-class proxy assessment + AI signal separation + verification chain + retention discipline + override pattern survive Texas §541, Florida §624.155, California Cumis. Architecture durable across jurisdictions.
• The L4 algorithm-inventory entries cover six surfaces. Shift Technology fraud-scoring; ISO ClaimSearch deterministic reference; Five Sigma claims-core templates (FNOL + coverage analysis + ROR + EUO + SIU referral + adverse determination); carrier enterprise LLM (RAG-bounded drafting); §4.2/§4.3/§4.4 compliance per surface.
• The full UW + claims + SIU chains produce parallel six-to-nine artifact §4 audit trails. UW chain (Acme): 6 artifacts. Claims chain (Atlanta water-damage): 3 artifacts at coverage-determination stage. SIU chain (Atlanta first-party theft): 5 artifacts at adverse-determination stage. Each chain operates under L2 prompt template + L3 RAG retrieval + L4 retention + L4 algorithm-inventory entry + L5 board narrative. Each survives §4 exam, state UCSPA enforcement, bad-faith litigation, FCRA-adjacent consumer-rights review.