SubrogationScreenerGKR

What it is

SubrogationScreenerGKR is the rules tool that turns a normalised claim payload, a resolved jurisdiction and the active corpus into a subrogation thesis. It reads records only: theories, defenses, gates, channel rules and the LOB spec's input_map with a source_path per fact. No threshold, pattern list, state rule or default lives in code, and a projection key a rule needs with no source path is a spec gap the generator reports, never a flat default .

Phase 0 gates

Gates run first and in order, before any theory is evaluated :

1. Comparative fault bar. Reads liability, then stated_fault_pct, then the clears_when conditions over the police determination and the citation. A further step of clears_when reads an admission, exercised by seed A20. The gate is evaluated against the jurisdiction's fault_system record, one record set shared with Property.
2. No-fault, per damage head. The per-head evaluator runs in Phase 0 and returns proceeds, barred or indeterminate for each head. The legacy file-level no-fault gate was deleted.
3. Waiver of subrogation. A not_stated on the waiver field is a verified absence and the gate clears; a not_returned leaves it indeterminate.
4. First-party payment. Fires the no_first_party_payment reason code when coverage does not respond or damages sit under the deductible. Unknown coverage never suppresses theory identification.
5. Candidate-target gating by roles. A theory targets party roles named in its own target_party_rule; a presence flag is published per targeted role, and an untargeted role is left without a flag so a trigger naming it evaluates indeterminate rather than false.

Coverage responds and waiver facts are read from core.policy, one path for all LOBs.

Theory evaluation

Each theory carries a requirement structure, not a flat flag. Requirements combine any_of and all_of groups, because a boolean cannot express a real evidentiary route: negligent entrustment needs one of two unfitness facts plus owner knowledge, with prior incidents as corroboration, and a flat all-required flag on alternative routes would guarantee the theory never fires. Fleet negligence is any_of the five narrative fields; if none is supplied the theory is indeterminate, not excluded .

Respondeat superior is narrative-required on trip_purpose_stated or assignment_stated; the route deviation question belongs to the frolic defense. Negligent entrustment requires an owner distinct from the driver, compared on party keys, not names: true only where an owner party and a driver party resolve to different keys, indeterminate where either is absent, so entrustment stays open pending a fact rather than being excluded on an absence.

Theory outcomes are confirmed, indeterminate or excluded, and an exclusion carries a closed reason code. The at-fault third-party driver theory (AUTO-006) carries a channel_rule record: arbitration_eligible false yields direct_demand, otherwise intercompany arbitration .

Defenses and caveat classes

Every defense in the active corpus is evaluated whether or not it fires, and carries a caveat_class. A routine_verification defense adds a routine condition to a referral; a material defense changes the referral posture. Waiver is routine; SOL bar, sovereign immunity, act of god and spoliation are material. A confirmed theory with a material defense open is viable with caveats and refers with a material condition .

The frolic defense (DEF-003) shows the ruled shape for a described-but-unproven signal: a described route deviation makes the defense indeterminate with a routine condition; it fires only on a stated deviation_personal_purpose_stated. The renderer must not relabel an indeterminate defense as not fired.

Money heads

Every head is one of stated, computed from stated inputs, pending, excluded or indeterminate. Nothing is estimated.

• recoverable_base is stated from the payload with the deductible shown alongside.
• legally_recoverable is pending until stated_fault_pct is supplied; the liability tool will move it to computed once it runs on this path.
• Rental is pending on rate and limit only. A rental total is never computed from rate times days; an adapter that did so was reverted twice by ruling.
• The bodily injury reserve is excluded at FNOL and shown separately from recovery.

Deadlines, verification items and the ledger

Deadlines are computed only for causes on the thesis, and each shows its anchor. A deadline with no date is never computed; the states are computed, accepted and overridden. Verification items name the fact that would resolve an open gate or defense, deduplicated. The completeness ledger lists every field each theory and gate read, marked stated with value and speaker, not stated, or not supplied, with decisive fields flagged; the receipt returned before the run must predict this ledger exactly.

What the trace records

Every gate, defense, exclusion and trigger appears in the trace with the fields it read and its outcome, fired or not. Every indeterminate names the fields it needed. The trace is the only source the thesis record is built from, by build_thesis_record(decision_trace), and the renderer reads only that record .

The Property path

The screener runs the same phases on a Property claim; the records differ .

• Gates. No-fault does not run. The comparative gate reads the fault_system record but is not_applicable when no fault facts are stated, which is the normal Property case. The waiver gate reads core.policy.waiver_of_subrogation_present and core.policy.lease_subrogation_waiver and applies each per theory through SUBRO-WAIVER-SCOPE-v1: the contract waiver bars contractor, tenant and landlord targets, the lease waiver bars tenant and landlord targets, and a stated waiver with a target outside its scope adds a material condition rather than a bar. The first-party payment gate reads the building coverage response.
• Theories. Six records, PROP-SUBRO-THEORY-001 to 006. Contractor and product confirm on a description of the work or the failure; utility, municipal and neighbouring-property theories require cause_attribution_to_party; tenant negligence requires tenant_conduct_described. Targets are gated by the roles the normalizer maps from the carrier vocabulary.
• Defenses. Five records, all material: waiver within scope (full bar), the property-damage limitation period (full bar), sovereign immunity (full bar), act of God (full bar only as the sole cause; with a party attribution the theory stands with a material condition) and spoliation on a discarded component (partial bar, theory viable_with_caveats).
• Conditions. A tenant or contractor target carries a routine condition to confirm it is not an insured or additional insured. Every material condition from a gate or theory reaches referral_decision.conditions, and a referral with a material condition is never the unconditioned refer label.
• Money. structure, contents and ALE stay as separate pd_components; recoverable_base is their sum. No Property deductible is in the carrier format, so none is printed.
• Deadlines. The property-damage limitation period from the date of loss. Statute of repose and government notice are not computed and are stated as not assessed.

Property Subrogation explains the theories as an adjuster reads them; Property Worked Examples walks eight seeds.

Worked example: Morales

Morales is a synthetic Texas auto claim, document-verified, and its canonical outcome is the acceptance test for the renderer . Referral: refer with conditions. Primary theory: respondeat superior confirmed against Rapid Bites LLC by the demand letter. At-fault driver Walsh confirmed, channel intercompany arbitration. Fleet negligence indeterminate: none of the five any_of fields was supplied. Negligent entrustment excluded because no owner other than the driver exists; Walsh owns the Sentra and Rapid Bites is the platform, not the owner. Product liability excluded. Comparative fault clears on the police determination and citation on the other driver; stated_fault_pct not supplied, so legally_recoverable is pending. Waiver indeterminate. Base 18,400 with deductible 500; rental pending at 40 per day, 30-day limit; medical reserve 28,000 to 35,000 excluded. One deadline: Texas negligence SOL 2027-07-22, two years from the 2025-07-22 date of loss. Two routine conditions.

Two lessons from getting it wrong. A seed once carried a BI figure of 12,500 that appears nowhere in the Morales documents; every figure on a brief must be in the source. And a sample payload sent not_stated for seven facts the canonical lists as not supplied, which cleared the waiver, excluded the contractor defense and excluded fleet negligence; the fields had to be omitted.

Open items

• The first-party payment gate emits not_fired where it should emit clears (a meaning error on the brief: "did not apply" on a claim with a stated payment).
• The exclusion reason code on negligent entrustment is free text and must be an enum.
• Defense caveat text in the trace is truncated mid-sentence; source unknown.
• JurisdictionResolverGKR returns product-liability and contract SOL rows for a claim with no such theory.
• Provenance: why seed A22 marks the Walsh employment facts as spoken by the insured.
• The comparative gate's source path after the retirement of police_faulted_other and other_party_cited must be verified and conformance re-run (A3).
• Property, carried in SL-FIX-07P: a stated waiver with a target outside its scope ends the run (exit 3) instead of adding a condition; the theory-level act-of-God defense still excludes when a party is named as the cause; the insured-target condition is not emitted; a discarded component reads as undetermined disposition and the decision cites the waiver instead of lost evidence; the brief prints "modified comparative" for every state instead of the label on the fault_system record.
• Property, outside SL-FIX-07P: statute of repose and government notice not wired; Florida property negligence period record (four years carried, two years from 2023-03-24); one candidate per theory under format v1.0.
• An AUTO spoliation defense on AUTO-004 with an A13 removal axis; posture enum and channel-rule evaluator debt; the "3e vehicle bridge" and vehicle-prefix split are runner mappings logged as debt.