Benchmark & limits

The current controlled build exercises six public carrier document families and 82 policy-structure fixtures, including authentic public Aggregating Specific Deductible, conditional deductible step-down, Bridge Renewal, advance-funding and mirroring structures. It achieved 514/514 expected resolved-field checks plus 3/3 explicit fail-closed checks. The reconciliation corpus contains 110 cases, including ASD arithmetic after the member Specific deductible/laser and a 100-row truth-labeled payment-linkage classification benchmark. Classification is 100/100 correct: 77 MATCHED, 12 AMBIGUOUS, and 11 UNMATCHED; ambiguous or unmatched reimbursement evidence holds fail-closed. Maximum arithmetic variance is $0, and there are zero positive reimbursement-exception outputs when a required predicate is unresolved.

Evidence-chain control artifact

Each evaluated member now carries explicit claim, submission, optional carrier acknowledgment, and payment evidence references plus factual linkage findings. The chain is called complete only when the submission explicitly identifies the reconciled claim references and the acknowledgment/payment links are evidenced; member/year proximity alone is insufficient. A missing carrier acknowledgment or claim-to-submission link is a control break; it does not reclassify the financial result, prove that money is owed, or override any unresolved reimbursement predicate.

What that does not prove

It does not establish production policy-packet extraction accuracy, real-customer exception prevalence, recoverable dollars, carrier acceptance, or authentic-packet reviewer time. Those require blinded/deidentified authentic files and buyer validation.

Fail-closed boundary

A six-case post-release carrier operational prerequisites benchmark covers Symetra separate-PBM Rx-experience receipt, Symetra final Plan Document acceptance, and QBE advance-reimbursement conditions. Three unresolved adversaries all hold, three resolved controls preserve the clean positive result, and the suite has zero unresolved-positive leakage. These carrier documents may add evidence holds but do not authorize policy terms.

Clinical, eligibility, contract-window, filing-window, COB/subrogation, denial, policy-term, reimbursement-evidence, carrier-operational, or operative deadline-exception uncertainty suppresses UNSUBMITTED_ELIGIBLE and UNDERPAID_REIMBURSEMENT conclusions. RFP/request text cannot authorize governing terms; ordinary run-out derived from claims basis is not treated as proof of terminal-liability election; renewal-only No New Laser language does not clear current lasers.

The Aggregating Specific Deductible is a decisive Specific-layer term. A positive ASD is consumed only after the applicable member Specific deductible or laser. When multiple members can consume a positive ASD, authenticated claimant allocation/chronology evidence is required; otherwise affected members hold. Conditional deductible step-downs, Bridge/Gapless renewal, advance/deemed-paid treatment, mirroring/policy-alignment and similar unsupported governing modifiers are detected and held rather than interpreted in V1.

Back