How It's Built
Lumis Life is not a spreadsheet with a web front-end. It is four interlocking actuarial systems — a health-adjusted mortality engine, a per-carrier underwriting estimator, a settlement-valuation engine, and a Social Security optimizer — built on public actuarial standards, validated against published literature, and engineered for reproducibility and audit. This page describes the architecture. It deliberately omits proprietary parameter values; what it does not omit is the scope and the rigor.
The mortality engine
The core is a health-adjusted mortality model built on the Society of Actuaries 2015 Valuation Basic Table (VBT) Ultimate as its baseline, projected forward with the MP-2021 mortality-improvement scale and extended into the oldest ages with a Kannisto logistic model where published tables run out. On top of that baseline sits an impairment layer that models 64 conditions across 15 clinical categories, each with severity tiers and treatment-status modifiers, plus functional impairment (activities of daily living) and a comorbidity layer of 79 condition-pair interactions where two conditions together are worse than the sum of their parts.
Every estimate is run through a Monte Carlo simulation that produces a full distribution of outcomes, not a single point — which is where the 90% confidence intervals (5th–95th percentile) on every report come from. The methodology is intentionally built on public standards (VBT, MP-2021) so it can be inspected and trusted; the calibration that maps conditions to mortality is the proprietary part, and it is validated against published clinical literature.
The underwriting estimator
Carrier underwriting is modeled as a pluggable architecture: each of the 18carriers is its own module implementing a shared contract, mapping that carrier's native class names onto a canonical 10-class taxonomy so placements are directly comparable. Rules are sourced exclusively from publicly available field underwriting guides, and each module carries a version stamp that tracks the guide it was built from. Adding carrier number 19 is an incremental module, not a rewrite — the architecture is built to scale. Read the underwriting methodology →
The settlement-valuation engine
For permanent policies, the platform runs a discounted-cash-flow settlement valuation: a two-year contestability gate, a buyer-conservative mortality margin, a projected premium stream and death-benefit stream, and a fee-and-tax stack — discounted across a band of buyer internal-rate-of-return scenarios to produce a defensible offer range. The same engine compares five strategies (hold, settle, surrender, reduced paid-up, and 1035 exchange) on an after-tax net-present-value basis.
Reproducibility & versioning
Fourteen components of the platform are independently versioned. Every assessment is stamped with the engine version, a SHA-256 hash of the complete parameter set, and the Monte Carlo seed — so any result the platform has ever produced can be reproduced exactly, and any change to the engine parameters is detectable as a change in the hash. Engine releases are git-tagged. This is the kind of audit trail that matters for fiduciary documentation and is rare in advisor-facing tooling. The snapshot powering the validation page was produced by engine v3.9.0.
Engineering rigor
- 2,500+ automated tests across the engine, underwriting modules, settlement math, and platform — including a regression test that guards against trade-secret leakage in any client-facing output.
- Encryption at rest and in transit (Fernet AES-128-CBC + HMAC-SHA256) for assessment data, plus CSRF protection and audit logging.
- A migration-chain validator that runs in CI to keep the database schema history single-headed and consistent.
- A clean separation between the public marketing surface and the private engine, so the proprietary calibration never ships to the browser.