Advisor-Platform Playbook: Verified Math Under the Advice

An advisor platform does not need a math vendor to give advice — that is the one thing it must never delegate. It needs the arithmetic under the advice to be current, defensible, and on the record.

Advisor platforms — planning suites, RIA dashboards, client portals — sit closer to regulated advice than any other vertical in this series, which makes their integration question unusually crisp. The recommendation layer is the advisor's: fiduciary or suitability obligations, client context, judgment. The computation layer underneath — claiming-age comparisons, Roth arithmetic, RMD divisors, refinance break-evens — is exactly what a verified-model dependency is for. This playbook is the division of labor and the mechanics of wiring it.

The models that fit the workflow

The advisor conversation runs on decision comparisons, and the catalog's retirement and tax families map onto its recurring moments: Social Security timing for the claiming conversation (in ratios of the client's own SSA-stated benefit), Roth-versus-traditional and the roth-conversion-ladder and backdoor-roth models for the tax-location conversations, RMD for the distribution calendar, capital-gains and quarterly-estimates for the realization conversations, long-term-care for the insurance-decision arithmetic around a real quote. The catalog's boundaries matter as much as its coverage here: no benefit estimation from earnings records, no premium invention, no suitability verdicts — the non-coverage piece (/writing/what-worthune-does-not-model) is worth an advisor-platform product team's full read, because every absence on that list is a place where the advisor's judgment or a specialist system belongs.

The surface: API, with envelopes in the client file

Advisor platforms should integrate through the backend API rather than embeds: the numbers feed documents, meeting prep, and client records, so the platform needs the full envelope, not a rendered frame. The load-bearing habit is filing the envelope with the client record whenever a computed number enters a plan, a meeting agenda, or a client-visible document. That single habit gives the platform the books-and-records posture (/writing/books-and-records-envelope) and the replay procedure (/writing/audit-replay) for free: when a client, a principal, or an examiner asks why a plan said what it said, the answer is a retrieval and a hash check, with the spec version and assumptions attached.

The advice boundary, operationalized

The platform's copy discipline does the boundary work. Model outputs render as what they are — the arithmetic of a scenario under stated assumptions — and recommendation language stays in the advisor's voice, entered by the advisor. The distinction is not cosmetic: a platform that auto-generates "you should claim at 70" from a timing comparison has moved the software into the recommendation seat, with everything that implies for the firm's compliance posture. The honest rendering is comparative — claiming at 70 yields 24 percent more monthly than at a full retirement age of 67, under the model's assumptions — with the recommendation left to the human whose name is on the relationship. The disclaimer that rides every envelope reinforces the same line, and platforms should keep it visible rather than designing it away.

Migration mechanics

Most advisor platforms arrive with homegrown calculators accumulated over years, which makes this vertical the natural consumer of the migration guides: shadow the legacy math against the models, classify the diffs, and cut over with the diff log as the compliance artifact (/writing/backfilling-legacy-calculator). Two vertical-specific notes. First, migrate the constants exposure before anything else — an advisor platform citing a stale contribution limit in hundreds of client plans is the highest-blast-radius version of the stale-constant failure, and pointing plan copy at registry-backed values is a smaller project than replacing engines. Second, sequence by conversation, not by code: the claiming-age and Roth conversations are where advisors most often get challenged on numbers, so verified math earns its keep there first.

• ✓Integrate via the API; file the envelope with the client record at every trust boundary.
• ✓Keep recommendation language human — models compare, advisors advise.
• ✓Surface spec links and assumptions in the advisor view, not just the client view.
• ✓Pin versions in CI; route bumps through your existing review workflow.
• ✓Migrate constants exposure first, then the highest-challenge conversations.
• ✓Attribution: visible Powered by Worthune where API results reach end clients, per the pricing page.

The trade this vertical is really making

An advisor platform that externalizes its arithmetic is not outsourcing its value; it is refusing to spend its credibility maintaining undifferentiated math. The advisor's edge was never the RMD divisor — it is the judgment about the client's life around it. A computation layer with published specs, sourced constants, and per-answer records lets the platform put its engineering where its differentiation is, and lets every number an advisor shows a client carry a provenance the firm can defend years later. This playbook is an integration guide, not compliance or legal advice; the firm's obligations stay with the firm.