Methodology & validation

Retirement tools notoriously disagree. The same plan can score 60% in one tool and 90% in another, and most tools won't show you why. RetireGolden's answer is to show its work.

One auditable ledger

Every result comes from a single year-by-year ledger you can open and read: the success rate, the ranked Roth-conversion schedules, the "how much can I spend" answer. There is no separate "estimate" engine. The Monte Carlo simulation, the optimizer, and the reports all price against that same projection. If two numbers in RetireGolden ever disagree, that's a bug, not a methodology difference.

Checked against independent implementations

RetireGolden's calculations are tested against third-party references, none of which share code with it:

• Federal and state taxes. Test cases are cross-checked against independent open-source tax engines maintained by policy-research communities.
• Roth-conversion optimization. A parity harness runs the same households through an independent open-source conversion optimizer, then prices both tools' schedules on RetireGolden's own ledger. It is re-verified against pinned releases on a maintenance cadence.
• Social Security. Claiming math is validated against the SSA's published rules (bend points, family maximums, survivor-benefit limits) and cross-referenced with open-source claiming calculators.

Asset-location invariance, proven

There is a bug class that inflates conversion benefits across retirement tools: quietly assuming Roth dollars are invested more aggressively than traditional dollars, so the reported "conversion benefit" is really a hidden allocation change. RetireGolden holds portfolio-wide asset allocation constant across account types, and a dedicated test suite proves it. A tax-free conversion between identically allocated accounts leaves every year's totals identical to the dollar, so the entire estate benefit of a conversion is the tax term, and a conversion candidate can never carry a hidden account or allocation change. If you choose different allocations per account, the engine prices that too, visibly, as your explicit setting.

Every parameter cites its source

Tax brackets, IRMAA tiers, ACA percentages, RMD tables, and Social Security parameters live in versioned data files with provenance. Each value traces to the Revenue Procedure, statute, or agency publication it came from, and is refreshed annually on a documented schedule. The app shows the sources and vintages it is using in its disclaimer and provenance panel.

The tax-rule registry

Every claim RetireGolden makes about tax law lives in one open registry record: quoted authority, a classification, and a verification date. A public page is generated from the same JSON the tests enforce, so no count on it can be maintained by hand.

Tax-rule registry: the verification chain

Every calculation catalogued

Every number the planner shows is an output family in a public census, and each family is being tied to a calculation record with a worksheet derived without running the engine, an evidence test, an executed mutation receipt, and an independent review. Today, 13 of 229 output families have a complete chain; the catalog page counts the rest by what is still missing, generated from the engine's own index so no count on it can be maintained by hand.

Calculation catalog: every output family and its evidence

What RetireGolden deliberately simplifies

Honest scope beats false precision. RetireGolden is planning-grade, not filing-grade:

• AMT is modeled as a planning screen, not to filing precision; per-lot basis and every credit phase-out are simplified by design.
• The optimizer's in-solve model uses documented simplifications (for example, a single long-term-gains rate inside the solve). Candidate schedules are then re-priced and validated on the full year-by-year ledger before you see them.
• Future-year tax parameters beyond published law are stand-ins, clearly marked.
• Nothing here is financial, tax, or legal advice. It's an educational model of current law. For the year you execute a major decision, verify with a professional.

Why it's free (and what that means for your data)

No accounts, no server, no data collection: your plan never leaves your device. There is nothing to sell because nothing is collected. See Sustainability & your data for what happens if the project ever winds down (short version: your data is a portable file you already have).