A model catalog earns as much trust from what it refuses as from what it ships. This piece is the refusals, in one place, with the reasoning that a spec's exclusions section states one model at a time.
Worthune specs carry assumptions-and-exclusions sections, every spec documents its known issues, and every model rejects inputs outside its published domain. But exclusions read one spec at a time understate a larger pattern: whole categories of financial computation are absent from the catalog on purpose. Naming the categories in one place does two jobs — it saves integrators discovery time, and it makes the boundary itself inspectable. Here is the map of what is not there, and why.
Lender-priced and underwritten numbers
The catalog computes loan mechanics — amortization, break-evens, comparisons — from rates the user supplies. It does not price credit: no credit-score modeling, no rate quoting, no approval-odds estimation, no debt-to-income qualification verdicts. The reason is structural. A lender's price comes from the lender's underwriting box, which is proprietary, changes without notice, and varies borrower by borrower. A published, versioned spec cannot honestly describe a number that a third party sets privately. The same logic covers insurance premiums: the long-term-care model takes a real quoted premium as an input and analyzes the decision, precisely because inventing a premium would be underwriting without an underwriter.
Forecasts dressed as computations
No model in the catalog predicts markets. Growth rates, returns, appreciation, and inflation are always user-supplied inputs with published domains — never outputs. A deterministic engine can compute what follows from an assumed 7 percent return; nothing can verify what next year's return will be, and a verification-first platform does not ship unverifiable outputs. The same discipline excludes Monte Carlo simulation from the verified catalog: the models compound explicit assumptions and say so in their assumptions arrays, rather than wrapping randomness in an authority it has not earned. (The FIRE calculator's illustrative probability bands are the exception that proves the rule — they live outside the spec-and-harness regime, and the FIRE spec says so, precisely because they cannot be verified the way the deterministic models are.)
Benefit estimation from personal records
The Social Security timing model compares claiming ages in ratios of a benefit the user brings from their SSA statement. It does not estimate benefits from earnings histories — that computation requires the bend points and the user's actual earnings record, which SSA holds and Worthune does not. The claiming-factors piece (/writing/social-security-claiming-factors) walks this boundary in detail; the pattern generalizes to any computation whose honest inputs live in a government or employer system of record.
| Absent category | Why it is absent | What exists instead |
|---|---|---|
| Credit scoring and rate quoting | Lender-proprietary, unverifiable by spec | Loan mechanics on user-supplied rates |
| Insurance premium estimation | Underwriting without an underwriter | Decision analysis on real quotes |
| Market forecasting | Unverifiable outputs | User-supplied assumptions with published domains |
| Monte Carlo layers | Randomness wearing authority | Deterministic compounding with declared assumptions |
| Benefit estimation from earnings records | The record lives at SSA | Claiming-age comparison in ratios |
| Tax preparation | Filing requires the full return | Planning-scoped tax models with named scopes |
The tax-preparation line
Several models compute tax-shaped quantities — capital-gains brackets, quarterly estimates, Roth comparisons — and every one is planning-scoped: a decision aid computed from a handful of inputs, not a filing computed from a return. The catalog does not model itemized deductions, credits, phase-outs beyond the ones its specs name, computed state income tax (where a state amount appears, it is a flat user-supplied rate, never computed state law), or filing mechanics. The boundary is stated in each spec's exclusions and enforced by the input surface itself — a model with seven inputs cannot silently be doing your taxes, and that inspectability is the point of keeping input lists short.
Absence as a maintained artifact
This list is descriptive of the catalog as it ships today, and the roadmap page — themed packs, no dates promised — is the only authority on what changes. If a category moves from absent to shipped, it arrives the way everything arrives: a spec, a second implementation, published domains, and a changelog entry. Until then, the absence is doing real work. Every category above is one where a confident wrong number does more damage than a stated refusal, and the catalog's refusal is written down precisely so that no surface built on it has to improvise one.
Sources
- [1] Worthune model catalog. https://worthune.com/models
- [2] Worthune roadmap. https://worthune.com/roadmap
- [3] Worthune writing: Social Security Claiming Factors. https://worthune.com/writing/social-security-claiming-factors