Equity Compensation: Three Models, Scoped to What Ships

Equity compensation is not one topic. It is a set of instruments with different tax treatments, and the fastest way to build a wrong tool is to model them as one. Worthune ships three equity-comp models; this is what each one actually computes.

The catalog's equity-compensation coverage is three models: rsu, espp, and mega-backdoor-roth. Incentive stock options, non-qualified options, AMT, and section 83(b) elections are not modeled — not because they are unimportant, but because each deserves its own spec rather than a corner of someone else's. That boundary is stated here before the arithmetic; each spec's assumptions section states what its own model excludes, and every response links its spec.

The RSU model: diversification, not withholding

The rsu model answers a portfolio question, not a payroll question. Its inputs: shares vesting per year, the share price, the holder's annual salary and filing status, a state tax rate, a holding period in years, the fraction of vesting shares sold each year, and a stock growth rate. Each year the model sells that fixed fraction of the vesting shares, computes the federal tax on the proceeds as the marginal slice stacked on top of salary — walked through the actual TY2026 progressive brackets, which live in the facts registry with their Revenue Procedure source — adds the flat state tax, and accumulates the after-tax proceeds. The output is the multi-year picture of what an annual sell-and-diversify discipline produces.

Two spec-documented quirks deserve an integrator's attention: the same number of shares vests and sells every year (nothing depletes a fixed grant pool), and the share price grows only after each year's sale is booked. Both are named in the spec as modeling conventions. What the model deliberately does not compute is the vest-time withholding gap — the difference between the flat supplemental withholding rate an employer applies and the holder's true marginal rate. That is a real and common RSU surprise, and it is outside this model's contract; nothing in the response estimates withholding.

The ESPP model: the immediate-sale case, priced exactly

The espp model prices one scenario — buy at the plan discount, sell immediately — because that scenario is the clean baseline every participation decision starts from. Inputs: the annual contribution, the discount percentage, the price at grant, the price at purchase, whether the plan has a lookback, and the holder's marginal ordinary rate. With a lookback, the discount applies to the lower of the grant and purchase prices; the model computes the purchase price, the shares bought, the bargain element, the ordinary-income tax on it, and the after-tax profit, with pre-tax and after-tax return rates on the contribution.

An immediate sale is a disqualifying disposition, and for that case the entire spread is ordinary income with essentially no capital gain — which is why one marginal rate is the only tax input the contract needs. Qualifying dispositions — holding two years from grant and one from purchase for partially favorable treatment — are a named v1 exclusion, not an implied feature.^[1] One more boundary the spec states plainly: the domain caps contributions near the twenty-five-thousand-dollar section 423 limit, but that statutory cap is measured in grant-date fair market value, which a discounted purchase can exceed on paper — a real plan would truncate the purchase, and the model does not recompute the cap.

The spec's FICA note is the kind of precision this catalog exists for: payroll taxes are excluded from the model correctly, because section 423 spreads are excluded from FICA wages by statute — regardless of disposition — under a rule added in 2004.^[2]

purchase_price = min(grant_price, purchase_price_at_date) × (1 − discount)  [lookback plans];  bargain_element = shares × market_price − contribution;  after_tax_profit = bargain_element × (1 − marginal_rate)

The mega-backdoor model: capacity and consequence

The mega-backdoor-roth model computes two things: how much after-tax 401(k) contribution room exists under the section 415(c) overall limit — seventy-two thousand dollars for 2026, read live from the facts registry, never hardcoded — after the employee's regular deferrals and the employer's contributions are counted; and what that contribution becomes over the years to retirement inside a Roth versus in a taxable account with annual tax drag. The comparison is the model's argument: the strategy's value is not the contribution, it is the compounding difference between tax-free and tax-dragged growth on the same dollars.

The model does not know whether a specific plan permits after-tax contributions or in-plan conversions — plan features are facts about the caller's plan document, not computable quantities; the caller must establish them before the capacity number means anything.

• ✓rsu — annual sell-and-diversify projection; progressive TY2026 federal + flat state; withholding gaps out of scope.
• ✓espp — immediate-sale disqualifying disposition; lookback honored; qualifying dispositions a named v1 exclusion.
• ✓mega-backdoor-roth — 415(c) capacity from the facts registry + Roth-vs-taxable growth comparison; plan support is the caller's fact to establish.
• ✓Not modeled anywhere: ISOs, NSOs, AMT, 83(b) elections — candidates only through the request mechanism on /roadmap.

Reading the responses

All three models return the standard envelope. The rsu response's facts array cites the TY2026 bracket-table registry markers; the mega-backdoor response always cites the 415(c) limit entry; the espp model consumes no registry constants and its facts array is empty. Each response carries its spec version, assumptions naming that version, and the record.sha256 over model, spec version, inputs, and outputs — so a stored vest-year projection or participation analysis can be recomputed and matched exactly when someone asks, years later, what the tool actually said.