Check the Worksheet

Usability findings on the homepage — 2026-09-16

Origin. These are findings about this site’s own homepage, surfaced as a byproduct of testing whether Claude’s design-usability skill (built and tuned against Shopify app screens) also works on a non-app, content-argument surface. The skill-fit question was the actual task; this file exists so the four things noticed about the live page don’t evaporate when that session ends. Not triaged, not fixed, and not a design-department deliverable — just a faithful record of what a Nielsen heuristic evaluation, run once, found and didn’t find.

Method: source-only read of index.html, _layouts/home.html → _layouts/default.html, _includes/header.html/footer.html/disclosure.html, assets/js/calculator.js, and assets/css/site.css. No rendered browser pass — see Scope note at the end.

Surface reviewed: the homepage only, as the page a cold, possibly skeptical first-time visitor lands on and decides within seconds whether to keep reading.

Findings

Each finding: heuristic, file:line, evidence, severity (Nielsen 0-4: 0 not a problem, 1 cosmetic, 2 minor, 3 major, 4 catastrophe), and what fixing it would take.

F1 — Numeral-pair orphaned from its context (Heuristic 6, Recognition rather than recall)

Severity: 1 (cosmetic).

index.html:342-352. The “40 → 57” numeral-pair (captioned “What Line 7e reports, in percent of income” → “The payor’s true share of net income when it finally does”) sits directly after the “What the Worksheet does that its text doesn’t say” heading and its one-line intro (index.html:336-339), but before the four finding cards. The card that actually explains these two numbers — f-hardship, “The hardship test reads a different income than the order pays from” — is the second of the four cards below it (index.html:377-396).

A first-time reader hits two bare percentages with only a caption, no link or heading tying them to the hardship finding, and has to keep scrolling to get the “why” — a brief, self-resolving disorientation rather than a real barrier.

Cheap fix: either move the numeral-pair to sit directly above/inside the f-hardship card, or add one clause to its wrapping context (e.g. an eyebrow line “the hardship test:”) so the pair reads as already-labeled rather than a preview a reader has to hold in mind.

F2 — No reset control on the calculator (Heuristic 3, User control and freedom)

Severity: 2 (minor).

index.html:58-292, the “Try it: compute your own two incomes” calculator. The entire page’s narrative and every stat claim on it are pinned to one worked example — $201,000 / $29,640 / 3 children, no child care. The calculator’s sliders default to those values, but nothing on screen restores them once a reader drags a slider away. The only paths back are (a) manually retyping all of higher-earner income, lower-earner income, children, custody box, and child care to their exact original values, or (b) a full page reload, which also loses scroll position.

Checked calculator.js for a reset/restore/default-value handler — none exists (grep -n "reset\|restore" assets/js/calculator.js returns nothing relevant).

Not severe — reload is a full, low-cost recovery path, and nothing is destroyed — but it’s a real gap on a tool whose whole point is comparing against a specific reference example.

Cheap fix: one small “Reset to the worked example” link/button next to the sliders that re-applies the six default values and re-renders.

F3 — Worksheet line numbers unglossed on first use (Heuristic 2, Match between system and the real world)

Severity: 2 (minor) — and see the caveat below.

index.html:126-131. The calculator’s readout labels a reader’s very first result as “Weekly order (Line 7d)” and “Line 7e’s reading,” with no inline explanation of what Line 7d or Line 7e is structurally, the first time either appears. A visitor who has never seen Massachusetts form CJ-D 304 meets a worksheet-internal citation before any plain-language gloss of what it means. Heuristically this is the same failure shape as a raw enum or internal exception name surfacing as UI copy — jargon presented as if the audience already has it.

This is not an oversight to fix on sight. The site’s stated method — see the disclosure at _includes/disclosure.html and its worked example (“The model behind every number here is model/worksheet.py… checked against the form’s own calculation scripts”) — is built entirely on traceability: every figure cites its exact line on the actual government form so a skeptical reader (or the form’s own author) can go check it. Citing “Line 7e” by name is that traceability promise being kept, not a leak. Removing or hiding the citation would undercut the site’s central credibility mechanism.

This is a judgment call for the site’s author, not a defect: the options are (a) leave it as-is, on the reasoning that the target reader for a policy-verification site is expected to either already know the form or be willing to look it up — traceability outranks approachability here; or (b) add a one-time inline gloss the first time “Line 7d”/”Line 7e” appears (e.g. a <dfn>/tooltip: “Line 7d — the worksheet’s own line for the weekly order amount”), which would help a genuinely cold reader without weakening the citation. Either is defensible; this note exists so the trade-off is visible, not to argue for either side.

F4 — The same qualifier repeated three times in the first screen (Heuristic 8, Aesthetic and minimalist design)

Severity: 1 (cosmetic).

The phrase “the same two incomes and three children” (or a close variant: “one set of incomes,” “one worked example”) appears three times before a reader has scrolled past the first exhibit pair:

Each instance is individually load-bearing — it’s the site’s key methodological caveat and the web-page-anatomy/disclosure convention calls for it to travel with every figure — but reading top-to-bottom it lands as the same sentence four times in one screen’s worth of content. Not a correctness problem, and arguably intentional given the disclosure convention; flagged because a first-time skim is exactly the moment repetition costs the most attention.

Cheap fix, if the author wants one: vary the wording on at least one of the four instances, or drop the exhibit-pair confidence caption where the lede/tag directly above the same section already said it seconds earlier.

Heuristics and checks with no violation found

Stated explicitly, per the evaluation method, so nobody re-audits this ground later:

A deliberate trade-off that was checked and NOT flagged

The site’s own wordmark (_includes/header.html, styled at assets/css/site.css:142-146) is a link to / with no visible link affordance — no underline, and its :hover only shifts text color. Read in isolation, the discoverability “signifiers, not affordances” check would flag this: it’s clickable but doesn’t look clickable.

It isn’t flagged as a finding here because header.html’s own comment already reasons through exactly this and mitigates it deliberately: “Home is a deliberate repeat of the wordmark. The wordmark is a link but does not look like one, so a reader has no obvious way back. Duplicated affordance beats a hidden one.” The nav bar carries an explicit “Home” text link right beside the wordmark for this reason. Checked that the mitigation is real (the “Home” link is present, styled as a normal nav link, and marked aria-current="page" on /) — this is a decision already made and defended in the code, not a gap to re-litigate.

Scope note

This was a source-only read: HTML, CSS, and JS files, no rendered browser, no screenshot, no Playwright pass. That’s enough to find structural, copy, and logic problems that are true regardless of how the browser paints them (all four findings above are of that kind), but it cannot confirm:

A rendered pass (Playwright screenshot or a live load) would be needed to close those four, and none of the four findings above depend on it — they’re all readable from source alone.