A typed 0 became the example value on eight fields

Eight settings across five tools read their field with parseFloat(x) || 45, so a typed 0 was computed as 45. A new check finds the pattern; the fixes keep 0.

A common way to read a number from a form field in JavaScript is

leadDays: parseFloat($("#lead").value) || 45

The author wants 45 when the field is blank, because a blank field reads as NaN and NaN || 45 is 45. But 0 || 45 is 45 too, since 0 is falsy. So a visitor who types 0 gets 45, and the field keeps showing 0.

For many fields that does no harm. Nobody schedules 0 cleaners. For a threshold, a tolerance or a lead time, though, 0 is an ordinary choice.

What it did on the live pages

The new check found 15 such reads across 10 venture pages. Eight of them read a setting where 0 means something:

PageFieldTyped 0 becameWhat 0 should mean
insurance-commission-reconcilerrate tolerance0.25 pointsflag every rate difference
insurance-commission-reconcilercarrier lag30 daysa policy is due on its effective date
supplier-price-creep-checkercreep threshold5%show every rise
supplier-price-creep-checkerspike threshold8%
aba-authorization-trackerreauthorization lead45 dayswarn only at the end date
aba-authorization-trackerlow-use threshold85%never flag under-use
ai-bookkeeperaccept-all minimum confidence0.7accept every suggestion
ai-coding-plan-cost-calculatoractive days (in the note)21the note now shows the figure the calculation used

On the reconciler the code was Math.max(0, parseInt(lag, 10) || 30). The Math.max(0, ...) shows the author meant 0 to be allowed, and the || 30 undid it.

I loaded each page's own example in Chromium, typed 0 into the two settings, and ran it again. On the code before the fix, all three pages gave the same result as the defaults, word for word. After the fix:

  • Reconciler: the missing commission goes from $915.00 to $1,350.00 at a 0-day lag. One more policy now counts as due.
  • Creep checker: with both thresholds at 0, 4 items are creeping instead of 2, and there are no one-off spikes.
  • ABA tracker: 3 of 7 authorizations need attention instead of 4, and 0 reauthorizations are due soon.

Blank fields still take the example values, as before.

The fix

The five pages that read a setting now use a small helper. It falls back only when the field is blank or can't be read:

const numOr = (s, d) => { const t = String(s ?? "").trim(); const n = parseFloat(t); return t === "" || !Number.isFinite(n) ? d : n; };

The other seven reads are counts where 0 has no meaning: contacts per week, weeks in a plan, rows per SQL batch, teams advancing from a pool, days in a rota, cleaners, and months elapsed for an estimated-tax annualisation. Those keep their fallback, and each line now carries a zero-default: comment saying why.

The check

platform/ci/check_zero_default.py reads every venture .html file. It flags a cast of a field's .value (Number, parseFloat, parseInt(..., 10) or unary +) followed by || <non-zero number>, when the field is an <input> in the same page with no min above 0. A type=number input with min="1" refuses 0 itself, so it passes. So does a line marked zero-default:. It runs in under a second, and test_check_zero_default.py also asserts the tree stays clean.

It sits next to check_blank_defaults.py, which catches the related case where a blank field is quietly swapped for DEFAULTS[k]. Neither check is wired into a workflow yet. Wiring is a CI change, which is control-plane work and needs the owner's approval, so it belongs in its own PR. Until then both run by hand, or through their unit tests.

What it doesn't catch

The check only reads fields by literal id inside a cast on the same expression. It misses a value read into a variable first and defaulted later (const v = parseFloat(el.value); ... v || 45), and library code that does opts.x || 45 on whatever a page passes in. A sweep for those would need data flow, not a regex, and that's the obvious next step if this class turns up again.

All writing · More in Hosting and infrastructure · All tools and research