1. A written brief before code
Each utility starts as a short problem brief: who does the job, what they already know, what they need back, which formula or data answers it, and what the utility must not claim. The brief names the sources that will be used and the cases that will be checked by hand. Where search data was used to choose the task, the brief says so — search demand is a reason to build, never evidence that a result is right.
2. Hand-worked test cases
Every engine ships with fixtures: inputs with expected results worked independently — by hand, from a published example, or from a separate reference computation — before the code existed. The test suite runs them on every change, together with browser tests that drive the real page. A utility whose fixtures fail cannot be built or deployed.
3. Primary sources, cited on the page
Where a result depends on data rather than a formula — a bag yield, a dressed lumber size, a sampling table, a density — the dataset carries the document it was read from, the edition or date, and the day it was accessed. The utility page shows that source, and the “Method & assumptions” section explains how the figure is used and what it does not cover. 12 of the 32 published utilities list their sources on the page; the rest rest on a stated formula alone (an area, a volume, a ratio) with no external dataset. Manufacturer documents are named but not linked where the manufacturer's site terms restrict linking.
4. No licensed table without a lawful review
Some engineering tables exist only inside paid standards. A utility that depends on such a table is not published until the table has been checked against a lawful copy of the standard or a public-domain equivalent. One utility was withdrawn under this rule and one was held until its table could be verified cell by cell against a public-domain edition. Reconstructing a table from its pattern is not verification — that mistake was made once, caught by this rule, and corrected before release.
5. Versioned datasets and a dated review
Datasets are versioned separately from the engine and the page. Each utility carries a “last reviewed” date — the last time its results, dataset, or method changed — not the date the site was rebuilt. When a source changes or a discrepancy is found, the dataset version is raised, the fixtures are re-sourced, and the change is recorded.
6. Independent review and an explicit release decision
Work is reviewed by a separate reviewer before release, and publication is a recorded decision, not a side effect of finishing the code. Typical-value datasets that a practitioner has not yet reviewed are labelled as such on the page and can be overridden. Safety-relevant figures carry their uncertainty in words (for example, a ±25 % scatter on bolt torque), and no page states compliance with a building code, a standard, or a regulation.
What we do not claim
No utility replaces the source you are accountable to — the standard, the data sheet, the code, or the engineer of record. Waste and allowance figures are yours; product figures are the manufacturer's; prices are whatever you enter. If you find a result you can show to be wrong, write to feedback@arzentiq.com with the page and the numbers; corrections are versioned and dated.