How the numbers are verified

The Learn section contains twenty fully worked examples — ten beam, ten truss. Each one is solved twice: once by hand using classical methods (statics, the method of joints, the three-moment equation, standard deflection formulae), and once by the solver that powers the calculators.

An automated harness re-solves all twenty through the real engine and compares every reaction, member force and diagram extreme against the published hand solution. A disagreement beyond round-off is treated as a stop-the-line defect, not a tolerance to widen.

Precisely: that harness is part of the release gate that is run before work is accepted — it is not wired into the JavaScript bundler step, so "the build fails" would overstate it. What is true is that all twenty currently agree, and that any disagreement stops the work.

Beams and trusses are solved by a direct-stiffness engine written in Rust and compiled to WebAssembly. It runs entirely in your browser: no model you enter is sent to a server.

Beyond the twenty examples, the engine is held to global statics — the formula-independent ground truth. Equilibrium of the whole structure is checked independently of the method that produced the answer, so an error in the solver cannot hide behind an error in the formula being used to check it. Rotational-invariance tests confirm that rotating an entire model rotates its reactions and displacements while leaving member forces unchanged.

The engine has been through an independent adversarial audit specifically looking for plausible-but-wrong outputs. Three real defects were found and fixed that way: inclined and vertical roller supports, duplicate loads on one node being overwritten instead of summed, and a sign error on loads applied directly at a support. Each was characterised against statics and fixed with a failing-test-first change.

The library holds 2,686 sections across six national standards. The editions used are:

• AS 4100 (Australia) — OneSteel Hot Rolled & Structural Steel Products, 7th edition
• AISC (USA) — AISC Steel Construction Manual shapes database (published Imperial columns kept alongside their exact SI conversions)
• EN 10365 (Europe) — EN 10365 published IPE / HEA / HEB / HEM / UPN tables
• BS 4 (UK) — BS 4-1:2005 property tables
• JIS G3192 (Japan) — JIS G3192
• IS 808 (India) — IS 808:1989 including Amendment 1

Every row is checked by automated invariants that do not depend on the source: area against mass and density, the major-axis second moment against the minor-axis one, radius of gyration against the named depth, and per-family samples pinned to published values. Thirty such tests run across the whole library.

Transcription errors have been found and corrected, and the corrections are recorded. The BS 4 tables had the major-axis second moment stored a factor of ten too small, together with a corrupt minor-axis column; both were retranscribed against British Steel's own datasheets. The IS 808 channel data had been taken from the wrong table and understated area and both second moments by up to about half. These were found by the invariants above, not by inspection.

• These tools perform analysis, not design. They compute reactions, member forces, shear, moment, deflection and section properties. They do not perform member capacity checks to AS 4100, AISC 360, EN 1993 or any other code, do not generate load combinations, and make no compliance statement. Sizing a member remains the engineer's judgement.
• Some published section data is still flagged as unverified. The minor-axis second moment of all 89 AS 4100 rectangular hollow sections is physically impossible as published — several rows state a minor-axis value greater than the major-axis one on a section deeper than it is wide. The same test passes 89 of 89 AS 4100 square hollow sections and 301 of 301 AISC rectangular HSS, so the problem is specific rather than a fault in the check. Those rows are recorded as open findings rather than silently corrected to whatever would make the test pass.
• Torsion and warping constants are not computed. The section tools report area, centroid, second moments, principal axes, radii of gyration and section moduli. The torsion constant J and warping constant C_w are deliberately absent rather than approximated.
• Analysis is linear-elastic and, for trusses, small-displacement. There is no second-order (P-Δ) analysis, no material nonlinearity and no buckling check.

Every worked example shows its full arithmetic, so any result on this site can be checked by hand. The same solver is available over HTTP through the public API, which returns the engine version with every solve — so a result can be tied to the exact build that produced it.

If you find a number you believe is wrong, please tell us. Errors found in published data have been corrected before and recorded; that is the process working, not an embarrassment.