How every engine is built
Every one of the 80 live engines on this platform follows the same process before it ships: a sourced formula, script-generated test vectors independently cross-checked by hand, and a stress-case vector chosen specifically to catch the kind of mistake a naive version of the same formula would make.
Sourced, not invented
Every formula traces to a named textbook, standard, or established methodology, cited on the engine's own page: engineering economy (Blank and Tarquin) for capital screening, Theory of Constraints (Goldratt and Cox) for bottleneck and constraint logic, inventory management (Silver, Pyke, and Peterson) for safety stock, and similarly sourced methods for every other family of engine. Nothing is an invented heuristic presented as established practice.
Test vectors, not spot checks
Each engine's test vectors are computed programmatically, by running a small script against the exact calculation formula, and at least one or two are independently cross-checked by hand as a sanity check before the engine ships. Every engine also includes at least one deliberate stress vector: a case specifically constructed so that a naive or oversimplified version of the same formula would give the wrong answer, proving the real logic (not just favorable inputs) governs the result.
Two commands, one gate
An engine's full automated test suite and a clean production build are the authoritative correctness gate before anything ships. This keeps validation rigorous while keeping the process fast enough to build toward a large, connected system without cutting corners on any individual engine.
Assumptions and limitations are stated, not hidden
Every engine page states its assumptions and limitations directly in its methodology section, alongside the formula and its sources. If a rule of thumb exists alongside the rigorous calculation, it's shown only as a supplementary note, never as the primary answer, and only when the two can disagree in an instructive way.
Connected by a versioned Decision Graph
Engines aren't published as isolated calculators. A shared registry and Decision Graph define which engines connect, and how a result from one can carry into the next. Engine components never hard-code downstream links themselves, so the graph stays the single source of truth for how the system fits together. See how your data is handled for the local-first storage model behind the graph's "My Operation" carryover feature.