Limits, assumptions, and validation

EngDatabase AI is built to return trustworthy engineering numbers, but trust has limits. This page explains what the product does, what it does not do, and how the results should be treated in a real engineering workflow.

Results are traceable

Every numerical result comes from either a reference table or a calculator. The assistant tells you which one. For lookups it names the table and the row. For calculator runs it lists the inputs and the method used. You can reproduce any answer by running the same tool yourself from the database page or the calculators page.

Interpolation is labeled

When a value is computed between two bracket rows of a table, the assistant says so and reports the rows it used. This is the standard way to read a reference table that has discrete entries, and it is also an approximation. The assistant does not hide that fact. You decide whether the brackets are close enough for the purpose.

The assistant does not fabricate reference values

If a value is not in the reference library, the assistant will say so rather than invent a number. This is intentional. A fabricating assistant is worse than no assistant, because it makes you cross check every result.

If a lookup fails, the assistant will point you at the closest table it could find and ask whether you want to use it. You stay in control of the source.

Reasonable engineering assumptions

Fabricating a reference value is not the same as making a reasonable engineering assumption. When you leave a calculation input out, the assistant will pick a sensible default and say so. If you ask it to size a pump without naming a fluid, it assumes water at standard conditions. If you ask for a beam check without naming a steel grade, it assumes a common grade. If you ask for a pressure drop without naming a roughness, it picks a standard value for the named pipe material.

Every assumption the assistant uses is reported back with the result. If the default does not match your situation, you can correct it in your next message and the assistant will rerun the calculation with the new input. Picking a default and reporting it is normal engineering practice. Fabricating a table value is not. The difference matters and the assistant treats them differently.

Calculators follow published methods

Each calculator follows a stated code, standard, or textbook method. The method is listed with the result. When a calculator references a particular code revision, that revision is named. When the method requires assumptions (a load factor, a safety factor, a material grade, an efficiency), those assumptions are either supplied by you or chosen to match common engineering practice, and the assistant reports them back so you can see what was used.

Units

The assistant handles mixed unit input. It reports results in the units you asked for, or in the native units of the source table if you did not specify. Unit mistakes are a common source of engineering errors, so the assistant always states the units it used. If it detects an ambiguous unit in your request, it asks.

Versioning of reference data

The reference library is updated as source standards update. Older code values are kept as historical tables where they matter. The assistant uses the current library by default and will tell you which version of a standard a given table is sourced from when that matters to the answer.

What you still have to do

Every result returned by the assistant or a calculator is an engineering calculation, not engineering judgment. A licensed engineer is still responsible for selecting the right method, accepting the assumptions, reviewing the result, and signing off on the design. The product is a tool to speed up the arithmetic and lookup side of engineering work. It does not replace the review step, and it is not a substitute for professional judgment.

What the product does not do

• It does not perform finite element analysis or computational fluid dynamics simulations.
• It does not produce stamped drawings or construction documents.
• It does not sign or seal engineering work.
• It does not make code selection choices for you. When a code revision matters, you name the revision.
• It does not train on your conversations. AI messages are not stored after the response finishes.

For the full data handling position, see the privacy policy.