How the AI uses structured engineering tables

The reference library contains 375 engineering tables. Each table has a strict structure: named columns, known units, a defined way to look up rows, and a fixed set of entries. The AI assistant uses that structure to return exact reference values instead of generating them from model memory.

A simple example

Ask the assistant: “What is the wall thickness of a 4 inch schedule 80 pipe?”

The assistant does not answer from memory. It:

1. Identifies the table that holds standard pipe dimensions.
2. Looks up the row for 4 inch nominal size at schedule 80.
3. Returns the wall thickness, in the units you asked for. If you then say “and in millimeters” it reconverts the same row.

The answer is the published value from the pipe dimension table. There is no rounding, no rewording, and no chance of a fabricated number.

Why structured tables beat plain text

A plain text document can contain the same information, but the AI has to parse it, interpret it, and hope it read the right column. A structured table removes all of that. The column set is known, the units are known, and the value is looked up rather than read.

This is the difference between asking an assistant to “find this value in a PDF” and asking it to “fetch row 5 from table X”. The first is a reading exercise with room for mistakes. The second is a deterministic lookup.

When there is no exact match

Engineering data often lives on a grid of bracket values. A stress table may be indexed by temperature in 50 degree steps. A friction factor table may be indexed by Reynolds number in decade steps.

When you ask for a value between two bracket rows, the system interpolates linearly between the nearest entries. The assistant tells you when this has happened and which rows were used, so you can see whether the brackets are close enough for the purpose.

Linear interpolation is the standard way to read an engineering reference table. It is not a shortcut; it is how the tables are designed to be used.

When the value is not in the library

The assistant reports that the value is not available. It does not fabricate a reference value. This is a deliberate design choice: an engineering assistant that fabricates lookups is not useful, because the whole point is to return numbers you can trust without cross checking every one of them.

This is distinct from the reasonable engineering assumptions the assistant makes for unspecified calculation inputs. Picking a standard fluid, a common material grade, or a typical efficiency when you did not name one is normal engineering practice. Those choices are always reported back. Fabricating a reference table row is not.

If a lookup fails, the assistant will suggest the closest table it could find and ask whether you want to look there instead. You stay in control of which source gets used.

Browsing the library

You do not have to go through the assistant. You can browse the reference library directly from the database page and run the same lookups the assistant would. The AI layer is an accelerator, not a wall in front of the data.