Large assemblies, skipped files, and whether the AI invents things

I uploaded 104 STEP files and only 12 were analysed

That is the importer working as intended, and it is worth explaining because the numbers look alarming.

A real assembly contains hundreds of individual parts, most of which are fasteners, washers, standard components, superseded revisions or exact duplicates. Opening and parsing every one of them would take minutes of processing to discover that 78 of them are M6 bolts, and it would not improve your case study by a word.

So every file is **indexed** — recorded, classified, named, sized, hashed, and shown to you — and a smaller number are **analysed in detail**. The selection is made by deterministic rules, not by guesswork: the likely top-level assembly, the latest revision in each chain, unique components, the design report, test data, and important images rank highest. Files in folders named "old", "archive" or "backup" rank low, as do superseded revisions and exact duplicates.

Every indexed file is still visible on the review screen and can still be kept with your project. Indexing is not exclusion.

Some of my files were skipped

Skipped means operating-system or temporary files — .DS_Store, Thumbs.db, ~$ lock files and the like. They are listed with the reason so you can see exactly what was left out.

Nothing you uploaded is deleted from your own computer, and nothing was published. If a file you care about was skipped or marked as needing attention, it is named on the review screen and you can still attach it to the project directly.

Does the AI invent information?

It is prevented from doing so by the way the draft is checked, not only by how the model is asked.

Every narrative section must cite the source files it was drawn from. When the draft comes back, each citation is checked against the files that actually exist in your import. A section whose citations do not resolve is emptied before you ever see it, and a question is put in its place.

That is why an outcome section is often blank on the first build. Your files usually do not record that the design saved 18% of the weight, or that you personally did the FEA — so the importer will not say either. It asks you instead.

The importer will state what it measured. "The assembly declares 61 components" and "7 likely revisions of the motor mount were detected" are facts read out of your files, and both are shown to you with the file they came from. What it will not do is explain why the mount was revised seven times, because nothing in the files says.

You should still read the draft before publishing it. It is your name on the page, it is created as a private draft precisely so you can, and it never becomes public on its own.

How big an import can be

Essential

200 files indexed · 8 analysed in detail · 250 MB total

Professional

800 files indexed · 16 analysed in detail · 1 GB total

Premium

2,000 files indexed · 28 analysed in detail · 3 GB total

90-Day Job Search Pass

800 files indexed · 16 analysed in detail · 1 GB total

File categories

The review screen groups your files under these headings: CAD files, 3D meshes, Drawings, Documents, Spreadsheets, Data and simulation files, Images, Archives, System files, Other files.