Looking for fai software that reads gd&t and applies ISO 2768 by feature size, not OCR that detects text? See how Mavlon reads six tolerance standards.
I watched a quality engineer balloon a four-sheet machined housing with a position callout that read true position 0.2 at maximum material condition, referenced to datums A, B, and C. The auto-ballooning tool he was using dropped a balloon on it, and into the characteristic field it wrote: 0.2. That was it. The frame symbol, the MMC modifier, the three datum letters, all of it gone. He spent the next forty minutes typing the rest of that one feature back in by hand, then did it again on the next position callout, and the one after that. The tool had detected the number. It had not read the control.
That is the gap nobody puts on a slide. Most first article inspection tools were built to find characters on a page and put a circle next to them. The drawing, the actual engineering intent encoded in the symbols and the title block and the notes, stays locked inside the picture. If you are searching for fai software that reads gd&t, you are not looking for a better circle-placer. You are looking for something that understands what the circle is pointing at. This guide is for the quality and estimating people who already know the difference, and are tired of doing the interpretation themselves.
This post goes deep on tolerance interpretation specifically: how a tool handles a full datum reference frame, how it applies AS9102 Form 3 for ISO 2768mk drawings, and how it expands a fit class into real limits. For the broader mechanics of how AI reads engineering drawings, there is a separate chapter. Here, we stay on the tolerances.
Why you are searching for fai software that reads gd&t
You have probably already tried the auto-ballooning route. You imported a PDF, watched the tool sprinkle balloons across the sheet, and then opened the export to find half the GD&T flattened into bare numbers. The datum frames came in wrong. The general tolerance note in the corner was ignored, so every untoleranced dimension landed in your report with no limits at all. The basic dimensions got treated like toleranced ones. You ended up auditing every characteristic the tool produced, which is most of the work you bought the tool to avoid.
There is a reason this keeps happening, and it is not that the vendors are careless. It is the underlying approach. Optical character recognition reads pixels and returns text. Text has no idea that the box around it is a feature control frame, that the letter to its right is a datum, or that a note three views away governs the limits on the dimension you just ballooned. The information a quality engineer needs lives in the relationships between marks on the page, and character detection throws those relationships out. So the human puts them back, one feature at a time.
The real question is not which OCR tool is best
Buyers frame this as a shootout: 1Factory versus High QA versus InspectionXpert, which auto-baller wins. That is the wrong axis. Look at how each one describes its own extraction engine, in its own marketing. 1Factory calls its auto-ballooning traditional OCR. High QA brands its version smart OCR. InspectionXpert, now part of Ideagen Quality Control, is OCR-based with manual point-and-click extraction. They are not hiding it. They are telling you, plainly, that the engine reads characters.
So the decision is not which OCR is best. The decision is whether to use OCR at all, or to use a tool that reads the drawing. That is a category split, not a feature difference. It is the same split the phone industry went through. The BlackBerry was a great BlackBerry. Then the category changed underneath it, and being the best at the old thing stopped mattering. Tolerance interpretation is having that moment now.
OCR versus AI, explained plainly
Here is the distinction with no jargon. OCR answers one question: what characters are in this region of the image? It returns 0.2, or A, or a position symbol it may or may not recognize. It does not know how those pieces relate. It cannot tell you that the 0.2 is a position tolerance, that the A is its primary datum, that the modifier means the tolerance grows as the feature departs from maximum material condition. To OCR, those are just shapes that happened to be near each other.
AI drawing intelligence answers a different question: what is this drawing telling an inspector to do? It reads the feature control frame as a single control with a characteristic type, a value, a modifier, and an ordered datum reference frame. It reads the title block to find which general tolerance standard governs the sheet. It reads the notes and understands that a note on sheet one can govern a tolerance on sheet three. It produces a characteristic, not a string. That is the whole difference, and it is the difference between a report you trust and a report you re-check.
This is why fai software that reads gd&t is a meaningfully different product from fai software that detects gd&t. Detection gives you the symbol as a picture or a guessed character. Reading gives you the structured control: type, value, modifiers, datums, in order, ready for the Form 3.
What it looks like on a real drawing
Abstract claims are cheap, so here is the concrete behavior on the kinds of callouts that actually slow a quality engineer down.
A general tolerance note that says ISO 2768-mK is not one tolerance. It is a table. The permitted deviation on a linear dimension depends on the nominal size: a dimension from 0.5 to 3 mm gets a tighter band than one from 30 to 120 mm, and a dimension over 1000 mm gets a wider one still. Reading the drawing means looking up the right row for each untoleranced dimension by its actual size and writing the correct plus-minus limits into that characteristic. Mavlon applies ISO 2768 linear, angular, and chamfer tolerances by feature size, per the class called out in that drawing's own title block. Detecting the note would just copy the text ISO 2768-mK into a field and leave the limits blank.
A fit callout like 9 G6 is a hole on a 9 mm nominal with an ISO 286 tolerance grade. On its own, G6 means nothing to an inspector standing at a CMM. It has to be expanded into real upper and lower limits in millimeters before anyone can measure to it. Mavlon expands fit classes into actual limits. If you want to see the underlying logic worked through, the post on ISO fit class G6 tolerance limits walks the full expansion.
A GD&T callout like position 0.2 at MMC to datums A, B, C is captured as a complete control. Characteristic type: position. Value: 0.2. Modifier: maximum material condition. Datum reference frame: A primary, B secondary, C tertiary, in that order. The order matters because it defines how the part is constrained for inspection, and a tool that drops the order has changed the requirement. Mavlon keeps the full frame.
A note carried across sheets is the one that catches everyone. The general note lives on sheet one. The feature it governs is dimensioned on sheet three. A quality engineer knows to carry that note forward. A character detector working sheet three in isolation has no way to know the note exists. Mavlon reads the drawing as one document and carries the note onto the tolerance it governs, wherever that tolerance sits.
What Mavlon is
Mavlon is AI drawing intelligence. You hand it a drawing the way you would hand it to a quality engineer, and it reads it. It auto-extracts the notes, auto-places the balloons, and fills out the AS9102 Form 3. It captures GD&T with full datum reference frames, applies ISO 2768 general tolerances by feature size, and expands fit classes like 9 G6 into real limits. It carries a note on one sheet onto the tolerance it governs on another. It reads scanned and photographed PDFs, and it handles European comma-decimal drawings without choking on the separator.
It reads six tolerance standards: ISO 2768 linear, angular, and chamfer; ASME Y14.5 GD&T; ASME decimal-place tolerance notes; ISO 286 fits; ISO 13920 for welded structures; and ISO 9013 for thermal cutting. It runs in any browser with no install. It reads a drawing in about two to three minutes at roughly ninety percent accuracy, and it flags the roughly ten percent it is unsure about so a human can confirm those specific characteristics instead of auditing the whole sheet. Two clicks. Those numbers are ours, measured on our side, and we present them as our claims.
Side by side
The honest comparison is not about who has more buttons. It is about whether the engine reads or detects, and what reaches your Form 3 because of it.
| Capability | Mavlon | Typical OCR-based FAI tool |
|---|---|---|
| Reads the drawing vs detects text | Reads and interprets | Detects characters |
| Auto-places balloons | Yes, two clicks | Yes, often with manual cleanup |
| Fills AS9102 Form 3 | Yes, with structured characteristics | Exports text fields for human completion |
| Applies ISO 2768 by feature size | Yes, correct row per nominal | Copies the note as text |
| Expands fit classes to real limits | Yes, ISO 286 limits in mm | Returns the class string |
| Full GD&T datum reference frame | Yes, ordered datums and modifiers | Often flattened to a number |
| Tolerance standards covered | Six | 1Factory publicly markets ISO 2768 coarse and medium |
| Deployment | Browser, no install | Desktop heritage or gated trial |
On that last standards row: Mavlon reads six tolerance standards. 1Factory publicly markets ISO 2768 coarse and medium for its general tolerance handling. That is what each vendor states, and the gap between six and what is publicly marketed is the point. We are not asserting what anyone else cannot do. We are stating what we read, and what they advertise.
Where the incumbents genuinely fit
This is the part most comparison posts skip, and skipping it is dishonest. The incumbents are real products that serious shops run for good reasons.
1Factory is a full quality management platform. SPC, gage management, PPAP, CAPA, the whole QMS spine. If you need that system of record across your plant, its ballooning being traditional OCR is a footnote in a much larger and genuinely capable suite. High QA is a modular quality suite with smart OCR ballooning, and shops that have standardized on its modules get a coherent toolset. InspectionXpert, now Ideagen Quality Control, has deep desktop FAI heritage; Ideagen also markets some machine learning to help populate measurements, so it is fairer to frame the difference as AI-native interpretation versus OCR-based extraction than to claim there is no AI in the building. DISCUS is the original desktop FAI tool, with its OCR add-on IDA, and its local ITAR-friendly install is exactly what some defense programs require, even though it is the weakest of the group on GD&T and notes.
Here is the honest summary: those platforms are a mile wide. Mavlon is a mile deep on one thing, reading the drawing and producing trustworthy tolerances. The two are not even in conflict. Mavlon reads the drawing, builds the Form 3 with correct GD&T frames and ISO 2768 limits, and you feed that Form 3 into whatever QMS you already run. You do not rip out 1Factory. You stop hand-typing datum frames into it.
What switching actually feels like
The experience difference shows up before you ever look at output. Most incumbents put friction in front of you on purpose. A sales call to see anything. Pricing gated behind contact us. Access negotiated up front in at least one case. A desktop install with point-and-click extraction in another. You are committing before you have seen a single one of your own drawings read.
Mavlon runs in a browser. You upload your drawing, you get balloons and a Form 3 in two to three minutes, and you judge it on your own part. It is self-serve, with nothing to install and no gate to clear before the tool proves itself. The ten percent it is unsure about is flagged, so your review is bounded and specific rather than a full re-audit. The contrast is not subtle. One path asks you to trust a demo. The other hands you the result on your drawing and lets the drawing decide.
Bring the drawing that took someone all day
If you are evaluating fai software that reads gd&t against the OCR-based incumbents, do not run the comparison on a clean sample part. Run it on the worst drawing you have. The four-sheet housing with the note on sheet one that governs sheet three. The position callout at MMC to three datums. The ISO 2768-mK block and the 9 G6 fit nobody wants to expand by hand. The European drawing with comma decimals. The scanned PDF that came in as a photograph.
Bring the drawing that took someone all day. Let Mavlon read it, and look at the datum frames, the ISO 2768 limits applied by size, and the fit class expanded into real numbers. Then decide whether you want a tool that detects your drawing, or one that reads it.
---
Have a drawing that ate someone's afternoon? Upload it and watch Mavlon read the GD&T and apply ISO 2768 in two clicks. Book a 30-minute demo.
FAQs: Reads GD&T + ISO 2768
Mavlon is the FAI software that reads GD&T automatically - capturing full feature control frames (geometric characteristic, tolerance zone, datum reference frame in order) rather than just detecting symbols. Per ASME Y14.5, the datum frame matters as much as the tolerance value.
Mavlon does. It reads the title block's general tolerance note (ISO 2768-m, ISO 2768-f, etc.), then walks every untoleranced dimension on the drawing and applies the correct band based on the feature size. 6mm gets one band, 120mm gets another.
Six: ISO 2768 linear, ISO 2768 angular, ISO 2768 chamfer; ASME Y14.5 GD&T; ASME decimal-place notes; ISO 286 fits; ISO 13920 for welded construction; and ISO 9013 for thermal cutting.
Yes. Fit-class shorthand like Ø9 G6 is expanded to real upper and lower limits - 9.005 to 9.014mm in this case - per ISO 286. The Form 3 gets the actual limits, not the shorthand.
OCR can usually detect the symbol and the value inside a feature control frame. AI reading captures the whole frame as one structured callout - symbol + tolerance zone + datum reference frame - because the datum frame is the grammar that makes the tolerance meaningful.
Upload your hardest drawing.
Watch it read.
Bring the seven-sheet one with GD&T across multiple datums, a buried tolerance note, fit classes, and a flag note or two. The drawing that took someone all day. We'll read it, balloon it, and fill the Form 3.
Upload a Drawing