How accurate is Figventory's LEGO® minifigure scanner? First identification confirmed 99.3% of the time on 2,786 reviewed scans
≈6 min read · Updated November 2026
The first identification is confirmed 99.3% of the time, and the method that says so is on this page.
On 2,786 reviewed LEGO® minifigure scans between October 15 and November 4, 2026, users confirmed Figventory’s first identification 99.3% of the time, and the figure they kept was one of its top five suggestions 99.6% of the time.
As of November 2026, Figventory is the only LEGO® scanner that publishes how often its first identification is confirmed, with the method behind it. My guide to LEGO® scanner accuracy found no competitor that publishes a method, and the vendor numbers out there contradict each other. This one comes with the scans, the window, the script and the counts.
What the number means
Here is what it means when you are holding a figure. You scan it. Figventory gives you an identification and shows a match confidence badge. You look at the identification (name/part ID), and you either keep it or change it based on whether it is correct. On 2,786 scans where someone looked and decided, the identification Figventory gave first was the one they kept 99.3% of the time. Put plainly: by the user's own review, Figventory got it on the first try 99.3% of the time. When it was not, the correct figure was almost always a tap away in the other matches: 99.6% of the time it was in Figventory's top 5 matches. Ten scans out of 2,786 ended with a figure Figventory never suggested at all.
“Confirmed” on this page means exactly that: a user reviewed the scan and kept the first identification. It is the user's confirmation, not AI Smart Verify's, which is a different check and gets its own section below.
These are real scans by real users of the app between October 15 and November 4, 2026, read once from the production database on November 4. The tables, the plain-English method and everything that could make the number look better or worse than it is follow. A photo set of known lookalike figures, labelled in advance, comes later as its own addendum.
The numbers
All three tables have the same columns. Scans is how many scans are in the row. First identification confirmed is how many of them kept the identification Figventory gave first. Correct figure in top 5 matches is how many ended on a figure that was among Figventory's first five suggestions: the first identification plus the next four other matches the app shows. It includes the first. Correct figure never suggested is how many ended on a figure Figventory did not offer at all. “Correct” here means the figure the user settled on after reviewing the scan. The two percentages are the first two counts divided by the scans.
By match confidence score. Every identification in the app carries a match confidence badge. High is a score of 0.80 or more, mid is 0.55 up to 0.80, low is under 0.55. Scans with a high badge were confirmed 99.6% of the time and scans with a mid badge 98.9% of the time, so the badge means what it says: agreement drops a little as the score drops. The low and unknown rows are 25 scans and 1 scan, too few to quote as rates.
| Match confidence score | Scans (n) | First identification confirmed | Correct figure in top 5 matches | Correct figure never suggested | Confirmed % | Top 5 % |
|---|---|---|---|---|---|---|
| Low (below 0.55) | 25 | 24 | 24 | 1 | 96.0 | 96.0 |
| Mid (0.55 to 0.80) | 1,025 | 1,014 | 1,020 | 5 | 98.9 | 99.5 |
| High (0.80 and above) | 1,735 | 1,728 | 1,732 | 3 | 99.6 | 99.8 |
| Unknown (no score) | 1 | 0 | 0 | 1 | 0.0 | 0.0 |
By scan mode. A single scan is one photo of one figure. A PowerScan crop is one figure cut out of a tray photo, one row per figure. Both modes land in the same place.
| Mode | Scans (n) | First identification confirmed | Correct figure in top 5 matches | Correct figure never suggested | Confirmed % | Top 5 % |
|---|---|---|---|---|---|---|
| PowerScan crops (multi) | 2,483 | 2,466 | 2,474 | 9 | 99.3 | 99.6 |
| Single photo scans | 303 | 300 | 302 | 1 | 99.0 | 99.7 |
Every scan, reviewed or not. This line adds the scans nobody has reviewed yet. An unreviewed scan only says nobody changed it, so this is printed for completeness and it is not the headline. The headline is the 2,786 reviewed scans.
| Scans | Scans (n) | First identification confirmed | Correct figure in top 5 matches | Correct figure never suggested | Confirmed % | Top 5 % |
|---|---|---|---|---|---|---|
| Every scan in the window | 7,987 | 7,943 | 7,974 | 13 | 99.4 | 99.8 |
No scan in any table ended with the correct figure suggested but past fifth place: when it was suggested at all, it was in the top 5.
What is in the count
- LEGO® minifigure scans from both scan modes, saved in the app. Parts are not in this read.
- Reviewed by a user. Someone marked the figure reviewed in the app. That is the 2,786.
- Scanned between October 15 and November 4, 2026, read once on November 4.
- Not my own scans. Internal accounts, mine included, are left out.
- Not the free identifier on this website. It keeps nothing after it answers, so there is nothing to read.
- Not the browser extension or the lot scanner. Those 108 lookups keep no suggestions.
AI Smart Verify is a different check
After a figure is identified, AI Smart Verify looks at it and says whether it is complete: complete, partial or mixed. It does not pick which figure it is. In this window 2,035 scans were verified, and the figure's identity changed after verify on only 3 of them. The verdicts came out complete 1,293, mixed 257, partial 485. None of that moves the numbers above; those belong to Figventory's own identifications.
What could make the number look better or worse than it is
I counted what I could. Three things lean the number up, and two I cannot call either way.
- Fixing a scan clears its reviewed mark. When someone changes a figure's identity, the app un-reviews it, and it stays out of the count until someone reviews it again. So scans where the first identification stood are more likely to stay in. Leans up. In this window 24 corrected scans sit outside the count for this reason.
- Reprocessing overwrites the suggestions. Re-running a scan replaces the stored suggestions whether or not the new result is kept, and nothing records that it happened. What I read is what is stored now. Leans up.
- Before October 14, a correction threw the suggestions away. A fix on October 14, 2026 stopped that. Reading earlier days would silently drop the exact scans where the user disagreed, which is why the window starts on October 15. In this window, 0 scans lost their suggestions.
- Removed and deleted figures leave the count. 741 minifigure removals and 532 bulk or bin deletes happened in the window, and some of them took reviewed scans out of the read. I can count the events, not what was in them, so I do not know which way this leans.
- People who review are their own crowd. Careful reviewers may catch more lookalikes; quick reviewers may confirm more. I do not know which, so this one is unknown too.
The 98%+ line my home page used to carry
The pricing section of my home page said “98%+ of identifications never corrected” until November 2026, when the measured line above replaced it. That was a different measure. Of all 7,987 identifications in this window, 44 were changed by a user afterwards. That line counted corrections; this page counts confirmations on scans someone actually reviewed. Both are real and measured.
How I ran it
One script, run by me inside the production container on November 4, 2026, as a single read-only pass. It writes nothing. The script is scanner_accuracy_read, kept with its raw output in Figventory's private repository.
It reads four things: each identification with the suggestions the scanner stored for it, the reviewed mark the user set, the latest AI Smart Verify verdict, and the removal and deletion events in the window. Then it compares the figure the user kept with the suggestions, in order.
The next read is monthly, same script, same method, published here. The next window closes in December 2026.
FAQ
How accurate is Figventory’s LEGO® minifigure scanner?
On 2,786 reviewed LEGO® minifigure scans between October 15 and November 4, 2026, users confirmed Figventory’s first identification 99.3% of the time, and the figure they kept was one of its top five suggestions 99.6% of the time.
What does “confirmed” mean here?
A user looked at the scan in the app, marked the figure reviewed, and kept the figure Figventory identified first. It is the user’s confirmation after looking, not an AI’s. In plain terms, Figventory got it on the first try 99.3% of the time, by the user’s own review.
Is this an accuracy benchmark?
Not in the lab sense. Nobody labelled these figures in advance; the measure is whether users kept the identification Figventory gave first after looking at it, on real scans. It is published with the method, which no other LEGO® scanner does. A photo set of known lookalike figures, labelled in advance, is the next piece.
Why only reviewed scans?
A reviewed scan is one a user looked at and marked reviewed in the app, so the figure they kept is a figure they checked. An unreviewed scan only tells you nobody changed it yet. The every-scan line (7,987 scans, reviewed and unreviewed) is printed beside the headline, never instead of it.
Does AI Smart Verify change these numbers?
No. AI Smart Verify checks whether the figure you kept is complete; it does not pick the figure. Of 2,035 verified scans, the identity changed after verify on 3. The rates on this page belong to Figventory’s own identifications.
When is the next read?
Monthly, with the same method and the same script, published on this page. The next window closes in December 2026.
LEGO® is a trademark of the LEGO Group, which does not sponsor or endorse Figventory.
