There's a quiet economy of knowledge in RailScanPro, and you're a contributor to it. Every time you fix a wrong manufacturer, correct a road name, or identify an obscure shortline livery that the AI guessed wrong — that correction flows back into the model and makes it more accurate for everyone who comes after you.
How Corrections Work
When the AI pre-fills your item form and gets something wrong, you don't need to do anything special:
- Edit the wrong field before you click Save
- Save the item
That's it. RailScanPro automatically logs the delta between what the AI suggested and what you actually saved. The photo, the wrong answer, and the right answer form a training triple that's used in the next model update.
The Feedback Buttons
For items already saved, a pair of feedback buttons appears at the bottom of any item detail page:
Thumbs up — you're telling the AI "yes, you got this one right." This is just as valuable as a correction — it reinforces which visual patterns correctly identify this manufacturer, model, and road.
Thumbs down — opens a quick correction form where you specify which field was wrong and what the correct value is.
Getting in the habit of rating AI identifications — even when they're right — noticeably improves accuracy over time for the equipment you model most.
High-Value Corrections
Some corrections have more impact than others:
Shortline and regional railroads — The AI has deep knowledge of Class I carriers but much thinner training on shortlines, regionals, and historical predecessors. If you model the Wheeling & Lake Erie, the Central Vermont, or the Colorado & Southern, your corrections for that equipment are extremely valuable because few other users are providing data in those categories.
Transition-era equipment — Locomotives from the 1950s–1970s, particularly early geeps and Alcos, have enormous variety in nose styles, cab configurations, and paint schemes. Precise corrections here help thousands of modelers of the transition era.
Custom-painted and re-lettered models — If you've applied a prototype paint scheme to a model that was originally produced in a different road, the AI may identify the wrong road. Correcting these teaches the AI to look past the original manufacturer's paint.
The Recognition Report
Go to Asset Management → AI Recognition Report to see:
- Your correction history
- Which items have low-confidence AI assessments (candidates for re-photographing with better light)
- A summary of how much your corrections have contributed to platform accuracy
Privacy and Anonymization
Your corrections are fully anonymized before being included in any training data. The training triple contains: the photo (stripped of EXIF metadata), the wrong AI answer, and the right answer. Your account identity is never attached.
Community Accuracy Over Time
When RailScanPro launched, AI recognition accuracy for HO steam locomotives was approximately 74%. After eighteen months of user corrections, it's above 91% for the same equipment categories. That improvement came entirely from collectors like you correcting misidentifications in their own collections.
Next Steps
- How AI Recognition Works — understand the technology
- Taking Photos for Best Results — better photos mean fewer corrections needed
- Start Contributing — join the community of collectors improving AI recognition