RailScanPro's AI learns from corrections. When you fix a misidentified manufacturer, road name, or model, that correction becomes part of improving the recognition model for future photos.
How to Correct a Misidentification
When the AI pre-fills your item form and something is wrong:
- Simply edit the incorrect field directly on the form before saving
- After you correct and save, a small Correction Logged message confirms the AI has noted the change
- That's it — no extra step required
Your correction is recorded alongside the photo and used in future model training rounds.
Correcting After Saving
If you already saved an item and realize the AI got it wrong:
- Open the item and click Edit
- Make your corrections
- In the Photos tab, find the photo that was scanned and click Report AI Error — this flags that specific image + field combination for higher-weight correction
What Corrections Fix
Corrections are most powerful for:
- Manufacturer names — "Athearn" vs. "Athearn Genesis" vs. "Athearn Blue Box" are distinct product lines
- Road name variants — "Penn Central" vs. "Pennsylvania Railroad" vs. "PRR" (all mean the same prototype, but different liveries)
- Model number precision — distinguishing an SD40 from an SD40-2 (different blade count, different nose details)
- Era — when the AI guesses Transition-era for a clearly Modern model
Feedback Ratings
On each saved item, you'll see AI Feedback buttons at the bottom of the detail page:
- Thumbs up — the AI got it right (helps reinforce correct patterns)
- Thumbs down — something was wrong (opens a quick correction form)
Even giving thumbs-up feedback for correct identifications helps train the model — it confirms the patterns the AI used are reliable.
Community Accuracy
Recognition improvements from your corrections benefit all users photographing similar equipment. The AI gets retrained on a rolling basis using accumulated corrections from across the platform.
Your corrections are anonymized before being included in training — only the photo metadata (manufacturer, model, road) is used, not anything personally identifying.
When to Give Up and Enter Manually
For some items, AI recognition may never be confident:
- Vintage tinplate with no prototype markings
- Hand-built or kit-bashed models
- Very small Z-scale with minimal visible detail
Manual entry is perfectly fine — not every item needs to go through AI scan.
Next Steps
- Photo Tips — better photos lead to better first-pass accuracy
- AI Not Recognizing Items — troubleshoot persistent recognition failures
- How AI Recognition Works — understand what the AI is doing