Fictional: a made-up site for Fruitful’s example. Nothing here is real.

That Orange Sitelinks for curious people
Show TOS: I photographed 14 months of grocery receipts to build my own inflation index
233 points by unitprice 5 hours ago | 8 comments
Since August last year I've photographed every grocery receipt and run it through OCR on my laptop. A small Python script matches the line items to a list of about 60 things I buy regularly (oats, eggs, one particular loaf, tinned tomatoes, coffee) and works out what that same basket would have cost each month.

My basket is up 9.4% over 14 months. The official figure for the same period was lower, which surprised me until I looked at what drove mine: coffee and olive oil did most of the damage, and I drink a lot of one and cook with a lot of the other.

OCR turned out to be the easy part. Matching is where the time went. The same product appears under four different abbreviations depending on which till printed it, and pack sizes change quietly, so everything is tracked by unit price rather than item price.

No link yet. It's about 400 lines of Python and a SQLite file, and I want to strip out my address and loyalty card number before I put it anywhere. Happy to answer questions about the matching.

Nice project. One thing you'll run into if you keep going: a fixed basket overstates how much your costs rose, because in real life you substitute. When coffee got expensive, did you keep buying the same coffee? Official indices try to account for that, with mixed success.
I didn't, and that's a fair point. I switched brands in March. The index keeps the old one for continuity, so it measures what my habits from last August would cost now, which is a slightly different question from what I'm actually spending. I track both, and the second number is lower.
That's the right way to do it. Most people who try this only track one, then argue with the official figure about the wrong thing.
How are you handling shrinkflation? Same price for a smaller pack is invisible on a receipt.
Badly, at first. Receipts rarely show the pack size, so I keep a small table of pack sizes per product and update it when I notice a change. I've caught three that way: crisps, a chocolate bar and, painfully, the coffee.
I tried something similar with a spreadsheet and gave up after two months, because typing in receipts is miserable. OCR is what makes this sustainable. What are you using?
Tesseract, with a preprocessing step that straightens and thresholds the photo. Thermal receipts fade fast, so I photograph them the same day or the numbers turn into guesswork.
Please publish the matching table, even if the code never comes. A shared map from till abbreviations to actual products would save a lot of people doing exactly this a lot of evenings.