Into the Scrape-Verse · Bright Data × WeMakeDevs
Bright Data repairs a broken scraper. Something has to notice it broke.
PriceGuard reads the rows a run came back with, not the job status, because a scraper that broke overnight still reports success and still returns a normal row count.
Open the boardHow it was measured
21 laptop SKUs63 tracked prices396 archived pages1117 tests
Why nothing catches it
The same collector, 10 minutes apart. Every health signal reads the same.
Apple MacBook Air 13.6" (M5, 16GB, 512GB, Sky Blue)
19:00 UTC
- Job status
- ready
- Rows returned
- 19 of 19
- Prices missing
- 0
₹1,39,39,900
Wrong by 100×
19:10 UTC · after one repair
- Job status
- ready
- Rows returned
- 19 of 19
- Prices missing
- 0
₹1,39,399
Correct
Read from examples/reliance-digital-capture*.json
| Laptop | Yours | Croma | Reliance Digital | Vijay Sales | Cheapest · gap to dearest |
|---|---|---|---|---|---|
| Apple2 SKUs | |||||
| M5 · 16GB · 1TB · MidnightApple MacBook Air 13.6" | ₹1,66,070 | ₹1,72,99015:45 today ⌄
| ₹1,72,87915:47 today ⌄out of stock
| ₹1,85,90015:48 today ⌄
| Reliance Digital−₹13,021 · 8% |
| M5 · 16GB · 1TB · Space BlackApple MacBook Pro 14.2" | ₹2,14,180 | ₹2,22,99015:45 today ⌄
| ₹2,23,09915:47 today ⌄
| ₹2,23,50015:48 today ⌄
| Croma−₹510 · 0% |
| HP1 SKU | |||||
| Core i5-13420H · 16GB · 512GB · RTX 3050HP Victus 15-fa2192tx | ₹73,430 | ₹78,99415:45 today ⌄
| ₹71,590▼3%15:47 today ⌄out of stock
| ₹75,49915:48 today ⌄
| Reliance Digital−₹7,404 · 10% |
| Lenovo1 SKU | |||||
| Core Ultra 5 125H · 16GB · 512GB · OLEDLenovo Yoga 7 2-in-1 14IML9 | ₹96,000 | ₹98,99015:45 today ⌄
| Reliance Digital does not stock Lenovo Yoga 7 2-in-1 14IML9 | ₹99,99915:48 today ⌄
| Croma−₹1,009 · 1% |
Showing the captures committed in this repository
Every price on this page carries the capture that confirmed it.
Built from the captures committed with this page (fixtures/live), so what you read is exactly what shipped.
A refresh re-runs the projection over a store of captures. It never runs the scraper, so it spends no Bright Data credits.
396 real archived pages, from a retailer's platform migration nobody here staged
10 of 10
caught when someone acts on the alert
6 of 10
caught if nobody acts
0 of 9
false alarms on ordinary months
The check every scraper already has caught 0 of 10
All came back marked ready, with normal row counts. How this was measured, and its 7 caveats.
Measured on fixtures/heldout/WAYBACK-CROMA
Three commands, on a fresh clone, no API key
npm test1117 tests. No network, no credits.
npm run heldoutRe-runs the number above on the archived pages behind it.
npm run watch:once -- --break rows-halvedBreaks a scraper on purpose. One check notices; the usual one does not.