Method
How it works
A language model does the searching and reading. It never gets the last word on a number. Every figure that reaches a recommendation is re-validated and recomputed in code.
- 01
We pin down the exact variant
iPhone 17 Pro is not a product. 256 GB in Deep Blue, sold in India, is. Storage, memory, colour and part number are all identified before anything is priced, because comparing two variants produces a confident wrong answer.
- 02
We read the prices off real pages
Amazon, Flipkart, Croma, Reliance Digital and Vijay Sales. Every price we keep has a URL behind it. Prices we cannot attribute are dropped rather than estimated, and a retailer that disagrees wildly with the others is excluded and reported as a conflict.
- 03
We work out what you actually pay
Bank card discounts, coupons and cashback come off the list price and form the effective price. Exchange value and no-cost EMI are listed separately: they are real, but they do not reduce the cash you hand over, and you may not qualify.
- 04
We measure today against its own history
Not against the MRP, which is meaningless, but against what this variant has actually sold for over the past year. That is what tells you whether today's price is good.
- 05
We price the value of waiting
The expected discount from the next sale, multiplied by the chance it happens and the chance this model is included, plus its own downward drift — minus the risk of losing today's offer, the chance the price climbs, and stock running out. If the gap is thin, we say buy.
- 06
Your deadline wins
A discount that lands after the date you need the thing is not a discount you can have. We say what you are giving up, and recommend buying anyway.
- 07
Confidence comes from evidence, not tone
How many retailers we could verify, how much history we found, how fresh it is, whether sources agreed, and whether the sale is announced or merely likely. A model is never asked how confident it feels.
What we will not do
- Invent a price, a sale date, a discount or a launch we could not verify.
- Average away a disagreement between sources instead of showing it to you.
- Claim high confidence on thin evidence.
- Let a commission move a verdict, a target price or a retailer’s position.