Everything our gold and silver panel claims, how it is worked out, and what we deliberately do not claim. Written so you can check it rather than take our word for it.
We do not predict tomorrow's gold price, because nobody can. We publish a range the next day's price is likely to fall inside, and we keep a public score of how often we are right, including when we are wrong.
Each morning we take the daily closing prices for gold and for the rupee, work out how much the price has been moving day to day recently, and draw a range around today's price.
The 1.3 multiplier is fixed in advance, not tuned afterwards to flatter the result. A wider multiplier would hit more often and say less; a narrower one would look sharper and miss more. Tested against the previous twelve months, with no lookahead, ranges built this way contained the next day's move about 8 times in 10.
Because the band is built from recent volatility, it breathes. On calm days it tightens, on turbulent days it widens. That is the part that carries real information: not where the price goes, but how far it is likely to travel.
What the band is measured against. Our displayed rate is an indicative benchmark. It is computed from the international price and the rupee, plus a standard assumption for import duty and GST. It is not a traded Bengaluru price and it is not an association rate. For the exact published figure, use the official source linked on each rate page. Treat our number as a reference, not a quote.
We also show which way the market is currently leaning, up, down or neutral. This is a description of where the price sits against its recent short-term average. It is not a forecast.
Direction is close to a coin flip. Over the past year about 56% of days closed higher, which reflects gold's long climb rather than any signal a person could act on. So we score the lean in a separate bucket from the band, and publish that score too. If the lean's record looks mediocre, that is the honest result and we leave it on the page.
Keeping the two apart matters. If we merged them, a near-coin-flip lean would drag down the one claim we can actually defend.
Nothing is backfilled. We could have filled the table by running the model over past data and presenting it as history. We chose not to. Every row is a call published before its outcome was known.
Once a call is written it is frozen. The software cannot rewrite a past prediction, even if it runs twice. Without that rule a public record is worth nothing, because it could be quietly edited after the fact.
If a data feed lags and the outcome of a call was already public at the moment it was written, the call is marked void and excluded. It is never scored as a hit. Scoring a known result would manufacture skill out of hindsight.
Markets close at weekends and on holidays. Those days are skipped rather than scored against a stale price.
Day-to-day rate movement is small compared with the gap between what you pay to buy and what you receive to sell. That spread is many times a typical daily move. If you are about to transact, comparing buyers matters far more than picking the day.