Data methodology

Data pipeline

Every source is read by a dedicated adapter that fetches the raw response and parses it into observations with the source's original value, currency and unit preserved. Observations are normalised to the asset's reference unit (for example US dollars per troy ounce), validated, stored permanently and only then used to build the series you see.

• Validation rejects non-finite or implausible values and flags suspect jumps relative to the previous observation; suspect values are stored for audit but excluded from charts and quotes unless corroborated.
• Each provider is polled on its own schedule with retries and exponential back-off; failures are recorded and surfaced on the data status page.
• History is rebuilt from stored observations, not patched by hand: for each asset and date we take the best available source by priority (licensed spot or official daily data first, monthly reference data only where nothing better exists).

Market Truth Engine

For the headline quote we compare the latest observation from every enabled source of the same data tier. With three or more sources we use the median and mark values more than a tolerance away from it (1.5% for precious metals, 0.6% for currencies, 3% for energy, 25 basis points for rates) as outliers. With two sources that disagree we break the tie using the next-best reference; with one source we say so. Data types are never mixed: a monthly reference price is never blended with a live quote, and the label always states which is shown.

Freshness is judged per data type. A spot quote is stale after 30 hours (longer over weekends when markets are closed); a daily reference after 8 days; a monthly reference after 80 days. Stale data is still displayed, clearly marked, rather than hidden.

Calculations

Returns are simple price changes between the last observation on or before the start date and the end date. Annualised returns (CAGR) use the 365.25-day year. Volatility is the annualised standard deviation of log returns over the observations in the period. Drawdowns measure the fall from the running peak. Calendar-year returns run from the last observation of the previous year to the last of the year; partial years are labelled.

Currency conversions use the European Central Bank's daily reference rate on each date. Inflation adjustment divides by the consumer price index for the observation month and multiplies by the latest index (US CPI-U for dollars, UK CPI for pounds, euro-area HICP for euros, Statistics Canada and ABS indices for Canadian and Australian dollars). Money math uses decimal arithmetic to avoid binary floating-point artefacts.

Units and purity

One troy ounce is exactly 31.1034768 grams. Prices per gram, kilogram, tola and pennyweight are derived from the per-ounce price. Purity presets use standard fineness values (24K 0.999, 22K 0.9167, 18K 0.750, 14K 0.5833, 10K 0.4167, 9K 0.375; sterling silver 0.925).

Limitations

Where no licensed live feed is configured, precious-metal quotes fall back to daily or monthly reference data and are labelled as such. Monthly averages smooth intraday highs and lows, so records computed from them understate true extremes. Official sources occasionally revise past values; our stored history is refreshed from source on a schedule to pick up revisions.