Data quality
How do we know a market result reflects real trading rather than gaps or bad prints in the data?
In plain words
Data quality is the practice of checking that the price series behind a result is complete and accurate. Missing hours, stale candles or a single bad price print (an erroneous trade or quote) can move an average, a high or a percentage return away from what actually happened.
Why it matters
Every risk figure — volatility, range, drawdown, correlation — is only as good as the inputs. A few missing hours can understate volatility or hide the true high; one bad print can invent a spike. The remedy is to check coverage, flag outliers, and refuse to publish comparisons the data cannot support.
In a real week: ETH, week of 19 February 2018
In the week of 19 February 2018, ETH data showed weeks_of_history: 1, so volume_vs_prior_12w and return_percentile_vs_history were reported as null rather than computed from a single observation. Funding fields were also null. Only one prior week was listed, with prior_weeks_down: 0 and prior_weeks_up: 1. The week's figures — a -7.66% return and a 22.27% range — were carried, while unsupported comparisons were withheld.
A common mistake
Treating a null as a zero or forcing a comparison from one week of data. A suppressed comparison is itself information: it means the evidence is not yet there.