Reproducibility
If I run an analysis again on the same data, why do I need to get exactly the same answer?
In plain words
Reproducibility means an analysis can be rerun from the same inputs and produce the same result. The data source, the time window, the time zone, the sampling interval and the calculation rules all have to be recorded and fixed.
Why it matters
Risk conclusions depend on small details. A weekly return computed from daily closes versus hourly candles, or a window boundary set in UTC versus another zone, can change the reported number. If those choices are not pinned down, a result cannot be checked or reused.
In a real week: BTC, week of 19 March 2018
For BTC, the week of 19 March 2018 (Mon) to 25 March 2018 (Sun) is defined on 1-hour candles from Binance spot BTCUSDT in UTC. With those choices fixed, the weekly return is 3.43%, the range is 13.46%, realised volatility is 90.0% annualised, and the best day is 2018-03-19 at 5.02%. The same inputs reproduce these figures exactly. Note the data holds only 5 weeks of history and 4 prior listed weeks, so percentile comparisons are left null rather than computed from an unstated sample.
A common mistake
Assuming a result is reproducible because the asset and the date range are named. Without the source, time zone, interval and calculation rule, two runs of 'the same week' can legitimately disagree.