Point-in-time data
A dataset that can be queried as it was known on a past date, rather than as it has since been revised.
A point-in-time dataset preserves what was known at each moment, including values that were later revised or corrected. Asking it about last March returns what was actually available last March, not the current best estimate of last March. Most operational datasets do the opposite: they overwrite, keeping only the latest view.
The distinction is decisive for anything tested against history. A backtest run on revised data quietly assumes the analyst knew figures that had not yet been published - a look-ahead bias that inflates results and does not survive contact with production. Physical data is particularly prone to it: production figures, permits, queue positions and trade statistics are all routinely restated after first publication.
Building it requires the provenance to be present in the first place, since a value can only be reconstructed as-of a date if the moment it was asserted was recorded. That is the same requirement an audit imposes, which is why the two are usually solved together rather than separately.
Evidence that keeps pace with the decision.
Evaluate Nuclir against the systems, markets, and decisions that matter to your organization. The result is current intelligence with the context needed to use it responsibly.