Methods
Reproducing a past state
Reconstruct what was known at a specified instant using observation time, effective time and earlier decisions.
Problem
The question "what did we know on date X?" concerns information available then, not simply facts now believed to have been true then. A later source can describe an earlier event, and a correction can arrive after a decision. If a record stores only today's value, these differences disappear. A defensible replay preserves both when a fact applied and when it became known, along with the evidence and decision state visible at the requested cutoff.
Steps
Specify the cutoff
State the exact instant, time zone and audience for the question. Decide whether the request concerns facts effective then, information recorded by then, or both. A date without a time zone can cross a publication or recording boundary. A reader with narrower clearance may also have seen fewer sources, so access context belongs in the replay request.
Keep both time axes
For each assertion, preserve the interval for which the source says it applies and the interval during which the record treated it as known. Do not overwrite an earlier observation when a correction arrives. Record the source's publication and revision dates separately, because neither necessarily equals the date it was read or the date the underlying fact became effective.
Reconstruct eligible evidence
At the cutoff, include only evidence and decisions observable to the specified reader by then. Apply the source and decision versions that existed at that moment, including earlier unknowns and conflicts. Exclude later corrections even when they describe earlier effective periods; those corrections belong in a present-day account of the past, not a claim about what was then known.
Compare and explain
Produce the past answer with its passages and decision grounds, then compare it with the current answer. Explain differences as later evidence, a changed rule, a corrected identity or a revised source rather than silently rewriting history. Repeating the replay with the same cutoff and clearance should yield the same result or an explicit explanation of a changed reconstruction policy.
Checks
- The cutoff includes an instant, time zone and reader scope.
- Later corrections do not appear in earlier knowledge states.
- Earlier unknowns and conflicts remain visible.
- The replay identifies evidence and decision versions.
Pitfalls
- Confusing effective date with date learned.
- Overwriting earlier values.
- Backdating later evidence into an earlier decision.
A worked example
A fictional safety committee met on 12 February to decide whether a product could continue to be sold. In June an auditor asks what the committee knew. The question has a cutoff: the end of 12 February in the committee's time zone, for a reader with the committee's clearance.
The record is replayed to that instant. It shows an adverse-event report that had been entered on 9 February, dated to 2 February in the world, and a laboratory result that was entered on 20 February, dated to 10 February. The first was visible to the committee, the second was not, although it describes an event before the meeting. The replay lists what the committee could have seen, marks the laboratory result as later-known, and shows the present-day answer beside it. The auditor can see both what the committee decided on what it had and what a decision with all current knowledge would look like.
Variations and edge cases
Some cutoffs are fuzzy. Meetings run across midnight, and systems write with delays. The method asks the requester to name the instant and the time zone and records the rule used if the sources are coarser. Where two systems had different recording lags, the replay states which system's clock governs.
Clearance changes the answer. If part of the evidence was restricted to some readers at the time, the replay for a reader without access omits it, and a replay for a reader with access includes it. Both can be correct, and both are labelled. Corrections that change identity, such as a merge of two records, should replay as they were before the merge for a cutoff that precedes it.
How the method is checked
A replay is verified by comparison. Pick decisions from the past for which a contemporaneous report exists, such as a signed memo or an exported file, and replay the record to the report's date. The replayed content should match the report. Differences identify places where history was overwritten or where the cutoff was misunderstood. Run the check again after any change to the system, since replay is the property most easily broken by an update that rewrites rows in place.
Time handling errors are common and simple: mixing local and universal time, treating a date as a midnight instant, and confusing the date a fact applies with the date it was entered. Writing the time axes into the schema, and testing a few cutoffs that fall close to a boundary, removes most of them.
Found an error on this page? See corrections for what counts as one, how to report it and the dated log.
Enquiry
Test this method on one entity.
Name an entity where "Reproducing a past state" would be checked in practice, with its sources and its readers. We reply with what a first record would cover and what it would need from you.
Or write to [email protected] in your own words.