Subsequent Events in Research: What You Owe a Decision After Fieldwork Closes

Answer first: your fieldwork closed three weeks before the readout, and the world kept moving in between. Auditing calls what happens in that window a subsequent event, and it attaches three different duties to three different windows: an active duty to search before you report, a duty to act but not to search after you report, and a duty to inform everyone who already acted on the old version. Research has a name for none of these, which is why findings get presented as current when they were overtaken a fortnight ago.

The short answer

Every study has two dates and most reports show only one. There is the date fieldwork closed, and the date the finding was presented. The gap between them is typically one to four weeks, and everything that happens in it is invisible.

ISA 560 defines the problem precisely. Subsequent events are "events occurring between the date of the financial statements and the date of the auditor's report, and facts that become known to the auditor after the date of the auditor's report."

Note that the definition covers two quite different things: events that happened in the gap, and facts about the original period that you found out later. Both can invalidate a conclusion, and they need different responses.

Why this is not insight decay

Our guide to insight decay and refresh cadence covers what happens to a finding over months: it ages at the speed of the thing it describes, and you re-validate it on triggers before reusing it.

This is a different failure with a different remedy. A subsequent event does not age a finding. It overtakes it, at full strength, while the finding is still brand new and before anyone has had a chance to consider it stale. Nobody applies a freshness check to a study that closed eleven days ago, and that is exactly the study most likely to be presented as though nothing has changed.

The two also fail in opposite directions. Decay is gradual, predictable, and manageable on a calendar. A subsequent event is discrete, unpredictable, and only catchable by looking deliberately in a specific window. A refresh cadence cannot catch it, because the report has not aged yet.

Three windows, three different duties

This is the structure worth stealing, and it is more sophisticated than anything research currently has. ISA 560 does not apply one rule. It applies three, and the duty weakens at each boundary while never disappearing.

The elegance is in the middle row. It solves a problem research teams handle badly by being unable to name: after you have presented, are you still on the hook? The audit answer is a clean split. You are not required to go looking. You are absolutely required to do something if you find out anyway. That single distinction resolves most of the awkwardness, because the objection to being on the hook forever is really an objection to open-ended monitoring, not to acting on what you learn.

The four procedures, translated

ISA 560.7 does not leave the search to judgment. It names required procedures, and each has a direct research analogue that takes minutes.

Four questions, fifteen minutes, run once immediately before the readout. Compared to the cost of the study itself this is a rounding error, and it is the only systematic defense against presenting a finding the world has already answered.

ISA 560.6 adds a sensible limit that prevents this becoming infinite: the auditor "is not, however, expected to perform additional audit procedures on matters to which previously applied audit procedures have provided satisfactory conclusions." You are looking for what changed, not re-doing the study.

Adjusting versus non-adjusting: the classification that decides what you do

Financial reporting splits subsequent events into two kinds, and the split transfers cleanly and usefully. The question is whether the event tells you something about the state of the world during your fieldwork, or reflects a change that happened after it.

The finding was wrong (adjusting). The event reveals that the conclusion did not hold even at the time you drew it. You learn after the readout that the two weeks of fieldwork coincided with a partial outage, and the frustration you attributed to the workflow was substantially about latency. That is not new information about a new world. It is a correction to your reading of the old one, and it requires re-opening the finding.

The finding stands but the context changed (non-adjusting). The conclusion was accurate for the period studied, and something has since happened that a reader must know. Your buyer research is sound, and a competitor shipped the exact capability your participants said was missing, three days after close. The finding is not wrong. Presenting it without the event attached is misleading.

The default in the third row is deliberate. Teams are strongly motivated to classify events as non-adjusting, because non-adjusting costs a paragraph and adjusting costs a re-run.

Dual dating: the trick for reporting without re-opening everything

Auditing has a device worth knowing. When something surfaces late and gets addressed, the report can carry its original date for everything except the specific matter, which carries a later one. The rest of the work is not silently re-dated as though it had all been current.

The research version is a two-line stamp on every readout:

Fieldwork closed: 4 February 2026. Reviewed for subsequent events to: 26 February 2026. One event noted, see page 2.

This is more honest than a single publication date, which implies currency the study does not have, and more honest than a fieldwork date alone, which leaves the gap unexplained. It also creates a small useful pressure: a team that has to write down the review date has to actually do the review.

The obligation nobody exercises

The third window is where the borrowed idea gets uncomfortable, and it is the reason it is worth writing down.

ISA 560.14 says that after a report has been issued, if a fact becomes known that would have changed it, the auditor must discuss it, determine whether amendment is needed, and inquire how it will be addressed. If the report is amended, ISA 560.15(b) requires reviewing the steps taken to ensure "anyone in receipt of the previously issued" version "is informed of the situation."

That is a recall. And research teams effectively never issue one. Findings are presented, absorbed into roadmaps, quoted in strategy documents and repeated in board decks long after the researcher has learned something that undermines them. The correction, if it happens at all, happens quietly in the next study rather than loudly against the old one.

The reason is not dishonesty. It is that no mechanism exists, and inventing one in the moment feels disproportionate. So build the mechanism before you need it:

• Keep a distribution list per finding. You cannot inform people you did not record. Most research repositories track where an insight came from and never track where it went.
• Set a threshold in advance. Not every new fact warrants a recall. The workable test is the audit one: would this have changed what the readout concluded, had we known it? If not, it goes into the next study. If so, it goes back to the people who acted on it.
• Make the correction as visible as the original. A finding presented in a leadership review is not corrected by a comment in a document nobody reopens.
• Treat withdrawal as a sign of health. A team that has never withdrawn a finding is not a team that has never had a wrong one. It is a team without a mechanism.

There is also a middle option between silence and withdrawal. Where the finding still holds but the event changes how it should be read, the right move is a prominent flag rather than a retraction, using the graded vocabulary set out in the graded hedge.

How a fast research cycle changes the problem

There is a genuine structural advantage worth stating plainly here, and it is about cost rather than capability.

The reason subsequent events are painful in traditional research is that every response is expensive. If your fieldwork takes three weeks to recruit and two to run, then discovering an adjusting event after the readout means the honest response costs another month, which means the honest response does not happen. The classification decision gets made by the budget rather than by the evidence.

When a study can be fielded and analyzed in a day, that pressure largely disappears. Re-opening a finding stops being a project and becomes an afternoon. In practice this is what makes the ISA 560 discipline usable rather than aspirational: the duty to act is only real if acting is affordable.

Three things make the check itself cheaper on Koji:

• Fieldwork dates are recorded per conversation, so the close date is a fact rather than a recollection, and the gap between close and readout is visible rather than assumed.
• A targeted follow-up study is cheap enough to run as a check. When an event is ambiguous between adjusting and non-adjusting, the fastest resolution is often to ask eight people directly rather than to argue about it. Conversations scoring below 3 on the 1 to 5 quality scale do not consume credits, so a small confirmatory study carries little downside.
• Structured questions make re-contact comparable. Koji supports six question types: open_ended, scale, single_choice, multiple_choice, ranking, and yes_no. Repeating the identical scale or single_choice item after an event gives you a clean before-and-after on the same instrument, rather than two qualitative readings you have to reconcile by hand. The structured questions guide covers designing items for reuse.

Common mistakes

• Publishing one date. A publication date on a study that closed three weeks earlier asserts a currency the evidence does not have.
• Assuming nothing happened because the study is recent. Recency is precisely why nobody checks.
• Classifying everything as non-adjusting. It is the cheap answer, which is why it needs a default that runs the other way.
• Treating the post-readout window as closed. No duty to search is not the same as no duty to act.
• Having no distribution list. You cannot correct a finding among people you never recorded as having received it.
• Running the check too early. Doing it the week before the readout leaves a fresh gap. Run it immediately before you present.

Frequently asked questions

What is a subsequent event in research?

It is something that happens between the close of fieldwork and the moment a finding is acted on, or a fact about the fieldwork period that you learn only afterwards. ISA 560 defines both as subsequent events. Examples include a competitor launch, a pricing change, a major outage during the fieldwork window that you did not know about, or a large customer churning between your last interview and the readout.

How is this different from research going stale?

Insight decay is gradual and predictable, and is managed with a refresh cadence. A subsequent event is discrete and immediate: it overtakes a finding that is still brand new. Cadence-based review cannot catch it, because the study has not aged yet. That is exactly why it slips through.

Am I still responsible for a finding after I have presented it?

Partly, and the audit split is the useful one. ISA 560.10 says there is no obligation to keep searching after the report date, but if a fact becomes known that would have changed the conclusion, you must act on it. You are not required to monitor indefinitely. You are required not to sit on what you learn.

What is the difference between an adjusting and a non-adjusting event?

An adjusting event reveals that the conclusion was already wrong during the fieldwork period, such as discovering an outage that explains the frustration you attributed to the design. It requires re-opening the finding. A non-adjusting event is a genuine change after fieldwork closed, such as a competitor launch. The finding stands but the event must be disclosed alongside it. When you cannot tell, treat it as adjusting.

What should a subsequent events check actually involve?

Adapt the four procedures in ISA 560.7. Ask the account, support and product leads what has changed since the fieldwork close date; read the changelog and release notes for the period; read current operational numbers such as support volume and churn; and confirm whether anyone is monitoring at all. It takes about fifteen minutes and should be run immediately before the readout, not a week earlier.

When should a finding actually be withdrawn?

Use the audit threshold: would this new fact have changed what the readout concluded, had it been known at the time? If yes, the people who acted on the finding need to be told, with the correction given the same visibility as the original. If no, it belongs in the next study. Build the distribution list before you need it, because the reason recalls do not happen is usually that no mechanism exists.

Related Resources

• How Long Is User Research Valid? Insight Decay and When to Re-Run a Study
• Structured Questions in AI Interviews
• Levels of Assurance in Research: How Much Confidence a Study Can Honestly Support
• How to Build a UX Research Repository: The Complete Guide
• Activating Research Insights: Turn Findings Into Product Decisions
• The Research Expectation Gap: Why Stakeholders Are Disappointed by Studies That Were Done Right
• The Graded Hedge: How to Flag a Finding You Cannot Fully Support
• Test the Process or Test the Output: When Research Controls Let You Review Less Data