Start with your own event log
Record the first sustained change in impressions and clicks, not the first anxious day in Analytics. Note releases, migrations, domain or URL changes, template edits, tracking changes, outages, content removals, and major link events. This is the site's causal timeline.
Google's Search Status Dashboard supplies a second timeline: announced ranking updates and system incidents. Aligning the two is useful only after both exist independently. Otherwise the public calendar supplies a ready-made story for every fluctuation.
Read rollout windows literally
An announced update has a start and completion status. Performance inside that window can be unstable, and the final direction may not match the first movement. Annotate the full window, wait for completion before judging durable winners and losers, and choose comparable periods that are not distorted by seasonality.
Google's history is intentionally selective: not every smaller system change is announced. The absence of a public update does not prove that Google changed nothing, and the presence of one does not prove that it changed your site.
Require a matching pattern
A useful update hypothesis predicts something observable. If an update supposedly changed how a class of content is evaluated, the affected pages and queries should show a coherent difference from the site's unaffected pages. If every page fell at the exact moment a robots rule changed, the public update is a distraction.
Compare result types, intent, evidence, freshness, original contribution, and site-level patterns. Use the hypothesis that best explains both the losses and the exceptions with the fewest unsupported assumptions.
Turn the timeline into a decision
The decision might be to repair an implementation, improve a weak page family, consolidate cannibalizing URLs, rebuild measurement, or wait for a volatile rollout to settle. “Recover from the latest update” is not a scope of work.
For a broad core rollout, continue with the core update investigation. For a general loss, return to the complete ranking-drop sequence.