On November 1, 2022, one of the four databases that persist student responses became unable to write an ID that ordered questions/responses in review and report modes only. The analytics features were potentially impacted by a maximum 25% of submitted sessions during the 2 hour and 9 minute duration of the incident and, because only a subset of those sessions used contributing features, only an estimated 1% to 3% of sessions were actually affected. No data was lost and only the order of displayed questions were temporarily changed.
Immediately after discovery, we began testing remediation options and the safest and quickest way to address the issue was to reset the impacted ID, as it played no other role in processing the data. This was theoretically unable to affect a typical assessment because questions and responses are captured upon initialization within milliseconds. However, a very small number of sessions using deferred loading features (such as sections and, in some cases, lazy loading or dynamically adding items via public method) were impacted--only if the ID was reset both after an initial batch of items were loaded and before a following batch of items were loaded. In this case, the sequential ids in the later batch were set to values lower than those in the earlier section batch, due to the reset. (Using sections as an example, this caused section two to appear before section one in a report.)
To summarize, this issue was relevant only when all of the following applied:
Additional long-term adjustments were made to prevent this issue from occurring again in the future. In addition to the immediate fix of resetting the ID, the review/reporting logic was improved to no longer require the ID for ordering questions/responses. As a result, all affected sessions were thereafter automatically displayed correctly without any need for rescoring or other alterations to the data.