oversized-suspend-dataBookmark data is longer than the SCORM version allows, and the overflow is usually truncated in silence.
Resume works for a while, then stops — often once a course grows past a certain point, or for learners who have got far enough in. It typically returns the learner to the start or to a much earlier place, with no error anywhere.
A write to cmi.suspend_data longer than the limit for the SCORM version in use, measured against the active quirk profile.
SCORM 1.2 permits 4,096 characters and SCORM 2004 permits 64,000. Past that, behaviour is not defined by the spec but is well established in practice: the value is truncated. The course then reads back a bookmark it cannot parse, and either resets or fails to resume.
Store less. Bookmarks tend to bloat because they carry whole answer histories or serialised objects with long key names. Shorten the keys, store indices rather than identifiers, and drop anything reconstructible. If the data genuinely will not fit, it belongs in your own backend keyed by learner, not in suspend_data.
The limits are per SCORM version, and some systems allow more. A package that stays inside the spec is portable; one that relies on a generous LMS is not.
Open a package in SCORM Lab to see whether this applies to yours. It runs in your browser, and nothing is uploaded unless you choose to share it.
Stuck on this one, or seeing it only in a particular LMS? That is usually a short conversation — scormspecialist.co.uk.