How does SCORM resume data work at the 4 KB boundary?

SCORM resume works when the course writes learner state to the LMS before exit and reads it back on the next launch of the same attempt. In SCORM 1.2, `cmi.suspend_data` has a 4,096-character specified maximum; SCORM 2004 raises it to 64,000 characters. When state grows past the smaller limit, a course must reduce what it saves or risk a rejected write. Intle's adapter uses a staged fallback, but an LMS exit-and-resume test is still required. The limit is expressed in characters; individual LMS storage and encoding behaviour can differ.

What is stored, and what can be lost?

The SCORM runtime can save a location, selected answers, activity state and assessment progress. The 1.2 adapter first tries the full state, then shorter answer representations and less detailed activity records. Its final fallback retains position and score while dropping answer detail if necessary. A learner may return to the right section without seeing every earlier choice restored. A rejected LMS write can be more severe, which is why the player must check the API result and a tester must relaunch the same attempt.

The observed local boundary cases

The repository's LMS validation ledger records a local API-stub test on 11 June 2026, build 80f9989a. For SCORM 1.2 it reports a 2 KB state saved in full, a 15 KB state reduced to an 8-character answer stage, and a 600-interaction case reduced to position plus score. Each recorded rung stayed at or below 4,096 characters and restored in the stub. For SCORM 2004 the same ledger records 30 KB full state, 120 KB reduced state and a 3,000-interaction position-plus-score fallback within 64,000 characters. Those are synthetic local cases, not a claim that every real course or LMS preserves every answer.

The ledger also records a 20 July 2026 SCORM Cloud run on build bcbde6a after a real persistence bug was fixed: a quiz resumed position in both editions on that reference platform. A later 27 September local build had 10/10 API-stub scenarios. These are separate test scopes and builds; neither silently updates the other. Read the dated reference-platform result.

Reproduce the risk in your LMS

1. Use the exact exported package and note its hash, build and SCORM edition.
2. Complete several answer modes and any longer text/activity tasks, then exit through the LMS player before the course ends.
3. Reopen the same learner attempt; check the section, score, selected answers, free text and any activity state separately.
4. Repeat near the longest realistic attempt, and inspect the LMS/API logs for refused or truncated suspend-data writes.
5. Record what remained and what disappeared. If full answer restoration is essential, do not accept a position-only fallback as a pass.

Choosing an edition

The larger SCORM 2004 allowance can preserve more state, but only where the receiving LMS supports that edition and the exact package passes. Moodle's current FAQ documents native 1.2 support and incomplete native 2004 support. Canvas documents both editions when its SCORM LTI is enabled. These are vendor statements, not current Intle validation on those platforms. Use the edition guide and test your own learner attempt.