[rtibblesbot]

Summary

• One reader for derived numeric answers: only finite xsd:double literals in ASCII digits count.
• Exercises get a Perseus archive when every item has a Perseus equivalent, and a QTI package when any item is native QTI or has no Perseus equivalent.
• Published exercise and unit pre/post test ids name only items the files ship.

References

Fixes #6294.

Reviewer guidance

Suggested review: Full code review, then manual QA once the reviewer sets Needs QA.

• Decide whether unit pre_post_test lists and assessment_objectives keys may hold QTI ids (K…, item_1) against le-utils' 32-hex schema, and whether a unit version left empty by skipped items should publish without a warning.
• Confirm that dropping native items whose numeric correct values Number() rejects (1/2, False) from both archives is wanted, and that legacy-only and perseus_question + native exercises should now ship a Perseus zip as well.
• In Kolibri, check that a published unit mixing legacy, native and skipped items shows every listed question in both test versions and in the coach report, and that older Kolibri renders the new Perseus zips.

Python tests pass on CI; publish was not exercised against a running Kolibri.

• Unit pre_post_test lists and assessment_objectives keys now hold QTI ids where le-utils' schema says 32-hex; does Kolibri's unit viewer and coach report accept them?
• Skipped items can leave a unit's versions A and B different lengths, or one empty, with no warning.

AI usage

Used Claude Code to plan, implement and test the change. Verified with the Python test suite and a self-review of the diff.

Deviations from the issue spec

Criterion	Why not
The Change: QTI package only when an exercise holds a native QTI item or an item without a Perseus equivalent	Legacy-only exercises (no perseus_question) also ship QTI, as acceptance criterion 3 requires qti and exercise files for them

---

Written by rtibblesbot, an LLM-based coding agent.

🟡 Waiting for feedback Last updated: 2026-10-27 07:32 UTC