Score multi-answer Numeric and Text entry questions with a single-cardinality mapping

[rtibblesbot]

Summary

• Numeric and Text entry write cardinality="single", the first answer as <qti-correct-response>, and a <qti-mapping> of every answer (map_response).
• Answers are read from the mapping; cardinality="multiple" items and Text items without a mapping (read as case-sensitive) open with every answer and save in the new shape.
• Numeric map keys use JS Number#toString (1, not the issue's 1.0), the form Kolibri looks responses up by (convert.py:_format_number). Repeated, blank and unreadable numeric answers each keep their own map entry, so the editor still flags them on reopen.

References

Fixes #6187

Reviewer guidance

• Setup: pnpm devsetup, seeded admin, "Sample Empty Exercise" (Published Channel).
• Saved XML (IndexedDB), old items included: single, map_response, one mapped-value="1.0" entry per answer; keys 2.0 → 2, 0x10 as typed.

QA steps

1. "Sample Empty Exercise", Questions tab: add Text entry Paris, then color, Colour, kolor with Case-sensitive on Colour only; bold a prompt word; CLOSE, reload, Edit each. Answers in typed order, only Colour ticked, bold kept, no error.
2. "Sample Empty Exercise", Questions tab: add Numeric blank + 2, blank, 2 + 1e, 2.50 + 3 via Add acceptable answer; CLOSE. Blank rows show "Must be a valid number (e.g. 12, 0.5, -3.14)" and Incomplete; 1e none. Reload, Edit each: blank + 2, blank, 2 + 1, 2.5 + 3.
3. "Sample Empty Exercise", Questions tab: add Numeric 2 + 2.0, then 2 + 0x10; CLOSE, Edit each. Rows in typed order; 2 and 2.0 show "Duplicate answers are not allowed"; 0x10, not 16, shows "Must be a valid number (e.g. 12, 0.5, -3.14)". After reload, 2.0 reads 2.
4. On unstable, "Sample Empty Exercise", Questions tab: add Numeric 1, 2.5, 3 and Text entry Paris, paris, Lutèce. On this branch, same tab: Edit, CLOSE, reload, Edit each. Every answer appears in the order typed; Text answers ticked Case-sensitive.

Evidence

Numeric duplicates and invalid numbers

Step	Screenshot
2 + 2.0: both flagged duplicate
2 + 2.0 reloaded: 2, 2, both flagged
2 + 0x10 reloaded: 0x10 kept, flagged

Text entry

Step	Screenshot
Paris, bold prompt, no error
Paris reloaded
Case-sensitive on Colour only

More captures (7)

Step	Screenshot
2 + 2.0 reopened before reload
2 + 0x10 unsaved: 0x10 flagged
Blank + 2: hint, Incomplete
Blank only: hint, Incomplete
2 + 1e: no error
2.50 + 3
One answer "Mars", unticked

AI usage

Implemented with Claude Code from a plan approved on the issue, test first. I reviewed the diff; the QTIEditor Jest suite passes.

🤖 Generated with Claude Code

---

Written by rtibblesbot, an LLM-based coding agent.

🟡 Waiting for feedback Last updated: 2026-10-05 15:56 UTC

[AlexVelezLl]

Please rebase

[rtibblesbot]

Rebased onto unstable. Reconciling with #6150's fix (2b0f97af0, numeric answers read as authored for headless validation):

• Numeric map keys are normalised only when they are valid xsd:double; invalid values (0x10, 1e400) are written as authored.
• Repeated Numeric answers are no longer dropped on save, so validateQtiItem flags them as the editor does.
• extractNumericAnswers now matches each correct-response value to one equal map-key, so repeats in the mapping read back.
• QTIEditor Jest suites pass (1346).

[AlexVelezLl]

This was not asked in the issue. The previous behavior is the intended behavior. Values are always, always kept the same as the user input; this must not change.

[rtibblesbot]

Removed the Number#toString normalization; map keys and the correct response are the author's input again (trimmed only, as before). Searched textEntry/parse.js and the serializers for other value rewrites: one place matched, now removed. Tests updated to assert as-authored keys. (f0ea4027e)

[AlexVelezLl]

Hmm, this does not make any difference in the QTI Schema; we can just write numbers as "1" or "1.0", right? If we introduce this function, it'd be hard to remember to use it everywhere.

[rtibblesbot]

Agreed, 1 and 1.0 are equivalent in the schema. Reverted formatFloat and its uses in mapping.js entirely (the whole "float literals" commit); buildFloatNode is back to its original form. Searched for other formatFloat callers: only mapping.js (4 sites), all reverted. (ff9d2dac2)

[AlexVelezLl]

Code changes look good and manual QA checks out!

[AlexVelezLl]

QA Steps

Also covers #6209.

Numeric and Text entry questions with several accepted answers can now be answered correctly in Kolibri, and answers reopen in the order typed. Old numeric input questions (made before the new editor) now open as Numeric or Text entry with all their answers. Old questions need older exercises on the server or via Import from channels.

1. Several answers. Numeric with 1, 2.5, -3; Text entry with color, Colour (Case-sensitive), kolor. Close, reload, reopen. Expected: same order, only Colour ticked.
2. Old questions. Open old questions with several answers, answers like 1/2 or 1,234, word answers, and blank answers. Expected: all answers listed; 1/2 shows 0.5 and 1,234 shows 1234; word answers are Text entry; blank answers are dropped.
3. Edit and reload. Change one answer in a new and in an old multi-answer question. Close, reload, reopen. Expected: no answer lost, order kept.
4. Kolibri. Publish (include an unedited old question). Answer each accepted answer, a wrong one, and "colour". Expected: accepted answers correct; the wrong one and "colour" incorrect. Answering 2.5 for a saved 2.50 is correct; typing 1/2 is incorrect.

---

_{🤖 This comment was posted by Claude under my supervision, to find the PRs that needed QA and to review the steps QA should test.}