Separate document numbering from scored item numbering

[dqnykamp]

Problem

isDescription marks a document as unscored and unnumbered. The question counter, credit extraction, shuffle anchoring, and the per-item attempt button all honored it already — but the item indexing did not.

getNumItems and getItemSequence counted descriptions as items, while extractActivityItemCredit excluded them. Those two indexings feed different fields of the same SPLICE.reportScoreAndState message:

• item_scores[].shuffledOrder — scored items only
• item_updated / new_doenet_state_idx — position in the document sequence, descriptions included

The host (DoenetApps) stores per-item state keyed by the scored numbering, and rebuilds doenetStates / itemAttemptNumbers from rows ordered by shuffledItemNumber. So with any description present, item state was written to the wrong item, and the two arrays were sized and indexed inconsistently.

Descriptions were never reachable from the host UI, so this never fired in production — but it blocks adding them, which is what DoenetApps wants to do.

Change

Split the two concepts that were sharing one sequence.

	Includes descriptions?	Drives
Document sequence — getDocSequence / getNumDocs	yes	mounting, pagination, checkHidden, checkKeepLive
Scored item sequence — getScoredItemSequence / getNumScoredItems	no	doenetStates, itemAttemptNumbers, item_updated, new_doenet_state_idx

The first pair is a rename of getItemSequence / getNumItems — no behavior change; the existing doc comment on getNumItems already said "the number of documents that will be rendered".

The scored ordering is the shuffled one, because the host rebuilds both arrays from rows ordered by shuffledItemNumber asc. getScoredItemSequence is implemented by filtering getDocSequence rather than by a separate recursion, which makes the subsequence invariant structural and sidesteps the slice(0, numToSelect) branch.

Other behavior:

• A description reports nothing on state update. It holds no slot in doenetStates and its state is not persisted, so emitting a report would cost a saveScoreAndState round trip per keystroke for a row loadState ignores. The host tolerates a missing item_updated, so this is a no-op there.
• new_attempt_for_item is unchanged. It is derived from item_scores, so it is already in original (unshuffled) order and already description-free. It legitimately differs from item_updated; a test pins that so it is not "fixed" later.
• Descriptions inside a select now throw instead of silently miscounting. extractSelectItemCredit scores a single-document select regardless of isDescription, and the select branch of propagateStateChangeToRoot averages over all selected children without filtering. Viewer already calls this inside a try, so it surfaces as the existing "Error in activity source" banner.
• A document with no scored slot is a no-op, never an error. Both updateSingleState and generateSingleDocSubActivityAttempt now return the state unchanged when the id is missing from the scored item sequence — whether because it is a description or because it is stale (an in-flight save from a just-regenerated attempt, or a select that re-picked its children). generateSingleDocSubActivityAttempt previously let such an id fall through and surface as an error banner.
• Activity.tsx falls back to itemAttemptNumber 1 and no answerResponseCounts for a description, instead of reading whatever itemAttemptNumbers[-1] gave (undefined → NaN attempts-left, masked only because the button is hidden).

Compatibility

No change to ExportedActivityState, isExportedActivityState, or sourceHash — only the length and meaning of two arrays narrows.

Old saved state cannot collide with the new indexing: createSourceHash hashes isDescription, and validateStateAndSource rejects a hash mismatch, so any source containing a description necessarily has a different hash and starts fresh. Sources without descriptions compile and index exactly as before.

Tests

npm run test — 57 pass (47 pre-existing unchanged, which is itself the compatibility evidence: every description-free fixture still asserts the same item_updated, new_doenet_state_idx, and itemAttemptNumbers).

New unit tests use the existing seqWithDes.json fixture:

• document vs scored counts, and the subsequence relation across several shuffled attempts
• the contract test — for every i, the shuffledOrder of scoredSeq[i] in extractActivityItemCredit equals i + 1. This is the exact invariant the host's shuffledItemNumber depends on; it catches any future reimplementation that drifts.
• a description state update is a no-op (identity-equal return, no postMessage)
• a scored document following a description reports item_updated strictly less than its document index and equal to its own shuffledOrder
• generateSingleDocSubActivityAttempt bumps only its own scored index, and new_attempt_for_item may differ from item_updated
• edge cases for the description/scored split, each checked over several shuffled attempts: a bare description at the root, a sequence of nothing but descriptions, a description as the last child (so the shufflable run ends at it), descriptions on both ends, and a sequence nested in a sequence with a description in each
• new selWithDes.json fixture → getNumScoredItems throws
• a generateSingleDocSubActivityAttempt for a description, and one for a document not in the activity at all, are both identity-equal no-ops that set no errMsg and post no message

New Cypress component test (ActivityViewer.descriptions.cy.tsx): page count includes the description, attempt buttons exist only for scored items, the dialog on the first problem reads "problem 1" (not 2), and reported item_updated is 1 with the description absent from item_scores.

Version

Bumped to 0.1.0-alpha-18. DoenetApps needs this published before its companion PR can be merged.