Export set row count: longest column is canonical (ruled 2026-08-25)

Both platforms size an export set by its FIRST column, silently
truncating longer later columns - and dropping a whole set when the
first container is empty, as the T1 export validation found on
motion_stopwatch (dtlist filled, tlist empty, zero rows in all six
formats). The maintainer overruled it: the longest column determines
the row count; shorter columns keep the existing NaN padding.

Recorded as export-set-row-count (decided - both platforms diverge from
the canonical identically), marked on the export docs, and the ruled
semantics stated in the set element's spec description. The T1 driver
already matches the ruling (rows required when any source buffer
filled), so no driver change - motion_stopwatch goes green when the
writers conform.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: Staacks

docs/file-format/index.md

@@ -257,7 +257,9 @@ The export block may hold one or more *set* blocks, grouping and naming multiple
 
 ### Block: set
 
-The set block will define a group of data-containers to be exported. The sets are represented in the final file: a CSV export results in a ZIP file containing a separate CSV file for each set, and an Excel export will contain a separate sheet for each set, in both cases named by the *name* attribute.
+The set block will define a group of data-containers to be exported. The sets are represented in the final file: a CSV export results in a ZIP file containing a separate CSV file for each set, and an Excel export will contain a separate sheet for each set, in both cases named by the *name* attribute. The number of rows is determined by the set's longest column; shorter columns are padded with NaN for the missing entries.
+
+{{inconsistency:export-set-row-count}}
 
 #### Tag: data
 

inconsistencies.yml

@@ -89,3 +89,31 @@
     A non-power-of-two number of input samples yields a different output
     length and different values on the two platforms; a power-of-two input
     gives identical results on both.
+
+- id: export-set-row-count
+  title: Row count of an export set
+  area: file-format
+  status: decided
+  canonical: >-
+    The number of rows a set exports is determined by its LONGEST column
+    (ruled 2026-08-25); shorter columns keep the existing NaN padding for
+    the missing entries. The current first-column rule silently drops
+    every value beyond the first column's length - motion_stopwatch's
+    dtlist data never appears in an export while tlist is empty, which is
+    data loss, not formatting.
+  affects:
+    android: >-
+      The first column determines the number of rows (DataExport.java,
+      the comment says so verbatim); longer later columns are truncated.
+    ios: >-
+      Same rule - rows iterate 0..<columns.first.count
+      (ExperimentExportSet.swift), shorter later columns are NaN-padded,
+      longer ones truncated.
+  summary: >-
+    Both platforms size an export set by its FIRST column, so a set whose
+    first data container holds fewer values than a later one silently
+    drops the excess - including the whole set when the first container
+    is empty. Found 2026-08-25 by the T1 export validation
+    (motion_stopwatch); ruled: the longest column determines the row
+    count. Identical on both platforms today, so this is a decided
+    behavior change rather than a divergence between them.

spec/root.yml

@@ -412,6 +412,12 @@ elements:
     parent: export
     since: "1.0"
     summary: One sheet or file of the export.
+    description: >-
+      One sheet (Excel) or file (CSV) of the export, holding its data
+      elements as columns. The number of rows is determined by the set's
+      longest column (decided 2026-08-25 - see the export-set-row-count
+      entry while both platforms still size by the first column); shorter
+      columns are padded with NaN for the missing entries.
     children: [data]
     attributes:
       - name: name