Pin RecordRewrite's clear arm at the renamed record (improvise-q2mq)

A blank edit on a measure a renamed drill row lacks must clear at the renamed
key. The phg1 sparse-row test covered only the write arm, so aiming the clear
at the unrenamed key left the workspace green. The reviewer's probe fails under
that rival and nothing else does.

Also corrects the phg1(c) evidence. The swap test cannot see the narrow rival
(eager clears, deferred writes). It does catch the wider one (eager clears and
writes).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdfEGXA7aJ6ftMsCc9PJze

context/test-evidence/command.md

@@ -5501,12 +5501,32 @@ nothing else in the workspace fails.
 ### every_drill_row_reads_the_model_before_any_row_clears_it
 
 improvise-phg1(c). `RecordRewrite::merged` puts every clear after every read.
-The rival lets each row's clears land before the next row reads the model, and
-the uo2p swap test does not see it: a row's own clears touch only its own old
-keys, which no other row reads. What separates the two is a clear aimed at
+The rival lets each row's clears land before the next row reads the model, but
+defers every write to the end. The uo2p swap test cannot see **that** rival: a
+row's clears aimed at its own old keys touch nothing another row reads. A wider
+rival, where each row's clears *and* writes land before the next row reads, is
+a different fault: the swap test does catch it (uo2p's rival 2, which fails
+`a_drill_that_swaps_two_records_coordinates_swaps_them` alone). So this test
+exists for the narrow rival only. What separates it is a clear aimed at
 **another row's** key. A blank edit on a measure the row lacks clears that
 measure at the row's *renamed* coordinates. So row 0 is renamed onto row 1's
 coordinates and blanks `Profit`, which only row 1 has, and row 1 is renamed
 away. Row 1's `Profit` must still arrive at `Food/Jan`. Under the rival it is
 cleared before row 1 reads it, `Food/Jan/Profit = 5` is lost, and nothing else
 in the workspace fails. This probe is the reviewer's, from the ticket.
+
+### a_blank_edit_on_a_measure_a_renamed_row_lacks_clears_it_at_the_renamed_record
+
+improvise-q2mq. `RecordRewrite::of`'s second loop handles the measures a row has
+no cell for. Its **clear** arm must aim at the *renamed* key, exactly as its
+write arm must (improvise-phg1(b)). The phg1 sparse-row test stages a value, so
+it covers only the write arm. Aiming the clear arm alone at
+`row.cell_key(measure)`, unrenamed, left the whole workspace green.
+
+The probe is the reviewer's. Row 0 (`Food/Jan`, `Revenue = 1`) is renamed onto
+`Clothing`, where row 1 already holds `Profit = 5`, and blanks `Profit`, which
+row 0 lacks. The blank belongs to row 0's final record at `Clothing/Jan`, so the
+store is exactly `{Clothing/Jan/Revenue = 1}`. Under the rival the clear lands
+on `Food/Jan/Profit`, where there is nothing to clear, and
+`Clothing/Jan/Profit = 5` survives. Injected alone into a full workspace run,
+the rival fails this test and nothing else.

crates/improvise-command/src/ui/effect.rs

@@ -3300,6 +3300,24 @@ mod tests {
         );
     }
 
+    #[test]
+    fn a_blank_edit_on_a_measure_a_renamed_row_lacks_clears_it_at_the_renamed_record() {
+        assert_drill_over(
+            &[
+                (drill_cell("Food", "Jan", "Revenue"), CellValue::Number(1.0)),
+                (
+                    drill_cell("Clothing", "Jan", "Profit"),
+                    CellValue::Number(5.0),
+                ),
+            ],
+            &[
+                (0, category("Type"), "Clothing"),
+                (0, measure("Profit"), ""),
+            ],
+            &["Clothing/Jan/Revenue = 1"],
+        );
+    }
+
     #[test]
     fn a_drill_row_renamed_twice_and_edited_lands_as_one_record() {
         assert_drill_commits_to(