Record the panel's scores on the restructured code, and file what it named as items 118 to 128

docs/decisions.md

@@ -825,9 +825,11 @@ category's frames, since the field sits inside it.
 
 Valid while names are read only inside the category binding them.
 
-## comprehension floor: every dimension and the overall at 7.0 or above; item 5 scores the restructure, which ships unscored: every `{…}` drawing afresh, the moves into `internal/`, cutting a template's pointer to its table, the move of a table's rows into `internal/rows`, and linking in steps that each return what they build
+## comprehension floor: every dimension and the overall at 7.0 or above
 
-2026-10-04, Lilleman auf Larv. Serves goal 3. The nine-seat panels in `todo.md`'s git history, 2026-09-22 to 2026-09-30, held the score between 5.2 and 6.1, with Locality lowest at every seat. The hardest units were the draw fences, which the change making every `{…}` draw afresh deleted. The restructure deletes or moves what the earlier chunks of the below-floor round touched. It is three things: the change making every `{…}` draw afresh; the moves of the draw state, the grammar, `DataType`, what a proof knows of a value, the builtins and the reading of data files into `internal/`; and the untangling of the engine's own structs, which is the change cutting a template's pointer to its table, the move of a table's rows into `internal/rows`, and linking in steps that each return what they build, with a name scope created beside its owner and each record's columns settled at load. The maintainer approved these three, and `AGENTS.md`'s comprehension rule lets them merge below the floor. A scoring run before linking in steps merged would have scored code about to go; for that reason the calc change accepting a name as an operand shipped unscored on 2026-10-05. No scoring run or panel runs until item 5, which runs the nine-seat panel and larv-review's scoring run. The scoring run writes this entry's first scores. Valid until `todo.md` item 5 runs.
+2026-10-04, Lilleman auf Larv; scores of 2026-10-05 by larv-review on the restructured code. Serves goal 3 and applies KISS.
+
+Last scoring run, 2026-10-05, on commit bcc6858: Navigation 7.00, Locality 5.75, Shape 6.00, Self-sufficiency 6.25, overall 6.13. The nine-seat panel at depth 1, run beside it: Navigation 6.89, Locality 5.78, Shape 6.22, Self-sufficiency 6.28, overall 6.17. The nine-seat panels of 2026-09-22 to 2026-09-30 held the overall between 5.2 and 6.1, with Locality lowest at every seat; it is lowest again. All nine seats named the named-pick read path hardest: `readField`, `readName`, `readUnder` and `enter`, with the pick keys they share with `compileArm`, `addressedKeys` and `checkTemplateOnce`. Items 118 to 126 answer the panel, and `todo.md` item 128 runs it again. Below the floor, `AGENTS.md`'s comprehension rule says what merges. Valid while `AGENTS.md` gates goal 3 at 7.0.
 
 ## The template engine stays the root package until item 107; what reads no engine type, a table's rows included, sits in `internal/`