docs: bound hand-item mutation proof to renderer consumer

docs/research/2026-09-01-overhaul/480-fix-round1-consumer-mutation.md

@@ -0,0 +1,25 @@
+# #480 fix round1 — put the renderer consumer before cache diagnostics
+
+Lead read the complete eight-file return and actual production path. No
+production error found; own Release0W0E and four changed suites79/79 PASS.
+The implementer's92/92 also includes LiveCollisionAssetPublisherTests and
+WalkProductionWorldDataTests. This is a narrow test-evidence correction only.
+
+The real cold OnSpawn test DOES check registry rows and the actual walk, but
+its mutation aborts first on prepared.GfxReads before reaching those consumer
+assertions. Original contract proof2 explicitly requires the missing-child-row
+failure. Move the first prepared-read/cache assertions to AFTER exact child
+CELLARRAY/part-row and WalkProductionWorldData assertions. Preserve every
+assertion and the control that materialization/attachment succeeds. No new
+test fixture, production edit or scope expansion. Production hashes frozen.
+
+Repeat both actual-production mutations (remove publication; use original
+Setup IDs instead of effective replacements). They must now fail on missing
+or incorrect child part entries, not producer read-count/ID diagnostics.
+Restore exact raw hashes, Release0W0E and same92-test affected lane. Return
+TRX/error messages/test hash; no other files except the existing test and this
+outcome. Lead repeats discriminating proof before independent reviews.
+
+Rawls implements after completing the separate #477 lifetime review. Both
+clients absent before each build/test; no client/commit/new agents. Root owns
+integration and ledger. No new code finding, deviation or extra renderer work.