Record in-app validation of ipywidgets branch and handoff closure

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

Author: maartenbreddels

notes.md

@@ -269,6 +269,27 @@ state/count remains the big client lever.
 The queue-drain (a2906a3d) and spinner fix (ecbf6384) are unambiguous keeps. The remaining client-side lever is reducing per-widget
 instantiation cost (vue/widget-manager), and reducing widget count.
 
+## In-app validation + handoff closure (leasing app agent, 2026-06-12 evening)
+
+The grotto leasing app agent profiled the real home-page load (yappi, PROFILE=1 boot hooks —
+endpoint-started yappi misses the pooled render thread):
+- ipywidgets perf-widget-construction branch validated IN-APP: get_state 0.383s -> 0.171s (-55%),
+  whole render cycle -28% CPU, wire bytes identical. (Their initial "38% Instance.__get__" profile
+  ran stock PyPI ipywidgets — the leasing venv had editable reacton/solara but not ipywidgets.)
+  This is the strongest case yet for upstreaming the branch.
+- close_widget_debug 5.5% turned out to be yappi cumulative incl. the real Layout.close;
+  self time ~0.05%. Not a target.
+- AGREED NEXT WIDGETTI PROJECT: template payload caching. With get_state fixed, Vue template
+  registration is the largest single remaining widget-construction item (~13% of cycle + ~480KB
+  byte-identical wire payload per page load). Tier 1: per-process cache of serialized template
+  state. Tier 2: content-hash + HTTP-cacheable endpoint (kills wire bytes + client parse; touches
+  ipyvue + solara + widget manager).
+- Biggest app-side staging win (not widgetti): 60s JWT expiry disabling the Redis auth fast path
+  (0.4-1.2s per page load).
+
+Env note: the leasing .venv now carries editable reacton, solara, and ipywidgets — plain
+`uv run` (without --no-sync) will silently clobber them.
+
 ## Stash (possible bugs / questionable behavior found along the way)
 
 - pandas 3.0 re-render on equal DataFrame (see above) — env pin papering over it for now.