arena: compwob follow-ups -- their 73-page crash ceiling, our ordering fix verified, the bookkeeping disease named

browser-use cannot load 73 of 101 composed pages (instrumentation crash: 28-task
coverage ceiling, 2 more false claims on what loads). Our ordering fix won its exact
predicted tasks; long-chain traces isolate the true frontier as lost mid-chain
bookkeeping, to be attacked with a rendered clause checklist (v24).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WsbS5x2rYsMDxP2kW3qqmQ

e2e/browser-v3/arena/ARENA.md

@@ -142,6 +142,16 @@ caught by canary before any number shipped. Gradient: 2-part 82%, 3-part 41%, 5+
 the long-horizon sequencing frontier, plus a traced-pending cluster of simple-pair losses.
 browser-use runs the identical 101 next (their first known CompWoB number).
 
+**Follow-ups measured:** browser-use CRASHES on 73/101 composed pages (their DOM instrumentation
+fails on the legacy engine: coverage ceiling 28/101; on the 28 that load, 21/28 with 2 false
+claims -- not comparable to a full-suite number and reported only as such). Our v23 rerun with
+the ordering fix: 65.3 again -- the fix won exactly its 2 predicted tasks, variance reclaimed 2
+others; the persistent structure is 3-part 36% and >=5-part 0/8, whose traces show clean
+execution for ~10 steps and then lost bookkeeping (scroll flailing, blind coordinate guess).
+Next mechanism: a rendered clause-checklist per turn. Toolchain note: the entire stack now
+lives in ~/.cache/arena after macOS's /tmp reaper deleted pyvenv.cfg mid-sweep -- a silent-error
+class now structurally closed.
+
 ## Positioning vs public generic-harness baselines (user-supplied 2026 survey)
 
 The comparable class is generic agents, NOT MiniWoB-specialized systems (HTML-T5++ 95.2 trained