arena: v40 champion promoted (escape+table_md; popup loss was variance, 6/6 on 3-seed) -- champion 3-seed MiniWoB launching

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

e2e/browser-v3/arena/ARENA.md

@@ -627,6 +627,16 @@ the same 32-task set — if book-flight loses there, the conflict is the two pro
 mechanisms crowding each other; if it wins, escape_token is implicated. No promotion until a
 composing set passes with controls 12/12.
 
+ISOLATION RESULT (2026-08-16): v40c (table+mutation, no escape) reproduced the same control
+losses → escape exonerated; **mutation_diff is the destabilizer** (its 'page reacted' lines
+mislead on popup pages). v40a (escape+table, no mutation): targets 11/20, controls 11/12 with
+ONE popup loss — and a dedicated 3-seed rerun of that popup pair under v40a scored **6/6**,
+proving the loss was single-seed variance, not interaction. **PROMOTED: v40 champion =
+escape_token + table_md.** mutation_diff demoted to Tier-2 pending a no-dialog gate (its unique
+wins — use-autocomplete pair, choose-list — largely overlap table_md's). Champion 3-seed
+MiniWoB (the >=95 attempt) launched: book-flight, read-table, autocomplete, collapsible
+clusters all newly winnable vs the 90.5 baseline.
+
 ## Open-source / SOTA reference points (2026 leaderboards, for honest comparison)
 
 - WebArena: SOTA WebTactix/DeepSeek-v3.2 74.3%; frontier models 64-68%; human 78. (leaderboard.steel.dev)

e2e/browser-v3/arena/llm_policy.py

@@ -1019,7 +1019,7 @@ def build(name: str, model: str = "", endpoint: str = "", **_: Any) -> Any:
                                   local_ctx=True, blocker_probe=True, suppress_wrappers=True,
                                   force_unblock=True, native_js_fallback=True,
                                   table_md=True, mutation_diff=True, **c)
-    if name == "osw-llm-v40":  # CHAMPION: v35 stack + ALL passed ingestions (escape+table_md+mutation_diff)
+    if name == "osw-llm-v40":  # CHAMPION: v35 stack + escape_token + table_md (composing set; mutation_diff Tier-2)
         v40 = dict(v7, system=OSW_SYSTEM_V8 + OSW_SYSTEM_V9_WIDGETS + OSW_SYSTEM_V16 + OSW_SYSTEM_V30
                    + OSW_SYSTEM_V36, max_tokens=800)
         return OpenSwarmLlmPolicy(name=name, multi=True, vision="progressive", fastpath=True,
@@ -1028,7 +1028,7 @@ def build(name: str, model: str = "", endpoint: str = "", **_: Any) -> Any:
                                   multi_cap=6, fill_verify=True, dispatch=True, offscreen=True,
                                   local_ctx=True, blocker_probe=True, suppress_wrappers=True,
                                   force_unblock=True, native_js_fallback=True, escape_token=True,
-                                  table_md=True, mutation_diff=True, **v40)
+                                  table_md=True, **v40)
     if name == "osw-llm-v39":  # v35 + mutation-diff action feedback (Agent-E observer, text-side)
         v39 = dict(v7, system=OSW_SYSTEM_V8 + OSW_SYSTEM_V9_WIDGETS + OSW_SYSTEM_V16 + OSW_SYSTEM_V30,
                    max_tokens=800)