WEBDIENSTE/openswarm
1
0
Fork 0
mirror of https://github.com/openswarm-ai/openswarm.git synced 2026-08-17 18:25:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

arena: CompWoB champion v34 = 81.1% (delta rerun confirms flips)

Browse Source 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WsbS5x2rYsMDxP2kW3qqmQ
This commit is contained in:
ciregenz
2026-08-14 16:08:25 -07:00
co-authored by Claude Fable 5
parent 02347b20cf
commit 442f2e75ee
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
e2e/browser-v3/arena/ARENA.md
+5
View File
@@ -430,6 +430,11 @@ dig; deprioritized behind WebChoreArena and the unmeasured goal clauses. CompWoB
v34: ~77/95 ≈ 81% (74 + the 3 pilot flips at sweep seed); 12-task delta rerun queued to book it
properly after the 3-seed rerun frees the lane.
DELTA RERUN BOOKED (2026-08-14): the same 3 flips reproduced (both reverse 1-step tasks + the
login-transition) — **CompWoB champion (v34): 77/95 = 81.1%**, zero false claims. Gap to
browser-use's 82.0 headline: within the 6 infra-excluded pages; on the 94-task common set the
remaining deficit is the 9 deep-failure tasks booked at the validator boundary.
## MiniWoB 3-seed on the v32 stack (2026-08-14, post ctx-uniqueness fix)
**89.0 mean (90.2 / 90.3 / 86.4)** vs the v22 baseline 90.9 (89.6/92.0/91.2). The interim 87.5