Quality Surface

DevFun Arena poker agent · Live signal from the agent's own decision log — not a mock.

Try it on Monad: 1 · Side with a verdict — your wallet writes to chain 2 · Watch rejections draft policy — nothing auto-applies 3 · Verify an anchor yourself — your browser checks the contract, not us
FROZEN frozen export, file date 2026-08-20 · latest decision #26175 · run tools/sync_live.sh to go live
READ-ONLY the agent you're supervising holds 0 MON, 28 txs

0x260e6103D23AAB166C56417F3bA28aF4239b348B · checked 2026-10-13 22:46:53 UTC · cached up to 15 seconds · txs is the outgoing transaction count (nonce)

The advisory loop is broken

6243 policy suggestions generated by the shadow-LLM reviewer, only 6 promoted into real policy (0.1%). Good ideas are being generated and never reaching production.

Stale by construction: these are frozen export, file date 2026-08-20, not the live log. The Promotion Pipeline adds the missing human side of the loop.

Shadow-LLM Agreement

84.3%

1497 / 1775 linked reviews agreed with the agent's action

Review Provider Health

35.7%

error rate across 10388 shadow-LLM review calls

  • clean6682
  • HTTP Error 403: Forbidden2364
  • parse: illegal498
  • The read operation timed out311
  • empty_response292
  • parse: no_action126
  • HTTP Error 402: Payment Required80
  • HTTP Error 500: Internal Server Error12
  • HTTP Error 429: Too Many Requests12
  • parse: unparseable11

Linked Reviews

2712

reviews with a 1:1 decision_id link (post-fix; historical rows predate this)

Promotion Pipeline

0

policy edits staged for review · 1 approved for application · 1 anchored on Monad

Review staged edits → · Verify Monad anchors →