Connected mode is fixed and committed (ea63aad) with a rebuilt dist/, but the live site isn't updated yet. This job doesn't publish: the task workflow publishes the committed dist/ after it completes, so I haven't seen the redeployed site.

Town Energy still won't rise for now, and that isn't a Squeekville bug. Squeekchain settlement has been stuck at round 1973 since 2026-10-26 16:20 UTC, about 1,000 rounds behind the clock, and the contract isn't paused. New players will connect and verify normally, then sit at 0 energy until settlement resumes. I added no fake rounds.

Root cause

The deployment was never the problem. The local record and SqueekCore ABI match the published files exactly, so I changed no metadata, and there was no upgrade to explain. 0x74665dAe…119f is still the verified proxy, and all reads go through it.

1. RPC rate limiting: thirdweb's public endpoint answers HTTP 429 with no CORS header, so the browser only sees TypeError: Failed to fetch. A regex in useTown.ts then turned that into "Sepolia could not be reached". I reproduced this from this machine's IP, but had no failing session from a player's browser to inspect.
2. Reload bug hidden behind it: a wallet restored after reload is still reconnecting, so the first chain read was dropped and never rescheduled. The town showed "Connected to Squeekchain" while reading nothing, with signing disabled.

What changed

• RPC: rpc.sepolia.ethpandaops.io is now the primary and only endpoint, and it's also what wallets are offered when adding Sepolia. There is no backup because no alternative passed every required request:
  • thirdweb returns the 429s above.
  • publicnode has pruned the historical state.
  • drpc's free plan doesn't serve Sepolia.
• verifyProxy() for Core and Mining: it now stops on an unknown implementation before calling through it, rejects malformed slot values, checks proxyAdmin() against the admin slot, and names each mismatch. All checks still run at the same finalized block as the round reads, and no check was removed.
• Connection diagnostics: there is a new expandable section labelled by failed stage, with wallet data, URLs and credentials stripped. RPC failures now read differently from deployment mismatches.
• One small layout change: once connected at 390 wide, the signing disclosure was pushing Sign mayor order off the first screen. On phones the button now comes first, with the disclosure directly under it. You asked to preserve the design, so this is the one change to review.

Tests that passed

• Requests from the deployed origin: eth_chainId, finalized block, individual and batched requests, contract calls and storage reads 100,000 blocks back all succeeded.
• Initialization: the first connect verified both proxies and started at round 1973 with 0 energy.
• Duplicate energy: energy, checkpoint and journal stayed unchanged after two reloads, a repeated read, and a wallet disconnect and reconnect. Two tabs reading the same rounds credited them once.
• Signed order: a signed mayor order ran once and did not replay on reload.

Limits of the testing

• Deployed site not loaded: the .eth.limo and .eth.link gateways failed from this machine and the IPFS gateways refused it. I served dist/ at the real hostname through browser interception, so the Origin, CORS and RPC traffic were real, but the gateway's own headers weren't observed. Recheck connected mode once it's published.
• No new round observed: because settlement is stalled, I

No reproducible defects found. .imd-findings.json contains {"findings":[]}. No application files changed.