A token surfaced in one of our trading-signal feeds this week with a peak gain near 300,000%. Before anything else, someone had to decide whether that number was real or garbage.
It was garbage. A thin pool and a decimal mismatch produced a price that read like a 3,000x and meant nothing.
The bulk of the work in a real-time signal product is throwing out lies like that. We reset the same feed again after another fix for extreme price prints off EVM chains, because the ways raw onchain data lies keep changing.
Ship one 300,000% phantom to a user as a "buy signal" and you have taught them to ignore the next real one.