The dashboard was green for eleven days.
I want to be precise about that, because "green" was the only metric that mattered to the people funding it. Twenty-three health checks. Four alerting rules. A Grafana board that someone had screenshotted for an investor update. Every panel nominal, for eleven consecutive days, across a stack that had just closed a nine-figure round.
On the twelfth day I opened the ingest worker and read fourteen lines of decoding logic.
The upstream feed was returning an empty payload. Not a revert. Not a timeout. An empty byte string, arriving under a success flag set to true. The consumer checked the flag, saw true, and wrote null through the rest of the pipeline โ into the risk engine, into the liquidation thresholds, into the dashboard. Every downstream system that received the null interpreted it as "no new information" and held its last known state.
The last known state was healthy. The system therefore reported healthy. The system therefore was healthy.
It was not healthy. A position that should have been liquidated at a 4% drawdown survived to 31%. Nobody noticed for eleven days, because nobody had asked the only question that mattered: what does an empty answer mean?
Poetic. And entirely preventable.
Context
This is not an exotic bug. It is the default behavior of the EVM, and it has been the default behavior since 2015.
When a contract executes a low-level call, it receives two things: a boolean and a byte array.