When a validator encounters an input it cannot interpret with certainty, the safe response is to reject it — not to make a best guess and hope every other validator guesses the same way. Lithosphere builds this fail-closed principle into its consensus-critical invariants directly.
Every consensus system eventually encounters an input it was not explicitly designed to handle — a malformed field, an edge case in an encoding scheme, a value at the boundary of what the specification anticipated. What a network does with that input is one of the more consequential design decisions in its security model, and the two options are not symmetric in risk. A system can choose to make a best-effort interpretation of the ambiguous input and proceed, or it can reject the input outright and require it to be resubmitted in an unambiguous form. The first option is more permissive. It is also the one that creates the conditions for consensus failure.
The risk with best-effort interpretation is not that any single validator will get it wrong in isolation — a well-implemented validator will usually interpret an ambiguous input consistently with its own logic. The risk is that different validator implementations, or even different instances of the same implementation running on different hardware or under different conditions, may not interpret the same ambiguous input identically. If validator A interprets an edge-case input one way and validator B interprets the same input a different way, the two validators derive different state from the same data — which is precisely the condition that causes a network to fork. The ambiguity does not need to be exploited by an attacker to cause harm. It only needs to exist and be encountered.
Lithosphere’s approach is to treat ambiguous inputs as invalid by design rather than as inputs to be interpreted. This is stated as one of the network’s core consensus-critical invariants: ambiguous inputs fail closed. A transaction, message, or state update that cannot be interpreted with certainty under the network’s canonical serialization rules is rejected rather than processed under a best guess. This removes the ambiguity from the validator’s decision entirely — there is no interpretation step to potentially diverge on, because inputs that would require interpretation are not accepted in the first place.
The trade-off this creates is a deliberate one. A fail-closed system will sometimes reject inputs that a more permissive system would have processed successfully, because the ambiguity in those inputs happened not to cause a practical problem this time. That is an acceptable cost. The alternative — a system that processes ambiguous inputs and relies on validators happening to interpret them consistently — is gambling the network’s consensus integrity on a coincidence holding across every validator, every time, indefinitely. Rejecting an occasionally-valid-but-ambiguous input is a minor inconvenience. A consensus fork caused by inconsistent interpretation is a systemic failure.
This principle becomes more consequential as Lithosphere’s execution surface expands. A network supporting a single, simple transaction format has a limited number of places where ambiguity can arise. A network supporting multi-VM execution, multiple cryptographic algorithms including both classical and post-quantum schemes, cross-chain message formats, and agent-generated transactions has a considerably larger surface where an edge case could theoretically be interpreted two different ways by two different validators. Fail-closed as a standing principle means each new capability added to the stack inherits the same discipline: define what is valid precisely, and reject everything that falls outside that precise definition, rather than attempting to handle every possible input gracefully.
The alternative philosophy — accept what you can, interpret generously, handle edge cases as they come up — is common in application software, where a mishandled edge case produces a bug report rather than a fork. Consensus-critical infrastructure does not get that margin. Lithosphere’s fail-closed invariant reflects an understanding that in this specific layer of the stack, being strict about what is valid is not excessive caution. It is the only way to guarantee that every validator, given the same input, reaches the same conclusion every time.
