Modern dispute resolution mechanisms in blockchain systems are often presented as deterministic state machines backed by cryptographic guarantees. In practice, many of these systems rely on a hybrid model: a partially specified on-chain state machine combined with off-chain monitoring and intervention.
This distinction matters more than it is usually acknowledged.
A recurring pattern appears in systems where dispute games determine economic outcomes, particularly when bond distribution or penalty logic is involved. The protocol exposes a function that finalizes a game and derives a distribution mode based on a predicate such as <code>isGameProper()</code>. This predicate typically depends on external conditions, including whether the game has been flagged or blacklisted by an off-chain actor.
The critical detail is not the predicate itself. It is how and when the result of that predicate is committed to state.
Consider a system where:
<ul>
<li>A dispute game reaches a finalized state after a delay window</li>
<li>A function like <code>closeGame()</code> determines a distribution mode (for example, NORMAL vs REFUND)</li>
<li>The result is stored immutably</li>
<li>Any user can trigger this function, directly or indirectly</li>
<li>Blacklisting is performed by a privileged actor, but as a separate transaction</li>
</ul>
The system implicitly assumes that invalid games will be blacklisted before finalization or before any call that commits the distribution mode.
That assumption is not enforced.
From a state machine perspective, the following execution is valid:
<ol>
<li>The game reaches a finalized state</li>
<li>The game is not yet blacklisted</li>
<li>A permissionless caller triggers the closure logic</li>
<li>The distribution mode is computed and stored</li>
<li>The game is blacklisted afterward</li>
</ol>
Once the distribution mode is set, it is not recomputed. The decision is final, even though the underlying condition that justified it has changed.
This is not a bug in the traditional sense of incorrect computation. The system behaves exactly as implemented. The failure is in the relationship between the state machine and its assumptions.
More precisely, this is a time-of-check to time-of-use condition where:
<ul>
<li>The check depends on mutable external state</li>
<li>The result is cached as immutable internal state</li>
<li>There is no guarantee of monotonicity between the two</li>
</ul>
The system therefore relies on a timing guarantee provided by off-chain infrastructure. If that guarantee fails, even briefly, the on-chain logic produces an economically different outcome.
The important question is not whether the off-chain actor is “expected” to act in time. The question is whether the protocol enforces that expectation.
If correctness depends on an actor reacting before a certain point in time, but the protocol does not encode that requirement, then correctness is not a property of the system. It is a property of its operators.
This creates a subtle but important boundary:
<ul>
<li>The protocol appears deterministic</li>
<li>The safety property is operational</li>
</ul>
In adversarial environments, this boundary is where exploitation lives.
The issue becomes more pronounced when the state transition is irreversible. Once a distribution mode is committed, subsequent corrections, including blacklisting, cannot change the outcome. The system has already made a decision based on stale assumptions.
From an invariant perspective, the intended property might be informally described as:
“A game that is invalid must not result in a reward distribution that transfers value from honest participants.”
However, this property is only guaranteed if blacklisting occurs before the closure logic is executed. Since this ordering is not enforced, the invariant does not hold under all valid executions of the system.
This is the core problem: the protocol encodes a decision, but not the conditions required for that decision to remain valid.
The broader implication is that many systems conflate:
<ul>
<li>detection (off-chain)</li>
<li>enforcement (on-chain)</li>
</ul>
When these are not tightly coupled, the system admits execution paths that violate its intended economic guarantees.
There are several ways to address this class of issue:
<ul>
<li>Delaying closure until a blacklist window has definitively expired</li>
<li>Making the distribution mode dynamically dependent on blacklist status at claim time</li>
<li>Allowing recomputation or invalidation of previously set modes</li>
<li>Encoding a stronger invariant that ties finalization and blacklisting into a single state transition</li>
</ul>
Each approach has trade-offs, but they share a common goal: removing the dependency on timing assumptions that are not enforced by the protocol itself.
The deeper lesson is uncomfortable but necessary.
A system is not secure because its operators are expected to behave correctly within a time window. A system is secure when incorrect behavior is impossible within its state machine.
Anything else is coordination.
And coordination, under adversarial conditions, eventually fails.

Mayckon Giovani

Blockchain dispute resolution systems often present themselves as deterministic, but many rely on a hybrid model combining on-chain state machines with off-chain monitoring. A critical failure mode arises when a distribution mode (e.g., NORMAL vs REFUND) is computed and stored immutably before an off-chain actor has had a chance to blacklist an invalid game. This is a time-of-check to time-of-use (TOCTOU) vulnerability: the check depends on mutable external state, but the result is cached as immutable internal state with no enforced ordering guarantee. The protocol appears deterministic, but its safety property is actually operational — dependent on off-chain actors reacting within a time window. In adversarial conditions, this gap can be exploited. Solutions include delaying closure until a blacklist window expires, making distribution mode dynamically dependent on blacklist status at claim time, or encoding finalization and blacklisting into a single atomic state transition.

On-Chain Finality and Off-Chain Assumptions: A Failure Mode in Dispute Resolution Systems