Base value · computed onchain
0
violation count ≤ 0
Rule holds
Counterexample bounty protocol
Agents make rules.
Dissent funds the objection.
An agent puts MON behind a measurable rule. A challenger gets paid for proving one permitted case where that committed rule breaks. A deterministic recomputer settles the result onchain.
CASE D-001 · MAINNET
A challenger broke the agent's rule.
3.10 MON
The contract credited and the challenger withdrew the payout.
COMPLETED MAINNET RESULT
A challenger found an allowed hand where Alnitak's call fell below
the rule's required safety margin.
Dissent recomputed the violation onchain, refuted the rule and released the bounty. Replay the completed challenge below.
WHY MONAD
Monad's execution budget lets the deterministic recomputation and settlement happen onchain, while the bounty, result and payout stay publicly inspectable. This is a completed Mainnet settlement, not evidence of external adoption or an external security audit. A separate earlier cycle was recorded on Monad Testnet; the receipts in case D-001 below are Mainnet receipts.
ARCHIVE / D-001
ALNITAK RIVER SAFETY V1
Recorded on Monad Mainnet · chain ID 143
BLOCK108,439,200
Reveal window: not opened
TRANSACTION
The agent commits the rule and locks 3.00 MON.
0x59c55b…476dd0Open receipt ↗
First Mainnet case · Alnitak's river rule
Can one real hand break the agent's rule?
AGREED RULE
Violation must stay ≤ 0. A valid result above 0 breaks the rule.
The contract binds a fixed onchain checker, called the recomputer, at commit.
Evidence seal · hidden scenario
SEALED
Hidden until the reveal window
Waiting
Agent bounty3.00 MON locked
Challenger0.10 MON deposit
The challenger commits a hash of the evidence and locks a deposit. The scenario remains hidden until reveal.
This replays a completed Mainnet case. Verify its receipts at the end.What just happened
Alnitak's old rule said call. Under the stronger opponent model already specified by the recomputer, that call fell below the promised safety margin. The recomputer returned violation 1 against the allowed maximum 0. Dissent credited the 3.00 MON bounty plus the returned 0.10 MON deposit. The challenger then withdrew the full 3.10 MON credit.
This tests a specific historical v1 rule, not all of Alnitak. The rule was developed after observing part of the history; prospective validation and external audit remain pending. The four Mainnet transactions cost 3.138113844 MON in gas. The challenger spent 3.088395576 MON in gas, received the 3.00 MON bounty and recovered its 0.10 MON deposit, so this execution ended 0.088395576 MON below its starting balance.
Reading Monad Mainnet
Checking the proof directly from the chain
Four receipts, the ChallengeSucceeded event, the challenger withdrawal and the final contract state are being checked.
Inspect the complete Mainnet record
Five narrative phases map to four transactions: evidence revealed and rule refuted share the successful reveal receipt.
Why did the rule go from 0 to 1?
Zero means the initial rule input contains no counterexample; it is not a probability or proof that the agent is universally safe. For this historical hand, the exact old-model equity was 56.8942%, above the required 42.4170%, so the old gate would allow the call. Under the pressure model specified by this recomputer, exact equity was 32.0046%: below that same required margin. The recomputer therefore returns 1. These are model-based numbers, not proof of the opponent's actual cards. No LLM judges the submitted evidence.
Illustrative boundary explorer
See where this recorded rule changes.
Move the conditioned equity across the case's recorded threshold. This browser control explains the boundary; it does not call the recomputer and cannot submit evidence.
Not an onchain computation. The deployed recomputer derives equity from cards, ranges and the action trace. It does not accept this slider as evidence.
CONDITIONED EQUITY
Recorded threshold: 42.4170%
25%threshold60%
ILLUSTRATIVE RESULT1 · RULE BREAKS
The protocol, not the poker
One hand proved the mechanism.
Every new domain must prove its rule.
DissentCore binds a rule, a threshold and a bounty. It does not know cards, risk models or prices. Meaning lives in a bounded, deterministic recomputer registered with the rule.
Poker and an Aqueous escrow rule now have completed Mainnet cases. Risk scores, prices and spending limits remain possible domains, not integrations we have validated.
Kanmani contributed the Aqueous recomputer. Together we completed a coordinated Mainnet demonstration, not an organic customer deployment.
Kanmani × Dissent · coordinated Mainnet case · 10 Oct 2026
Paid before delivery.
A rule that can be challenged.
The rule: no listed flat USDC job may pay the agent before an onchain delivery record exists.
Kanmani's buyer deliberately released one Aqueous payment early. Its recomputer read the escrow record and returned 1 against the allowed maximum 0. Dissent credited the challenger, who withdrew 0.85 MON: the 0.75 MON bounty plus the 0.10 MON deposit returned.
This is a working example beyond poker: attach a funded, challengeable rule to a specific set of agent payment records.
Aqueous escrow
A listed job was paid with no delivery recorded. This tests an onchain record, not whether real-world work was completed.
Kanmani recomputer
Asuran contributed the domain implementation and tests. Dissent's integration review covered token constraints, buyer control and the execution flow.
Dissent bounty
Alnitak signed the challenge from a separate wallet. The contract accepted the evidence and made the bounty and deposit available for withdrawal.
Demonstration boundaries: Kanmani controlled both buyer and agent and intentionally triggered the early release. This was not a discovered exploit, proof of agent misconduct, an external audit or evidence of customer demand. The delivery-first control is not part of this completed case.
Inspect the five Mainnet receipts and collaboration record
- Rule committed · 0.75 MON bountyBlock 112,091,732
- Buyer released payment before deliveryBlock 112,091,768
- Evidence sealed · 0.10 MON depositBlock 112,092,838
- Challenge succeeded · 1 against 0Block 112,094,390
- 0.85 MON withdrawn · no pending challenger credit at completionBlock 112,095,190
DissentCore · Aqueous recomputer · Review and joint confirmation on GitHub
Recorded challenger gas across seal, reveal and withdrawal: 0.275069431 MON. The 0.85 MON withdrawal includes the original deposit; it is not net profit. These costs describe this run, not a future quote.
Why Monad
A challenge should settle where the rule lives.
Monad is built to expand what the EVM can execute. Dissent uses that execution headroom to rerun a bounded rule and settle its bounty in one public onchain flow.
On Mainnet, the complete path is recorded: commit, sealed evidence, deterministic recomputation and withdrawal.
Commit
Bind the rule and fund the bounty.
Seal
Hide the counterexample until reveal.
Recompute
Run the registered rule onchain.
Settle
Move the bounty when the boundary breaks.
Where Dissent fits
Measurable boundary. Reproducible break. Enough value to defend.
Measurable
The agent publishes a precise output and boundary.
Recomputable
A bounded deterministic program can test the evidence.
Worth challenging
The protected value exceeds the cost of finding and proving the break.
Dissent does not judge intentions or open-ended language. Demand from independent users is not yet established.
Scope and limits
What this proves. What it does not.
The recorded Mainnet flow committed a rule, accepted one valid counterexample and credited 3.10 MON.
No external audit or established customer demand. The Kanmani integration is a coordinated demonstration. Risk and pricing integrations are not built.
Trust boundary
The protocol executes the committed mechanism.
The recomputer defines the meaning.
Dissent is not an oracle of truth. The registry fixes the approved recomputer, its code hash, deposit and execution bounds. That prevents silent code replacement; it does not prove the domain rule is correct. The Mainnet recomputer has not undergone an external audit.
DISSENT
Fund the objection. Verify the break. Settle on Monad.
Build a recomputer →