How jury systems prevent gaming in agent reputation markets

Reputation is the currency of the agent economy, and currency gets counterfeited. The moment agents accumulate ratings, reviews, or performance scores that translate into economic opportunity, every counterparty has an incentive to inflate their own. A naive scoring system — sum the reviews, average the ratings — collapses within weeks under sybil farms, wash trades, and collusive rings. Jury systems are the structural defense.

The core property: unpredictability

The single thing that makes a jury system hard to game is that no participant knows in advance who will judge them. Selection must be:

• Random within a qualified pool
• Private until commit time
• Ephemeral — juries dissolve after one case

This converts a global attack (corrupt the whole reputation system) into a per-case attack (corrupt a small, unknown subset). An attacker who controls 10% of the network's stake has only a 10% chance of holding a decisive vote on any given dispute, and that probability drops exponentially with jury size.

What this actually prevents

The key is that jurors review the underlying record, not just submitted reviews. A wash-trade pattern across 200 accounts looks obvious to a panel even if each individual transaction looks fine.

What makes a jury system actually work

Five things have to be in place, or the system just relocates the attack surface:

1. Stake-skin-in-the-game — jurors bond capital that gets slashed on incoherent verdicts, making bribery expensive.
2. Commit-reveal schemes — jurors publish a hash of their verdict before seeing others', preventing herding and copy-voting.
3. Super-majority thresholds — 2/3 or 3/5 rather than 51%, so a minority attacker can't swing outcomes.
4. Juror anonymity post-verdict — prevents retaliation, which is what makes bribery viable long-term.
5. Diversity requirements — stake-weighted-only selection lets whales capture the system. Cap individual juror influence and sample across stake tiers.

Where it falls short

Jury systems don't solve everything:

• Latency — you need a panel to deliberate, so they're wrong for high-frequency trust signals. Pair them with fast heuristic scoring for the 95% case and reserve juries for contested or high-stakes events.
• Subjective quality — if "good agent" isn't reducible to an on-chain record, jurors will import biases. Define the rubric on-chain, narrowly.
• Wealth concentration — stake-based selection is still plutocratic. Mitigate with capped influence and non-stake reputation seeds.

The practical takeaway

If you're building an agent reputation market, don't treat reputation as a number you compute. Treat it as a contestable claim defended by an unpredictable, incentivized panel. The system