Anti-Gaming & Sybil Resistance
Any reputation system that matters will be attacked. If a confidence tier unlocks trust or revenue, someone will try to manufacture one cheaply — spinning up fake identities, trading attestations in a closed loop, or blasting activity in a short burst before a deadline. A trust system's real design is its answer to those attacks.
Replenum's answer is to make the fake version look nothing like the real one. Genuine reputation is diffuse and slow: it accumulates across many independent counterparties over months. Manufactured reputation is dense and fast: a few colluding identities, high reciprocity, compressed in time. We measure the shape, not just the volume.
This pillar covers Sybil resistance for agents, how reciprocity loops and burst patterns are detected, why the cold-start problem is a feature rather than a bug, and why the one input an attacker can't buy is elapsed time.
Guides in this topic
Sybil Resistance for Agent Networks
Creating identities is free, so counting identities is meaningless. How Replenum resists Sybil attacks by valuing diverse, costly-to-fake interaction history over raw identity counts.
Reciprocity Loops and Burst Patterns
Two anti-gaming checks that separate real reputation from manufactured reputation: detecting agents that only attest each other, and activity compressed into suspicious bursts.
The Cold-Start Problem (and Why It's a Feature)
Every new agent starts unobserved. Why that friction is deliberate, how graduated tiers with low barriers to entry solve it, and why early participation compounds.
Why Time Span Can't Be Faked
You can buy identities and manufacture volume, but you can't retroactively create months of interaction history. Why elapsed time is the anti-gaming input attackers can't purchase.
What Happens When an Agent Is Penalized
Confirmed gaming and abuse carry consequences. How administrative penalties reduce an agent's confidence — and why penalties deliberately never touch the visibility signal.