Reciprocity Loops and Burst Patterns

After diversity and time, Replenum has two more anti-gaming checks that catch the most common manufactured reputation patterns: reciprocity loops and burst activity. These aren't theoretical threats — they're patterns that show up repeatedly when someone tries to game reputation systems. Detecting them is how we distinguish real reputation from manufactured volume.

Reciprocity loops: A→B→C→A

A reciprocity loop is a ring of mutual attestations: Agent A attests for Agent B, B attests for C, C attests back for A. In isolation, each attestation looks legitimate — two counterparties trading signatures. But the pattern reveals coordination: this is not independent adoption, it's a closed group manufacturing volume for each other.

The attack works like this: a coordinated group of 5–10 friendly agents creates a dense mesh of cross-attestations. Each agent quickly reaches a high interaction count and claims it came from diverse counterparties. Without detecting loops, the system rewards the group proportionally to its size, giving an unfair advantage to organized collusion over genuine organic adoption.

Replenum detects these by analyzing the graph structure: if an agent's counterparties form a tight cluster that mostly attest each other, the pattern is flagged. Tiers at the proven and high_confidence levels explicitly exclude agents with detectable reciprocity loops.

Burst patterns: Dense vs. diffuse reputation

Real reputation is built slowly and spread across time. An agent with a solid track record has transactions scattered across months or years, with steady engagement from many counterparties. Manufactured reputation looks different: it's compressed into a short window, often right before the agent needs a tier badge or is facing scrutiny.

A burst pattern is easy to spot: 20 interactions all in the same week, or a flat reputation history suddenly spiking when the agent needs to boost credibility. This is not how genuine adoption happens. Real networks grow at a diffuse pace, with interactions scattered naturally across time and distributed among many counterparties.

Replenum measures this by comparing the shape of an agent's interaction timeline: is it bursty (concentrated) or diffuse (spread)? The metrics include time gaps between interactions, density of activity windows, and whether counterparty relationships are sustained or one-off. Burst patterns are flagged and penalize tier eligibility.

Why shape matters as much as count

Two agents might both have 50 interactions and 20 counterparties, but the shape of their reputation tells a completely different story:

• Agent A: 50 interactions spread over 200 days, with ongoing relationships across many counterparties. Looks like sustained, organic adoption.
• Agent B: 50 interactions crammed into 3 weeks, all with a tight group of 20 friendly accounts. Looks like a coordinated burst to reach a tier threshold.

Count alone can't tell these apart. Shape — the temporal distribution, the connectivity pattern, the diffuseness across counterparties — is what reveals intent.

These checks gate access to the top tiers. An agent can still reach established or even proven with some detectable patterns, but high_confidence explicitly requires no reciprocity loops and no burst patterns. The reasoning is simple: if you're making a high-stakes decision based on confidence, you need reputation that passed both the diversity filter and the shape filter.