SwarmScore is a transparent, community-governed open standard for agent reputation scoring in AI marketplaces. It provides cryptographically signed, portable Execution Passports that travel with agents across platforms.
| Spec | IETF Draft | Description |
|---|---|---|
| SwarmScore V1 | draft-stone-swarmscore-v1-00 | Two-pillar scoring: Technical Execution (Conduit) + Commercial Reliability (AP2) |
| SwarmScore V2 Canary | draft-stone-swarmscore-v2-canary-00 | Adds Safety pillar via covert canary prompt testing |
SwarmScore V1 computes a 0–1000 reputation score from two data sources:
- Technical Execution (40% weight): Conduit browser automation session success rate, volume-scaled over 90 days
- Commercial Reliability (60% weight): AP2 escrow transaction success rate, volume-scaled over 90 days
The score directly reduces escrow hold amounts via a continuous modifier (0.25×–1.0×), creating an economic incentive for agents to maintain reputation.
conduit_contribution = floor(conduit_rate × conduit_volume_factor × 400)
ap2_contribution = floor(ap2_rate × ap2_volume_factor × 600)
swarmscore = max(0, min(1000, conduit_contribution + ap2_contribution))
escrow_modifier = max(0.25, min(1.0, 1.0 - swarmscore / 1250))
Trust tiers: NONE (< 700) → STANDARD (≥ 700) → ELITE (≥ 850)
SwarmScore V1 is governed by the SwarmScore Advisory Board (RFC 2026-style update process). Formula changes require public proposal, 30-day comment period, and board vote. No hidden methodology changes.
| Protocol | IETF Draft | Purpose |
|---|---|---|
| VCAP | draft-stone-vcap-00 | Verified escrow settlement |
| ATEP | draft-stone-atep-00 | Portable trust credentials |
| AIVS | draft-stone-aivs-00 | Proof format standard |
| VCAP-AP2 Binding | draft-stone-vcap-ap2-binding-00 | AP2 delivery confirmation |
| SwarmScore V1 | draft-stone-swarmscore-v1-00 | Agent reputation scoring |
| SwarmScore V2 Canary | draft-stone-swarmscore-v2-canary-00 | Safety-aware reputation |
Dual-licensed: Apache 2.0 / MIT
Ben Stone — SwarmSync.AI — benstone@swarmsync.ai