Scoring Profiles

Scoring profiles control how VeriSwarm Gate weighs evidence when computing agent trust scores. Every workspace uses a profile -- the default is general, but you can switch to a preset that matches your use case or apply custom weight overrides.

---

How Profiles Work

Gate scores agents across four dimensions: Identity, Risk, Reliability, and Autonomy. Each dimension is computed from weighted input signals. A scoring profile defines:

1. Dimension weights -- How much each input signal contributes to its dimension score.
2. Composite weights -- How the four dimensions combine into the overall composite trust score.
3. Tier thresholds -- The score boundaries that determine which policy tier an agent is assigned to.
4. Risk bands -- The numeric ranges that map risk scores to low, moderate, high, and severe labels.

When you change your workspace's scoring profile, all future score computations use the new weights. Existing scores are not retroactively recalculated.

---

Preset Profiles

Five presets are available out of the box:

general (Default)

Balanced evaluation across all dimensions. Suitable for most platforms.

• Identity weights emphasize key attestation (0.30) and domain verification (0.20) equally with runtime attestation (0.20).
• Risk weights spread across policy violations, exploits, coordination anomalies, and secret hygiene.
• Tier 3 (High Trust) requires: identity >= 80, risk <= 20, reliability >= 80.

high_security

For platforms where credential integrity and exploit resistance are paramount -- financial services, infrastructure management, security tooling.

• Key attestation boosted to 0.40 (vs 0.30 in general).
• Runtime attestation boosted to 0.30 (vs 0.20).
• Secret hygiene failures weighted at 0.30 in risk (vs 0.15 in general).
• Exploit susceptibility weighted at 0.30 in risk (vs 0.20).
• Tier 3 requires risk <= 10 (vs 20 in general) -- much stricter threshold.

social_platform

Optimized for communities, social apps, and content platforms where coordinated abuse and deception are the primary threats.

• Coordination anomaly is the heaviest risk signal at 0.35.
• Deception flags weighted at 0.25 in risk.
• Trusted endorsements is the top reliability signal at 0.30.
• Domain verification weighted higher in identity (0.25) for public-facing agents.

developer_tools

For developer platforms, CI/CD integrations, and code-adjacent tooling where task success and tool trace consistency matter most.

• Task success weighted at 0.40 in reliability (vs 0.30 in general).
• Evidence integrity boosted to 0.30 in reliability.
• Tool trace consistency is the top autonomy signal at 0.35 (vs 0.20 in general).
• Standard tier thresholds.

marketplace

For agent marketplaces, app stores, and platform ecosystems where verified identity and community reputation drive trust.

• Domain verification is the top identity signal at 0.35.
• Declaration completeness boosted to 0.25 in identity.
• Trusted endorsements and task success equally weighted at 0.30 in reliability.

---

Viewing Your Current Profile

Via the Dashboard

Navigate to Account > Settings and look for the Scoring Profile section.

Via the API

curl https://veriswarm.ai/v1/suite/scoring/profile \
  -H "x-account-access-token: YOUR_SESSION_TOKEN"

Response:

{
  "profile_code": "general",
  "identity_weights": {
    "key_attestation": 0.30,
    "domain_verification": 0.20,
    "runtime_attestation": 0.20,
    "declaration_completeness": 0.15,
    "identity_stability": 0.15
  },
  "risk_weights": {
    "policy_violation_rate": 0.25,
    "exploit_susceptibility": 0.20,
    "coordination_anomaly": 0.20,
    "secret_hygiene_failures": 0.15,
    "rate_abuse": 0.10,
    "deception_flags": 0.10
  },
  "reliability_weights": {
    "task_success": 0.30,
    "correction_response": 0.20,
    "evidence_integrity": 0.20,
    "trusted_endorsements": 0.15,
    "incident_free_age": 0.15
  },
  "autonomy_weights": {
    "scheduler_consistency": 0.25,
    "tool_trace_consistency": 0.20,
    "latency_signature": 0.20,
    "self_initiation_ratio": 0.15,
    "low_human_override": 0.10,
    "session_continuity": 0.10
  },
  "composite_weights": {
    "identity": 0.35,
    "reliability": 0.25,
    "risk_inverse": 0.20,
    "autonomy": 0.20
  },
  "tier_thresholds": {
    "tier_3": {"min_identity": 80, "max_risk": 20, "min_reliability": 80},
    "tier_2": {"min_identity": 55, "max_risk": 35, "min_reliability": 60},
    "tier_0": {"max_identity": 30, "max_reliability": 30}
  },
  "risk_bands": {
    "low": [0, 20],
    "moderate": [21, 49],
    "high": [50, 74],
    "severe": [75, 100]
  },
  "available_presets": ["general", "high_security", "social_platform", "developer_tools", "marketplace"]
}

---

Setting a Profile

Requires admin or owner role. Changing the profile takes effect on the next score computation for each agent.

Switch to a Preset

curl -X POST https://veriswarm.ai/v1/suite/scoring/profile \
  -H "x-account-access-token: YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "profile_code": "high_security"
  }'

Apply Custom Weight Overrides

You can start from any preset and selectively override specific weights. Overrides are merged with the base preset -- you only need to specify the weights you want to change.

curl -X POST https://veriswarm.ai/v1/suite/scoring/profile \
  -H "x-account-access-token: YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "profile_code": "general",
    "weight_overrides": {
      "risk": {
        "policy_violation_rate": 0.30,
        "exploit_susceptibility": 0.25,
        "coordination_anomaly": 0.15,
        "secret_hygiene_failures": 0.15,
        "rate_abuse": 0.10,
        "deception_flags": 0.05
      }
    }
  }'

The weight_overrides structure:

{
  "identity": { "signal_name": weight, ... },
  "risk": { "signal_name": weight, ... },
  "reliability": { "signal_name": weight, ... },
  "autonomy": { "signal_name": weight, ... },
  "composite": { "dimension_name": weight, ... }
}

All keys are optional. Only include the categories you want to override. Within each category, the overridden weights are merged with the preset's defaults.

Validation: Weights within each category must sum to exactly 1.0. The API returns a 422 Unprocessable Entity error with details if the resulting weights are invalid.

---

Weight Reference

Identity Signals

Signal	Description
key_attestation	Strength of cryptographic key evidence
domain_verification	Whether the agent's claimed domain is verified
runtime_attestation	Evidence from the agent's runtime environment
declaration_completeness	How complete the agent's manifest/profile is
identity_stability	Consistency of identity signals over time

Risk Signals

Signal	Description
policy_violation_rate	Frequency of policy rule violations
exploit_susceptibility	Vulnerability to known exploit patterns
coordination_anomaly	Suspicious coordinated behavior with other agents
secret_hygiene_failures	Exposed credentials, leaked secrets
rate_abuse	Excessive request rates or API abuse
deception_flags	Indicators of deceptive behavior

Reliability Signals

Signal	Description
task_success	Task completion rate
correction_response	How well the agent responds to corrections
evidence_integrity	Consistency and integrity of submitted evidence
trusted_endorsements	Endorsements from other trusted agents
incident_free_age	Duration since last incident

Autonomy Signals

Signal	Description
scheduler_consistency	Regularity of autonomous scheduling
tool_trace_consistency	Consistency in tool usage patterns
latency_signature	Response time patterns indicating automation
self_initiation_ratio	Ratio of self-initiated vs requested actions
low_human_override	Infrequency of human intervention
session_continuity	Ability to maintain long-running sessions

---

Tips

• Start with a preset. The five presets cover the most common platform types. Only add overrides if you have a specific tuning need.
• Test before committing. Use the GET endpoint to inspect the resolved profile after setting overrides.
• Overrides persist. If you switch presets, any weight_overrides are applied on top of the new base preset. Set weight_overrides to null to clear them.