git.stella-ops.org/docs/VEXER_SCORRING.md

Status

This document tracks the future-looking risk scoring model for Vexer. The calculation below is not active yet; Sprint 7 work will add the required schema fields, policy controls, and services. Until that ships, Vexer emits consensus statuses without numeric scores.

Scoring model (target state)

S = Gate(VEX_status) × W_trust(source) × [Severity_base × (1 + α·KEV + β·EPSS)]

• Gate(VEX_status): affected/under_investigation → 1, not_affected/fixed → 0. A trusted “not affected” or “fixed” still zeroes the score.
• W_trust(source): normalized policy weight (baseline 0‒1). Policies may opt into >1 boosts for signed vendor feeds once Phase 1 closes.
• Severity_base: canonical numeric severity from Feedser (CVSS or org-defined scale).
• KEV flag: 0/1 boost when CISA Known Exploited Vulnerabilities applies.
• EPSS: probability [0,1]; bounded multiplier.
• α, β: configurable coefficients (default α=0.25, β=0.5) stored in policy.

Safeguards: freeze boosts when product identity is unknown, clamp outputs ≥0, and log every factor in the audit trail.

Implementation roadmap

Phase	Scope	Artifacts
Phase 1 – Schema foundations	Extend Vexer consensus/claims and Feedser canonical advisories with severity, KEV, EPSS, and expose α/β + weight ceilings in policy.	Sprint 7 tasks VEXER-CORE-02-001, VEXER-POLICY-02-001, VEXER-STORAGE-02-001, FEEDCORE-ENGINE-07-001.
Phase 2 – Deterministic score engine	Implement a scoring component that executes alongside consensus and persists score envelopes with hashes.	Planned task VEXER-CORE-02-002 (backlog).
Phase 3 – Surfacing & enforcement	Expose scores via WebService/CLI, integrate with Feedser noise priors, and enforce policy-based suppressions.	To be scheduled after Phase 2.

Data model (after Phase 1)

{
  "vulnerabilityId": "CVE-2025-12345",
  "product": "pkg:name@version",
  "consensus": {
    "status": "affected",
    "policyRevisionId": "rev-12",
    "policyDigest": "0D9AEC…"
  },
  "signals": {
    "severity": {"scheme": "CVSS:3.1", "score": 7.5},
    "kev": true,
    "epss": 0.40
  },
  "policy": {
    "weight": 1.15,
    "alpha": 0.25,
    "beta": 0.5
  },
  "score": {
    "value": 10.8,
    "generatedAt": "2025-11-05T14:12:30Z",
    "audit": [
      "gate:affected",
      "weight:1.15",
      "severity:7.5",
      "kev:1",
      "epss:0.40"
    ]
  }
}

Operational guidance

• Inputs: Feedser delivers severity/KEV/EPSS via the advisory event log; Vexer connectors load VEX statements. Policy owns trust tiers and coefficients.
• Processing: the scoring engine (Phase 2) runs next to consensus, storing results with deterministic hashes so exports and attestations can reference them.
• Consumption: WebService/CLI will return consensus plus score; scanners may suppress findings only when policy-authorized VEX gating and signed score envelopes agree.

Pseudocode (Phase 2 preview)

def risk_score(gate, weight, severity, kev, epss, alpha, beta, freeze_boosts=False):
    if gate == 0:
        return 0
    if freeze_boosts:
        kev, epss = 0, 0
    boost = 1 + alpha * kev + beta * epss
    return max(0, weight * severity * boost)

FAQ

• Can operators opt out? Set α=β=0 or keep weights ≤1.0 via policy.
• What about missing signals? Treat them as zero and log the omission.
• When will this ship? Phase 1 is planned for Sprint 7; later phases depend on connector coverage and attestation delivery.