Decision Output

Machine-Readable Response

{
  "decision": "ALLOW",
  "confidence": "HIGH",
  "constraints": [],
  "retryAfter": null,
  "ruleIds": ["allow_strong_builder"],
  "version": "v1"
}

Field	Description
decision	ALLOW, DENY, or ALLOW_WITH_LIMITS
confidence	LOW, MEDIUM, HIGH, or VERY_HIGH
ruleIds	Array of strict rule IDs that triggered the decision
explain	Human-readable array of reasons

Human-Readable Explanation

{
  "summary": "User is eligible",
  "explain": [
    "Expert-level builder credibility",
    "High social trust",
    "No negative trust indicators"
  ]
}

Confidence Mapping

Rules produce a confidenceDelta (numeric). The final output uses a categorical confidence tier.

function mapConfidence(numericConfidence: number): ConfidenceTier {
  if (numericConfidence >= 80) return "VERY_HIGH";
  if (numericConfidence >= 60) return "HIGH";
  if (numericConfidence >= 40) return "MEDIUM";
  return "LOW";
}

type ConfidenceTier = "LOW" | "MEDIUM" | "HIGH" | "VERY_HIGH";

Note: confidenceDelta in rules is additive to a base of 50. The final numeric value is mapped to a tier for the API response.

Decision Logging (Minimal & Optional)

BaseCred does not log score history.

Optionally, systems may log decision metadata only:

interface DecisionLog {
  subjectHash: string; // hashed identity
  context: string;
  decision: "ALLOW" | "DENY" | "ALLOW_WITH_LIMITS";
  confidence: ConfidenceTier;
  ruleIds: string[];
  signalCoverage: number;
  timestamp: number;
}

This enables audits, analytics, and rule evaluation — without storing raw scores or signals.

Constraints Reference

When a decision is ALLOW_WITH_LIMITS, one or more constraints are returned based on the Rule ID.

Rule ID	Constraints Applied	Meaning
probation_inactive	reduced_access, activity_required	User needs to be active/verify again.
probation_new_user	probation_period, limited_actions	New user trial period.
probation_mixed_signals	review_required	Manual review flagged.
limit_partial_signals	reduced_access	Limited features due to data gaps.
limit_comment_new	rate_limited	E.g. 5 comments/hour.
limit_publish_unverified	review_queue	Content goes to mod queue.
limit_governance_inactive	reduced_weight	Vote counts for less.