Trust Scoring

Every SignalStack verification returns a trust_score — a 0.0 to 1.0 value representing the system's confidence in the result. This score is computed from four weighted dimensions, each contributing a sub-score.

The four dimensions

Dimension	Weight	Description
Source Authority	35%	Evaluates the credibility of each source based on domain authority, citation count, and editorial standards.
Cross-Reference	30%	Measures agreement across independent sources. High agreement = higher confidence.
Recency	15%	Favors newer sources. Information older than 1 year is penalized; older than 5 years heavily discounted.
LLM Analysis	20%	Uses a specialized LLM to evaluate logical consistency, contradiction detection, and semantic alignment.

Score thresholds and verdicts

Score Range	Verdict	Meaning
0.80 – 1.00	`"true"`	Strongly supported by multiple authoritative sources
0.60 – 0.79	`"likely_true"`	Supported but with some caveats or weaker sources
0.40 – 0.59	`"unverifiable"`	Insufficient evidence to confirm or refute
0.20 – 0.39	`"conflicting"`	Sources disagree or evidence is contradictory
0.00 – 0.19	`"false"`	Strongly contradicted by authoritative sources

Evidence chain

Each verification response includes an evidence array with the sources used. Every evidence item contains:

Field	Type	Description
`source`	string	Name of the data source (e.g. "Wikipedia", "SEC EDGAR")
`url`	string	Direct URL to the supporting page or document
`snippet`	string	Relevant excerpt from the source
`supports_claim`	boolean	Whether this source supports or refutes the claim
`authority_score`	number	0.0–1.0 authority rating for this specific source
`retrieved_at`	string (ISO-8601)	Timestamp when the source was accessed

Data sources

SignalStack aggregates data from 30+ sources:

Source	Type	Endpoints
Companies House	Government register	Business
SEC EDGAR	Government filings	Business
GLEIF	Legal entity identifier	Business
Companies House	Government register	Business
SEC EDGAR	Government filings	Business
GLEIF	Legal entity identifier	Business
Sanctions Network	Free OFAC/UN/EU sanctions	Business
VAT BusinessPress	Free EU VAT validation	Business
libphonenumber	Free phone validation	Business
OpenCorporates	Corporate database	Enterprise
OpenSanctions	Sanctions screening	Business
Abstract API	Company enrichment	Business
Google Fact Check Tools	Claim database	Claim
Wikipedia	Encyclopedic	Claim
GDELT	News event database	Claim, Media
World News API	News articles	Claim
Groq	LLM inference	All

Custom scoring (Business & Enterprise)

Business and Enterprise plans can customize scoring weights and thresholds via the dashboard or API:

// Custom scoring configuration via API
{
  "weights": {
    "source_authority": 0.40,
    "cross_reference": 0.25,
    "recency": 0.10,
    "llm_analysis": 0.25
  },
  "thresholds": {
    "true": 0.75,
    "likely_true": 0.50,
    "unverifiable": 0.35,
    "conflicting": 0.20,
    "false": 0.00
  }
}

# Python
client.configure_scoring(
    weights={"source_authority": 0.40, "cross_reference": 0.25},
    thresholds={"true": 0.75}
)

Interpreting scores in your agent

# Python — make agent decisions based on trust score
result = client.verify.claim(claim="...")

if result.trust_score >= 0.80:
    agent.proceed()          # High confidence, take action
elif result.trust_score >= 0.40:
    agent.ask_clarification()  # Uncertain, ask user
else:
    agent.refute()           # Likely false, do not act

Next steps

Explore the API Reference for endpoint details
Learn about Webhooks for async verification
Check Error Handling