Trust Signals
Trust signals are normalized, queryable representations of verified activity derived from trust events.
Signal derivation
Every signal MUST trace to a source event except governed manual attestations recorded as events first.
Signal anatomy
| Field | Purpose |
|---|---|
signal_id | Unique identifier |
signal_type | Catalogued classification |
pti_id | Affected subject |
context_id | Life-area scope |
polarity | positive, negative, neutral |
weight | Normalized influence (0.0–1.0) |
source_event_id | Provenance anchor |
effective_at | When signal becomes active |
expires_at | Optional TTL |
Signal types (examples)
| signal_type | context_id | Typical polarity |
|---|---|---|
repayment.on_time | lending | positive |
repayment.delinquent_30d | lending | negative |
lease.completed | rental | positive |
employment.tenure_12m | employment | positive |
chargeback.opened | merchant | negative |
endorsement.peer | informal_sector | positive |
Normalization rules
The event normalizer applies:
- Type mapping: partner payload → canonical
signal_type - Polarity assignment: based on event outcome fields
- Weight calibration: per context policy tables
- Deduplication: collapse duplicate business actions via
idempotency_key
Context binding
Signals are strictly context-bound. A merchant chargeback signal MUST NOT directly alter lending scores unless a published lens derivation rule explicitly allows cross-context aggregation.
Decay and aggregation
| Mechanism | Description |
|---|---|
| Time decay | Older signals contribute less to scores |
| Recency boost | Recent positive activity may accelerate recovery |
| Contradiction handling | Negative signals offset positives within same domain |
| Floor/ceiling | Context policies cap single-signal contribution |
Signal quality indicators
The intelligence engine tracks:
- Freshness: time since latest signal
- Diversity: count of distinct producers
- Verification depth: document vs self-reported
- Consistency: contradictory signal detection
These feed coverage_gaps and confidence bands in lookup outcomes.
Anti-poisoning
Exchange policy SHOULD detect:
- Abnormal volume spikes from single producer
- Context mismatch patterns
- Circular endorsement rings
Suspicious signals MAY be quarantined pending review.