Trust Quotient™ v2.0

Every memory stored in Synapse Layer receives a Trust Quotient (TQ) score between 0.0 and 1.0. This score determines how memories are ranked during recall.

On this page

What is TQ?

Trust Quotient (TQ) is a per-memory confidence score computed at store time. It represents how much the system should "trust" a given memory when deciding what to surface during recall.

TQ is not a static value — it can evolve over the memory’s lifecycle based on access patterns, consistency signals, and agent feedback.

The score is returned in every StoreResultand RecallResult, making it fully observable to the calling agent.

Scoring Factors

TQ v2.0 is computed from multiple weighted factors:

FactorDescription
ConfidenceThe agent’s self-reported confidence when storing the memory (0.0–1.0)
RecencyTime decay — newer memories score higher, with configurable decay curve
FrequencyHow often the memory has been accessed (recall count)
SourceSource type classification from the Intent Validation stage

Note

The specific weights assigned to each factor are proprietary and dynamically calibrated. The full algorithm is available under Enterprise license.

How TQ Affects Recall

TQ directly influences recall behavior in two ways:

Priority Recall Mode

When recall_mode="priority" is used, memories are sorted by TQ score (highest first). This surfaces the most trusted memories regardless of semantic similarity or recency.

Hybrid Recall Mode

In recall_mode="hybrid", TQ is one of the ranking signals combined with semantic similarity and temporal recency to produce a blended result set.

Observability

TQ is designed to be transparent to agents:

  • Returned in StoreResult.trust_quotient after every store
  • Returned in RecallResult.trust_quotient for each recalled memory
  • Visible in the Forge Dashboard under each memory’s detail view
  • Stored as tqScore and tqVersion in the database

Boundaries

What TQ is NOT:

  • Not a factual accuracy score — TQ reflects system confidence, not truth
  • Not immutable — scores can change as access patterns evolve
  • Not comparable across tenants — scores are meaningful only within a tenant context
  • Not a guarantee of recall order — other recall modes may not use TQ at all

Next Steps

Intent Engine

How memories are classified into intent categories.

Differential Privacy

The mathematics behind embedding noise injection.