What is a trust envelope?
A trust envelope is a structured wrapper attached to every AI answer, carrying its confidence, sources, evidence, reasoning, data-mode, freshness, timestamp, and compliance status. It lets a caller judge how much to rely on the answer.
The answer plus its provenance
Most AI systems return an answer and nothing about where it came from. A trust envelope returns the answer together with its provenance in a fixed schema. That includes a confidence value and label, the sources it drew on, the evidence points behind it, a one-line reasoning, whether it ran on live or sample data, and a compliance block.
Because the schema is consistent, any consumer (a person, a web app, or an agent) can read the same fields the same way on every answer.
Why it serves every buyer at once
The same envelope serves retail trust, IR compliance, developer reliability, and enterprise auditability. A retail user sees confidence and sources; a compliance officer sees the no-advice and no-prediction checks; a developer sees the data-mode flag; an auditor gets a record that can be persisted and verified.
Honesty about the data is deliberate. The data-mode field reads sample whenever a mock provider is in play, so a demo is never mistaken for a live read.
How LyraMind implements it
Every LyraMind answer is built with a single build_trust function that assembles the envelope: a trust version, confidence and its label (high, moderate, low, or unknown), sources, filtered evidence, reasoning, data-mode, freshness, a generated-at UTC timestamp, and a compliance object. The compliance checks assert no advice, no prediction, and sources cited, plus a grounded flag.
Some surfaces override the compliance status for their context, for example an issuer-facing legibility scan marks itself editorial-legibility-no-advice, and the execution gate marks itself an execution gate that does not execute. The envelope is what the Trust Ledger persists for each recorded answer.
Related concepts
The trust envelope is the per-answer proof; the Trust Ledger is the persisted history of those envelopes. Grounded financial reasoning is the discipline the envelope makes visible, and the readiness gates read its fields directly.
- A fixed-schema wrapper on every answer: confidence, sources, evidence, reasoning
- Adds data-mode, freshness, a UTC timestamp, and a compliance block
- Confidence labels: high, moderate, low, or unknown
- Data-mode reads sample for mock providers, so demos are never mistaken for live
- Compliance checks assert no advice, no prediction, sources cited, plus a grounded flag
- Some surfaces override compliance status for their context (issuer, execution gate)
Questions
What is in the trust envelope?
Confidence and its label, sources, evidence, reasoning, data-mode, freshness, a timestamp, and a compliance block with grounded status and the checks it passed.
How does it show live versus sample data?
The data-mode field reads live or sample by inspecting the provider. A mock provider is honestly labeled sample so a demo is never mistaken for a live feed.
Is the trust envelope the same as the Trust Ledger?
No. The envelope is the proof attached to one answer. The ledger is the append-only, hash-chained history of those envelopes over time.