Evidence
In Practice
● ILLUSTRATIVE CASE STUDY
Inside The Decision
Description
FICTIONAL ISSUER · NOT A CUSTOMER RESULT
Northstar Systems
Q2 earnings release
The illustrative release states: “Diluted earnings per share were 2.11. Revenue was 1.28 billion.” Five model classes review the same two fields.
| Model class | EPS | Revenue |
|---|---|---|
| Large generalist A | 2.11 | 1.28B |
| Large generalist B | 2.11 | 1.28B |
| Document specialist | 2.11 | 1.28B |
| Small fast model | 2.11 | 1.28B |
| Dissenter | 2.10 | 1.28B |
80% agreement.
The dissent stays visible.
With an EPS tolerance of 0.005, the 2.10 reading is outside the majority cluster. Four of five jurors agree on EPS; all five agree on revenue. The sample’s headline agreement is 80%.
The planned production record includes document and model-set hashes, evidence references and attestation commitments. This example uses sample data and does not issue an attestation or chain transaction.
Open this sample →What The Split Changes
Description
The minority reading directs further review to three areas: the cited source, the rounding convention and the schema tolerance. The agreement rate alone does not establish accuracy.
If fewer than four jurors agreed on a required field, the result would remain HUNG. The feed would retain its last accepted verdict, with its original timestamp, pending resolution.
Inspect a HUNG example →
Work With Evidence