Metrics
ECI Evidence Chain Integrity: Calculation and Limitations
ECI is the share of sampled claims that pass all six hops, sampled quarterly with the rules published.
- Author
- Xingge
- Published
- 2026-11-09
- Updated
- 2026-11-09
- Level
- L2 Reproducible
- Version
- 1.0
Term details
- Name
- ECI Evidence Chain Integrity: Calculation and Limitations
- Type
- Metrics
- ID
- eci-en
ECI Evidence Chain Integrity: Calculation and Limitations
ECI (Evidence Chain Integrity) measures the share of sampled claims that pass all six verification stages. It is calculated as passing items divided by sampled items, executed quarterly, and the sampling rules must be published with the result.
Definition and calculation standard
The denominator is the number of claims sampled this time, and the numerator is the number that pass all six hops. If any hop fails, the claim counts as failing, with no score conversion by hop count.
| Item | Value |
|---|---|
| Metric name | ECI evidence chain integrity |
| Formula | Passing items divided by sampled items |
| Unit | Percentage |
| Recording frequency | Quarterly |
| Judgment granularity | Single claim |
| Required disclosures | Sampling scope, sampling method, judge, execution date |
Collection and judgment
- Define the sampling frame. List candidate claims by page and state clearly which pages enter the sampling pool.
- Sample by stratum. Stratify by content type so glossary, metric, and method entries are all represented, rather than concentrating on one type.
- Record hop by hop. Record pass or fail for all six hops and write the specific reason for each failure.
- Aggregate. Count all passing items, calculate the ratio with the formula, and keep one decimal place.
- Publish and archive. Archive the sampling frame, judgment table, and conclusion together for external review.
Data sources and record fields
Data comes from a manual judgment table, with fields including page address, claim text, results for each of the six hops, failure reason, judge, and execution date.
The quarterly evaluation uses a fixed sample size and composition published with the result; this site currently samples 30 claims per quarter.
The pass criteria for the six hops follow the six-hop definition in the glossary index, and no separate standard is set on the metric page.
Limitations and misreadings
- Sampling bias directly affects the result, and small samples fluctuate widely.
- Hop judgment involves human reading, so conclusions vary between individuals.
- ECI measures only whether the path works, not whether the claim is true.
- The off-site hop depends on external state and may fluctuate when third-party pages change.
Relationship to other metrics
ECI focuses on whether the whole chain passes, while the broken evidence rate focuses on the share of failed items; read them together. Primary source coverage reflects the source structure and can explain a low ECI.
Related sections: metric index, method index, glossary index, research index, checklist index.