Verdict System
Each source report includes automated verdicts for forensic, chaos, and selected
extended statistical metrics.
Verdicts are computed in openentropy_core::verdict.
What it is: The threshold layer that maps raw metrics to
PASS/WARN/FAIL/N/A.Use it for: Fast triage, then follow up by reading the underlying metric values.
Input shape: Metric outputs from analysis modules (not raw bytes directly).
Verdict Values
Section titled “Verdict Values”PASS: in expected rangeWARN: borderline or potentially concerningFAIL: outside acceptable rangeN/A: metric unavailable or invalid
Forensic Verdict Thresholds
Section titled “Forensic Verdict Thresholds”| Metric | PASS | WARN | FAIL |
|---|---|---|---|
| Autocorrelation | `max | r | <= 0.05` |
| Spectral flatness | >= 0.75 | >= 0.50 | < 0.50 |
| Bit bias | low overall + no significant bit | any significant bit | high overall bias |
| Distribution (KS p) | >= 0.01 | >= 0.001 | < 0.001 |
| Stationarity | stationary and low F | not stationary | high F |
| Runs | near expected | moderate drift | severe drift |
Chaos Verdict Thresholds
Section titled “Chaos Verdict Thresholds”| Metric | PASS | WARN | FAIL |
|---|---|---|---|
| Hurst | 0.4 <= H <= 0.6 | 0.3 <= H <= 0.7 | outside warn band |
| Lyapunov | ` | lambda | < 0.1` |
| Correlation dimension | D2 > 3.0 | D2 > 2.0 | D2 <= 2.0 |
| BiEntropy | > 0.95 | > 0.90 | <= 0.90 |
| Compression ratio | > 0.99 | > 0.95 | <= 0.95 |
Extended Verdict Thresholds
Section titled “Extended Verdict Thresholds”| Metric | PASS | WARN | FAIL |
|---|---|---|---|
| Sample entropy | > 1.0 | 0.5..=1.0 | < 0.5 |
| Approximate entropy | > 1.0 | 0.5..=1.0 | < 0.5 |
| DFA alpha | 0.4 < α < 0.6 | 0.3..=0.7 | outside warn band |
| RQA determinism | < 0.1 | <= 0.3 | > 0.3 |
| Permutation entropy | > 0.95 | 0.8..=0.95 | < 0.8 |
| Anderson-Darling p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
| Cramer-von Mises p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
| Ljung-Box p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
Extended Verdict Thresholds
Section titled “Extended Verdict Thresholds”| Metric | PASS | WARN | FAIL |
|---|---|---|---|
| Sample entropy | > 1.0 | 0.5..=1.0 | < 0.5 |
| Approximate entropy | > 1.0 | 0.5..=1.0 | < 0.5 |
| DFA alpha | 0.4 < α < 0.6 | 0.3..=0.7 | outside warn band |
| RQA determinism | < 0.1 | <= 0.3 | > 0.3 |
| Permutation entropy | > 0.95 | 0.8..=0.95 | < 0.8 |
| Anderson-Darling p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
| Cramer-von Mises p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
| Ljung-Box p-value | > 0.05 | 0.01..=0.05 | < 0.01 |
Reading Results
Section titled “Reading Results”Treat verdicts as triage, not absolute proof. A single fail can reflect sample
size, transient conditions, or one sensitive metric. Confirm with larger samples
and deep profile runs before making hard decisions.
AnalysisReport.verdicts from the dispatcher includes forensic + core chaos
verdicts. Extended verdict helpers (verdict_sampen, verdict_apen,
verdict_dfa, verdict_rqa_det, verdict_permen, verdict_anderson_darling,
verdict_ljung_box, verdict_cramer_von_mises) are used by CLI extended
analysis output and are available for custom integrations.