Evidence Audit
Session jdg-2026053001 — all evidence nodes
| Domain | Evidence ID | Claim | Bayes K | Weight | Source | Ingested | Status | Actions | |
|---|---|---|---|---|---|---|---|---|---|
| linguistic | evd-001 | Perplexity score of 2.3 significantly below human baseline of 12.7, indicating low token-level surprise consistent with LLM generation patterns. | 142.7Decisive | 0.88 | automated | 2026-05-30 12:44:11 | Clean | ||
| statistical | evd-002 | Burstiness index of 0.12 falls in the bottom 8th percentile of human writing samples, consistent with uniform LLM output cadence. | 87.4Strong | 0.72 | automated | 2026-05-30 12:44:23 | Clean | ||
| statistical | evd-009 | Zipf law deviation coefficient 0.18 indicates highly regular word frequency distribution, inconsistent with natural human authorship. | 74.6Strong | 0.70 | automated | 2026-05-30 12:46:44 | Clean | ||
| syntactic | evd-005 | Sentence length variance of 4.2 characters std-dev is atypically low; human writers average 18.3 std-dev, strongly indicating LLM output. | 63.1Strong | 0.65 | automated | 2026-05-30 12:45:02 | Clean | ||
| structural | evd-012 | Paragraph length distribution entropy 2.14 bits is lower than human baseline of 2.89 bits, consistent with templated LLM generation. | 51.8Strong | 0.62 | automated | 2026-05-30 12:47:55 | Clean | ||
| semantic | evd-003 | Semantic coherence score 0.94 across paragraph transitions exceeds typical human writing (0.71) — consistent with both LLM and highly-edited human text. | 34.2Moderate | 0.54 | automated | 2026-05-30 12:44:38 | Clean | ||
| lexical | evd-011 | Rare word usage (frequency < 1/10M) at 0.8% of tokens is within LLM generation range; GPT-4 averages 0.6–1.1%. | 29.3Moderate | 0.48 | automated | 2026-05-30 12:47:28 | Clean | ||
| structural | evd-006 | Repetition pattern frequency 0.08 per 100 tokens; human prose averages 0.31. Low repetition is characteristic of GPT-class models. | 22.9Moderate | 0.45 | automated | 2026-05-30 12:45:14 | Clean |