STEALTHGPT SUPER VS PANGRAM 4.1 — NOVEMBER 15, 2026

Nine AI-written original passages. One StealthGPT Super rewrite per passage.
All 18 before/after scans were retained. Seven first outputs received Human Written
(0% AI, 0% AI-assisted, 100% human). One received Mostly Human (5% AI, 95% human).
One remained AI Generated (100% AI). Every original was classified 100% AI.
No retries and no manual edits were used before the output scans.

Files
- results.csv: every before/after classification and Pangram word count.
- corpus.json: complete exact texts, SHA-256 hashes, settings, method and results.
- texts/: exact originals and first Super outputs, saved without editorial edits.
- screenshots/: native captures of all nine rewritten results and one original.

All passages are synthetic writing fixtures. Business anecdotes and academic
studies, researchers and findings are invented for the test. No real author's
published writing, confidential customer material or private research was used.
This is a StealthGPT-run demonstration, not an independent or representative study.
Detector shares are text classifications, not authorship probabilities.
Rewrites may change details or inferences; review the full passages before reuse.

Publication hub: https://www.stealthgpt.ai/blog/ai-humanizer-benchmark-pangram
Product page: https://www.stealthgpt.ai/bypass/pangram
