Calibrating heuristics

PROTRAILBLAZER

AI Detectors · Signal Forensics
TOKENS0
AVG PRED.00
SENT VAR.00
REPETITION.00
FALSE POS
FALSE NEG
MODEL-LIKE SCORE0
Heuristic proxies · simulated
Specimen · analyzed locally, text never leaves this page
TAP A TOKEN TO INSPECT ITS PREDICTABILITY · BAR = HOW GUESSABLE THE WORD IS · RED = HIGH
Sentence rhythm · burstinessCV .00
CaveatThese signals are probabilistic proxies, not proof of authorship. False positives and false negatives are unavoidable at every threshold, so a score should never be sole evidence in an academic, employment, or disciplinary decision. Real detectors estimate perplexity with a language model; this page uses transparent heuristics, but the failure modes are shared.
SENSITIVITY TESTS
Detector · heuristic, not a product
0
MODEL-LIKE SCORE
±0 · illustrative
TRUTH: –
Signals · toggles reweight the score
.00
.00
.00
.00
Labeled demo set · synthetic
FLAGGED
PASSED
MODEL
TP
FN
HUMAN
FP
TN
RUN DATASET TO POPULATE OUTCOMES
CaveatThese signals are probabilistic proxies, not proof of authorship. False positives and false negatives are unavoidable at every threshold, so a score should never be sole evidence in an academic, employment, or disciplinary decision. Real detectors estimate perplexity with a language model; this page uses transparent heuristics, but the failure modes are shared.