What Are Hidden Signals in Documents?
Every document tells two stories: the one the author intended, and the one hidden in its structure. The second story is where the most valuable — and most dangerous — information lives.
What counts as a hidden signal?
A hidden signal is anything that deviates from what a careful reader would expect, but that's easy to miss at human reading speed. Common categories include:
- Numerical anomalies — a value that's an order of magnitude off from its peers
- Contradictions — two statements that can't both be true
- Missing references — a "see Appendix B" with no Appendix B
- Repeated entities — one name or topic that dominates the narrative
- Terminology drift — a term whose meaning quietly shifts
Why humans miss them
People read linearly and trust context. A figure on page 3 and a contradicting figure on page 14 rarely collide in working memory. Statistical methods don't have that limitation — they hold the whole document at once.
How detection works
Hidden In Numbers runs eight independent detectors. Each is a small, transparent algorithm: Z-score and IQR for outliers, Jaccard similarity for duplication, rule engines for contradictions. No language model, no guessing — just math you can audit.
The result is a ranked list of findings, each tied back to the exact evidence in your document.