The Geometry of Sameness: Riemannian Equivalence of Translation and Distance for Semantic Detection (Geometric Intelligence Book 2)

When is something really the same?

That question sits at the heart of every face recognition system, every fraud detector, every platform that decides whether two records represent one person or two. Get it wrong and someone loses access to their account, gets flagged as a threat, or disappears into a bureaucratic black hole.

Most machine learning systems answer this question twice: once with translators stitching together messy real-world data, and once with clean geometric models that look beautiful in papers but never touch production. Those two answers almost never talk to each other.

Until now.

The Geometry of Sameness proves that these two approaches—translator graphs and Riemannian manifolds—are secretly telling the same story, as long as you're willing to measure how smoothly they stitch together. The result: error guarantees that survive coordinate changes, detection systems you can audit across representations, and accountability when the schemas drift and someone swaps out your encoder at 3am.

What's inside:

This is a mathematics book. It contains theorems, proofs, and explicit error bounds. But it's mathematics in service of real problems—written by someone who has spent twenty-seven years in places where getting sameness wrong means someone doesn't come home.

✦ Riemannian foundations : Manifolds, geodesics, parallel transport, and curvature—built from first principles for readers who want rigor without hand-waving

✦ Translator graph theory : How production ML systems actually stitch representations together, and why the geometry hiding in those stitches matters

✦ The Equivalence Theorem : The core result proving that bounded translator distortion and bounded geometric curvature are two views of the same constraint

✦ Explicit slack terms : Every bound in this book comes with explicit constants. No there exists an ε without telling you what ε actually is.

✦ Four domain case studies : Identity verification, payment reconciliation, telemetry drift detection, and content moderation—showing how the theory applies where it counts

✦ Auditable guarantees : How to construct detection systems that regulators and survivors can actually trace, not just the model said so

Who this book is for:

→ ML engineers who want invariant guarantees that hold when encoders change, schemas drift, and production gets messy

→ Mathematicians who want their differential geometry and category theory applied to systems that actually run

→ Security researchers building detection systems that need to work across representation boundaries

→ Technical leaders who need to explain—to regulators, to courts, to survivors—exactly how their system decided two things were the same

→ Anyone who has asked: are we still talking about the same thing after we changed the code?

From the author:

I wrote this book because I kept watching systems fail in ways that the math should have prevented. Face recognition that couldn't survive a camera change. Fraud detection that broke when someone updated the embedding model. Identity systems that confidently declared two people were the same—or different—with no traceable justification.

The geometry was always there. The translator graphs were always there. No one had proven they were the same theory.

Now they are. And now you can build systems that know—mathematically, auditably, defensibly—when sameness holds and when it doesn't.

—Bee Davis

Part of The Geometry Intelligence series.

Bee Rosa Davis is Principal Adversarial Intelligence Engineer at IBM X-Force Red, a NASA contractor, and the inventor of the Davis Topological Processor. She holds an M.S. in Digital Forensics from Brown University and serves as Overall Mother of the Kiki House of Diwa. Six patents pending (2025).