---
title: "A Measure-Theoretic Account of Semantic Caching: When Approx"
emoji: ♻️
colorFrom: indigo
colorTo: gray
sdk: static
app_file: index.html
pinned: false
license: cc-by-4.0
tags:
  - semantic-cache
  - llm-serving
  - caching
  - theory
---

# A Measure-Theoretic Account of Semantic Caching: When Approximate Reuse Is Provably Safe

A measure-theoretic safety criterion for semantic caches: the exact condition under which a cache hit cannot change the answer, and its false-reuse bound.

Authors: Agon Avdimetaj, TMLS AI Research. Published 2026-07-13.

## What this paper is, and what it is not

This paper contains **no experiments of our own.** We did not train, serve, or
benchmark any model for it. Every empirical number is cited to a published
source with a working URL. Every curve we construct is labeled as an
**analytical model** with its assumptions stated, and figures that replot a
published study name that study in the caption.

The **derivations and proofs are original work**, carried out in full and shown
step by step. That is the contribution.

## Corrections

If you find a wrong proof step, a mis-attributed number, or a citation that does
not say what we claim, please open a discussion. We will correct the paper and
say what changed.

## Related

Canonical version: [tmls.nyc/research/semantic-caching-measure-theory](https://tmls.nyc/research/semantic-caching-measure-theory)
All five papers in this series: [huggingface.co/TMLSNYC](https://huggingface.co/TMLSNYC)

Text licensed CC-BY-4.0.