---
title: "How memory accuracy degrades with scale"
description: "Memory accuracy measured at 100K, 500K, 1M and 10M events: the published splits, which abilities hold at scale, and where systems actually lose accuracy."
canonical: https://past.dev/blog/memory-accuracy-at-scale
date: 2026-09-01
category: Research
authors: The past.dev team
---
# How memory accuracy degrades with scale

Memory accuracy is usually reported at one corpus size, and the number tells you little about the size you will actually run at. We publish the full curve. On BEAM, past.dev scores 91.50% at 100K events, 87.97% at 500K, 86.18% at 1M, and 82.61% at 10M, each measured on the complete split. This post reports where those points come from and, more usefully, which abilities lose accuracy as history grows and which do not.

## The published splits

Every figure below comes from the run of 26 August 2026. Each scale is a complete split rather than a sampled slice, and each row lists the question and conversation counts behind it, because a score is only checkable when its denominator is visible.

| Scale | Overall | Questions | Conversations |
|---|---|---|---|
| 100K events | 91.50% | 400 | 20 |
| 500K events | 87.97% | 700 | 35 |
| 1M events | 86.18% | 700 | 35 |
| 10M events | 82.61% | 200 | 10 |

The [benchmarks page](/benchmarks) carries the same data with the per-ability breakdown at every scale, alongside the published results other systems report for themselves. mem0, for comparison, reports 64.1% at 1M and 48.6% at 10M in [its own benchmark writeup](https://mem0.ai/blog/ai-memory-benchmarks-in-2026).

## Where memory accuracy is lost

The overall curve drops 8.89 points across a 100x increase in history. The drop is concentrated in specific abilities. Averages hide this, so here is the profile.

**What holds at scale.** Event ordering stays at 99.3% or above at every scale we measure, including 100% at 10M (n=20). Preference following stays at 96.55% or above everywhere. Abstention, the ability to decline when the stored evidence does not establish an answer, is 100% at 100K and still 92.5% at 10M. These abilities depend on the temporal structure of storage rather than on search over a larger haystack, which is why scale barely touches them.

**What degrades.** Knowledge update, answering with the current value after a fact changed, falls from 75.63% at 100K to 62.14% at 1M. Multi-session reasoning, combining evidence spread across separate conversations, falls from 74.97% at 100K to 43% at 10M (n=20), the steepest decline in the table. Contradiction resolution sits in the low seventies at every scale. These three are the frontier: they are hard at every size and harder with more history.

One caveat we would want stated if someone else published this table: the 10M split has 20 questions per ability category, against 70 at 1M. A category score at n=20 is a coarser measurement, and single-category swings at 10M should be read with that in mind.

## What this means if you are evaluating memory

Ask any memory vendor, including us, three things. First, the full curve rather than one point: a system tuned for a demo corpus can score well at 100K and collapse later, and the [500K and 1M splits](/benchmarks) are where degradation becomes visible. Second, complete splits with visible denominators, since sampled slices and hidden n make scores unfalsifiable. Third, the per-ability profile, because your workload is not an average: an agent that mostly needs event ordering has different risk than one that lives on knowledge updates.

The abilities that degrade are exactly the ones that depend on [tracking facts that change over time](/guides/facts-that-change-over-time): knowledge updates need [supersession](/glossary/fact-supersession), and multi-session reasoning needs [entity resolution](/guides/entity-resolution) to connect the same people across conversations. Systems that treat memory as similarity search over chunks degrade fastest on precisely these questions.

The full methodology, judges, and configurations are documented on the methodology page. [How to reproduce this →](/benchmarks/methodology)