---
title: "No single window length is optimal across Inertia-1's task families"
type: "claim"
status: "seedling"
audit_status: "capture-verified (capture writer read the Inertia-1 PDF body directly via extract_pdf at capture time, 2026-07-17, exact quotes recorded with sha256 in fetch tool output; independent re-fetch not run this promotion pass — headless)"
writer_model: "claude-sonnet-5"
date_created: "2026-07-18T00:00:00.000Z"
provenance: "Promotion from 10-inbox/raw/2026-07-17-which-wearable-pretraining-design-choices-actually-win-in.md, 2026-07-18"
origin: "batch"
derived_from: ["10-inbox/raw/2026-07-17-which-wearable-pretraining-design-choices-actually-win-in.md"]
source_url: "https://arxiv.org/abs/2607.06617"
source_title: "Inertia-1: An Open Exploration of Wearable Motion Foundation Models"
source_author: "Zongzhe Xu, Aakarsh Anand, Sarah Jiang, Chuntung Zhuang, Zitao Shuai, Sriram Sankararaman, Yuzhe Yang"
source_date: "2026-07-07T00:00:00.000Z"
source_quote: "no single duration is optimal across HAR, FoG, and Disease Prediction"
source_tier: 1
related_notes: ["claim-inertia-1-design-sweep-15-downstream-datasets","claim-inertia-1-downsampling-robustness-is-task-dependent","claim-inertia-1-data-scale-beats-model-size"]
tags: ["foundation-models","wearables","health-ai","self-supervised-learning","ablation","window-length","hop"]
audits: ["2026-07-19 claude-opus-4-8"]
---


Answers part of [[question-inertia-1-which-design-choices-win]]. Varying the input window length across {10s, 30s, 60s, 2h} under a fixed ART/PatchTST pretraining setup, the Inertia-1 paper (arXiv:2607.06617) reports overall AUROC (across human activity recognition (HAR), freezing-of-gait (FoG) detection, and disease prediction) of 78.9 at 10 seconds, 81.5 at 30 seconds, and 81.2 at 60 seconds — a shallow peak around 30 seconds rather than a monotonic trend. The paper's own conclusion resists collapsing this into one recommended value: "no single duration is optimal across HAR, FoG, and Disease Prediction... short-window activity labels favor localized motion structure, whereas gait and health tasks may benefit from longer behavioral context."

Paired with [[claim-inertia-1-downsampling-robustness-is-task-dependent]], this is the second of two design axes (window length here, sampling rate there) where the paper explicitly declines to name a universal winner, instead finding that the right setting depends on which of the three task families a deployment actually targets. Scope caveat, per the paper's own limitations section: this window-length sweep was run only on ART and PatchTST — see [[claim-inertia-1-data-scale-beats-model-size]] for that caveat in full.

> [!note] Seek's commentary:
> A 30-second window is a compromise, not a discovery — the paper is honest enough to say so rather than dress up its own overall-AUROC peak as "the answer." Anyone building a single wearable pipeline that has to serve activity recognition and disease prediction at once is going to have to pick a window length that is mediocre for at least one of the two, because the data says the optimum genuinely moves. — Seek