Skip to content
#edge computing Open access

The Minimum Swarm House: A material operating system that grows with its owner — the smallest swarm that keeps a house habitable, the joint that keeps its matter recoverable, and fifty-five years of one household measured on three ways of building the same plan

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

Three questions are asked of one runnable house and answered from logged runs. How few elements of a building need to reason before the building holds the life inside it? What has to be true of the joint between two pieces of matter for a plan to stay open for fifty years? And what happens to a house whose owner ages from twenty-five to eighty while the walls stay where a builder left them? One structured file declares 24 floor cells at 1.5 m, 207 modules, 58 edge slots, 12 sensors, 38 actuators and 8 activity contracts; five layers read that same file and read only it. Reasoning is then treated as a resource with a fixed total and placed five ways — on hubs, on districts of equal area, on partition lines, on the modules that carry services, and in every cell. At one shared budget of 960 operations a single element holds 0.630 of the day's contract-hours and thirty-two elements hold as little as 0.290, so thinner reasoning at equal compute buys less house. At one budget each, quality holds from one element to sixteen while resilience climbs from 0.000 to 0.847, which bounds the minimum between 4 and 8 reasoning elements for this plan, one for every 6.75 to 13.50 square metres. A survival ladder then removes intelligence rung by rung: a hosted model holds 0.696, one computer in the house holds 0.696 at 6.00 seconds less latency, district elements hold 0.609, a fixed table of rules holds 0.261, and bare matter with the power off holds 0.261 while keeping 54.00 square metres of sheltered floor, a 6.00 m span, 6 openable edges and 1.10 air changes an hour. Safe occupancy holds at every rung. One rung is indistinguishable from having no reasoning at all: elements that reach a single neighbour hold 0.290 against the fixed table's 0.261, with a spread of 0.082 across seeds and one seed reading 0.174, because a local commitment forecloses a later demand. Reversibility is then graded off the file into four classes of joint, and a fixed family of 60 future demands is asked of four histories of the same plan: as built it meets 0.667, after four service modules leave 0.650, and after eight edges are sealed 0.067 — a fall of 10 times, which locates future value in the joint rather than in the equipment. Ten life states from age twenty-five to eighty are then run on three construction systems holding identical geometry. A cast system holds 0.395 of the demand, spends 256 304 euro on change, discards 1 443 kg and loses 500.88 days; an actuated system holds 0.709, spends 24 euro, discards 0 kg and loses 0.35 days. The sharpest result belongs to neither intelligence nor reversibility: at 850 mm clear openings the state in which an occupant uses a wheelchair reads 0.00 on every system, and at 900 mm it reads 0.76, while over the whole trajectory the actuated household moves house 1 time at 850 mm and 0 times at 900 mm, losing 62.34 days against 0.35 and spending 34 024 euro against 24. Fifty millimetres chosen once, fifty years early, outweighs every layer above it. The paper closes with Swarm Room 0: nine cells, 14 priced lines, 11 995 euro, 592 euro a square metre, reasoning at 2.91 per cent of the build, and 15 acceptance tests. Melo de Magalhães, Planta Smart Homes, 2026 1 What this paper settles – Minimum reasoning is an empirical quantity rather than a design taste. For this plan it sits between 4 and 8 elements, one for every 6.75 to 13.50 square metres, and the criterion is a pair of separate tests on quality, tolerance and command rather than a weighted score. – More intelligence is worse under a fixed compute budget, and flat under a per-element budget. What extra elements actually buy is tolerance of failure: resilience rises monotonically from 0.000 at one element to 0.847 at sixteen, and no single failure at all is absorbed at one element. – Partial intelligence is the dangerous regime. A house whose elements reach one neighbour reads 0.290 against 0.261 for a house with no reasoning at all, and less than half of the 0.609 the full district layer delivers. – Safe occupancy is separable from intelligence and holds with the power off, because release is mechanical: springs and gravity rather than code. – A joint decides the future. Sealing eight edges costs 10 times more future passibility than losing four service modules. – Rooms are guesses. Conditions replanned each hour hold 0.696 of the day against 0.261 for one fixed allocation of named rooms, a factor of 2.67. – Labour, rather than actuation, is where the expense of Author ORCID: 0009-0008-6255-7724. License: Creative Commons Attribution 4.0 International.

View source

Similar papers

#computer vision Review Sep 2017

Agile Software Development Methods: Review and Analysis

This publication proposes a definition and a classification of agile software development approaches and analyses ten software development methods that can be characterized as being "agile" against the defined criterion.

P. Abrahamsson, O. Salo, Jussi Ronkainen et al. · 727 citations · ⚡54
#computer vision Jun 2008

The impact of agile practices on communication in software development

The study shows that agile practices improve both informal and formal communication, but indicates that, in larger development situations involving multiple external stakeholders, a mismatch of adequate communication mechanisms can sometimes even hinder the communication.

M. Pikkarainen, Jukka Haikara, O. Salo et al. · 401 citations · ⚡48
#machine learning Review Open access Oct 2014

Software development in startup companies: A systematic mapping study

The results indicate that software engineering work practices are chosen opportunistically, adapted and configured to provide value under the constrains imposed by the startup context.

Nicolò Paternoster, Carmine Giardino, M. Unterkalmsteiner et al. · 394 citations · ⚡54

Related blog posts

Microsoft Research Blog Sep 29, 2026

Introducing Quine: An AI research system designed for the complexity of biology

Biology doesn't operate in silos, and neither should the AI representation of it. Quine is an early-stage research effort to create a multimodal world model of biology. By connecting insights across biological scales and modalities, Quine helps scientists computationally search a space far larger than intuition allows and prioritize hypotheses before they reach the lab. Experimental results provide important feedback, helping researchers sharpen future research directions. The post Introducing Q…

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.