
ML Systems · Circular economy through construction · the reading list
Take the house apart on purpose. Put the material back into the next one.
Most teardowns end in a landfill. ML Systems deconstructs instead of demolishing, then rebuilds larger from the recovered material. The order is Reuse › Resale › Recycle, and every recovered member becomes a claim on the house record — the salvage bank. One record per (property, cycle). Cycle N+1 underwrites against the verified record of Cycle N. Physical improvement, not market timing.
Where the order comes from: I Stole RRR from Elon Musk and SpaceX — the three letters are his, the order Reuse › Resale › Recycle is the work, and a recovery rate is not a reusability rate.
This page is what we found on Hugging Face that bears on that loop, with a line on why each one matters to it, and a short list of what does not exist here yet. Labels on every claim of ours: MEASURED exists today, MODELED is a calibrated projection, ASPIRATIONAL is a goal not in system logic. No ML Systems deconstruction has been performed yet; the 80–90% recovery target is MODELED.
Demolition, deconstruction and what comes out of a building
The end of a building is where the material either becomes a claim or becomes waste.
- outpostvertical/Building-Materials-and-Demolition-Scrap SYNTHETIC
A synthetic computer-vision evaluation set for building materials and demolition scrap. Synthetic is the honest word: the scrap coming off a 1920s Rhode Island two-family has not been photographed at scale by anyone yet. - EmbedEarth/demolition-permits
Multi-city building-removal permit records. A demolition permit is the public record of a building that is about to stop existing — the earliest signal a deconstruction crew could act on. - french-open-data/realisation-de-diagnostics-plomb-termites-amiante-prealables-aux-travaux-de-demolition-partielle
French open data on lead, termite and asbestos diagnostics required before partial demolition. The hazard survey is the first claim on any deconstruction ledger; France publishes the shape of it. - Hygros-LCA/IFC-Bridge-LCA-KBOB
Life-cycle assessment pairs built on IFC (openBIM) model data against the Swiss KBOB material database. The closest thing found here to a machine-readable link between a building model and the carbon of its members. - paper arXiv:2506.11061
Probabilistic Assessment of Engineered Timber Reusability after Moisture Exposure — the reuse question asked properly: is this member still fit to carry load, stated as a probability, not a mood. - paper arXiv:2310.18288
Sustainable Concrete via Bayesian Optimization — the recycle end of Reuse › Resale › Recycle, treated as an optimization with measured outcomes. - paper arXiv:2405.10452
Exploring Public Attention in the Circular Economy through Topic Modeling — what people mean when they say circular economy, measured from what they write.
Construction sites, safety and cost
Deconstruction is construction run in reverse order; the same site, the same crews, the same hazards.
Buildings, floor plans and the model of the house
You cannot recover what you have not measured. The model of the house is where every claim gets its address.
Waste, recycling and the sorting line
Whatever is not reused or resold has to be sorted, and sorting is where computer vision already earns its keep.
Materials, climate, policy and the money
A circular loop is only as honest as its accounting. These are the accountants.
What does not exist here yet
Three things we looked for and did not find. Each is a dataset someone could publish; the labels below are ours.
A PUBLIC RECORD OF WHAT COMES OUT OF A HOUSE
Not synthetic scrap. The members of a real wood-frame house, one row each, with who measured it, what it was, what condition it was in and where it went — Reuse, Resale or Recycle. Every value carrying its author and an evidence grade. ML Systems is designed to produce exactly this record and has not yet run a deconstruction; when it does, the record is the asset, and the labeled parts of it will be published.
THE TASK SEQUENCE OF TAKING A HOUSE APART
Construction has task-sequence data; deconstruction has almost none. The ML Systems ontology began as three blue notepads — two-letter task codes, a definition the first time each appeared, a score at the bottom of the day — and is documented as a collective ontology in the public repo. The deconstruction sequence that mirrors it is MODELED until a crew has run it.
A GRADED VOCABULARY, NOT A CLASSIFIER
Most of what is here classifies images. The loop needs something upstream of that: an agreed vocabulary in which the estimator, the lender, the appraiser, the town record and the crew describe the same building without contradiction. That is the Master Ledger's job, and it is written in English before it is written in code.
Read the system: MLSystemsRI/ml-systems-public · the writing: Insights · the company: huggingface.co/MLsystems. If you publish something that belongs on this list, tell us: salparvez@mlsystemsri.com.