Industries / Energy, water and waste / Environmental services / Recycling
Collection and contamination
Collection is how recyclables get from homes and shops to a sorting plant, and contamination is anything in that stream that the plant cannot sell as the material it is sorting for.
7
Milestones achieved, from 1995 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1995
Households sort, makers pay
Japan's Containers and Packaging Recycling Act of 1995 split the work three ways: households sort their waste, municipalities collect it in separate streams, and the companies that make or use containers pay for recycling them. The law took effect in 1997 and covers eight kinds of containers and packaging, among them PET bottles and glass bottles sorted into colorless, amber and other colors. Containers and packaging then made up about 60% of household waste by volume, according to the Japan Containers and Packaging Recycling Association (JCPRA), which runs the recycling on the companies' behalf.
Credit
Government of Japan; JCPRA
1997
Four partners share the work
Taiwan's Environmental Protection Administration began its 4-in-1 Recycling Program in January 1997. Residents separate their waste, private collectors buy recyclables, local governments run collection teams, and a fund paid by the makers and importers of new products subsidizes the collectors and recyclers that meet the agency's standards. A December 2012 case study published by the US Environmental Protection Agency reports that waste quantities in Taiwan fell from 1.14 kg per person per day in 1997 to 0.4 kg in 2011, and that the number of formal electronics recyclers rose from none in 1997 to 19 at the end of 2011.
Credit
Taiwan Environmental Protection Administration; its Recycling Fund Management Board
2000s
One cart for everything
In single-stream collection, households put paper, cans, plastic and glass into one cart, which is easier for them and suits automated trucks. In December 2009 the Container Recycling Institute (CRI) said many US municipalities had switched over the previous decade, and its report found that mixing the materials brings cross-contamination and broken glass that raise costs for paper mills, glass plants and plastics recyclers. A March 2021 University of Miami study for the REMADE Institute found that in one US region, loads arriving at sorting plants from single-stream programs averaged 25.9% contamination, against 10.9% from dual-stream programs, which collect paper in a separate container.
LessonCheaper collection can move cost downstream, and CRI said some systems were "merely shifting costs" to the mills and recyclers that buy the material.
Credit
US cities and haulers; CRI (report by Clarissa Morawski); Temitope Runsewe and Nurcin Celik (University of Miami)
2013
Bogotá pays its waste pickers
On 21 March 2013 Bogotá made its first payments to waste pickers, called recicladores, as providers of a public service: 87,000 Colombian pesos (about US$48) per ton of recyclables collected and transported, the rate the city paid private firms per ton taken to landfill. Colombia's Constitutional Court had recognized waste pickers as needing special state protection since 2003, and in 2011 it ordered the city to include them in its waste system (Order 275). The money comes from the cleaning fee residents pay, and 14,400 waste pickers were registered in the city's census.
LessonPaying people for a public service needs bank accounts, identity papers and official weighing points; the first payment reached 790 of the 4,000 waste pickers given accounts, and WIEGO, a research network that supported the ARB, wrote that more of all three were needed.
Credit
Asociación de Recicladores de Bogotá (ARB), led by Nohra Padilla; Constitutional Court of Colombia; City of Bogotá
2016
Lithuania's returns nearly triple
Lithuania started a deposit-return system in February 2016: buyers pay €0.10 on each drink container and get it back when they return the empty, mostly at reverse vending machines in shops. Before the system about one-third of drink containers were returned; 74.3% were returned in 2016, against a first-year goal of 55%, and 91.9% by the end of 2017. The nonprofit operator, Užstato sistemos administratorius (USAD), reported the figures through TOMRA, which supplied about 1,000 of the machines.
Credit
Government of Lithuania; USAD; TOMRA
2023
Scotland's deposit system collapses
Circularity Scotland, the industry-funded company set up to run Scotland's deposit-return system, called in administrators on 20 June 2023. The UK government had allowed only a temporary, limited exclusion from the UK Internal Market Act that left glass out, the start was pushed from March 2024 to October 2025 to match the rest of the UK, and drinks producers withdrew their funding. The administrators found the company owed more than £86 million, £65 million of it to the logistics contractor Biffa.
LessonA deposit system whose scope needs two governments to agree can fail before its first return, leaving its contractors unpaid.
Credit
Scottish Government; UK Government; Circularity Scotland
2024
The US measures its losses
The Recycling Partnership, a US nonprofit, found in its 2024 State of Recycling report that only 21% of recyclable material from US homes is captured for recycling, and that 76% of recyclables are lost at the household level, which it says calls for recycling service for every household and clear guidance on how to recycle locally.
Credit
The Recycling Partnership
Watch for
Poland's new deposit system
Poland's deposit-return system started on 1 October 2025, with a refundable PLN 0.50 on PET bottles up to 3 liters and cans up to 1 liter and PLN 1 on reusable glass bottles up to 1.5 liters. Drink producers must reach collection rates of 77% in 2025, 81% in 2026, 84% in 2027, 87% in 2028 and 90% from 2029. The Ministry of Climate and Environment said 520 million bottles and cans had been returned by the end of March 2026 at 52,000 locations, 78% of them through reverse vending machines.
Achieved whenofficial figures for a full year show the collection rate met its target, with the method published.
Claimed by
Poland's Ministry of Climate and Environment, April 2026
Open
01
Test chatbots on local rules
A public test of AI answers to "which bin does this go in?" for at least ten cities, scored against each city's own published rules. For each city, list 100 common items (pizza box, plastic bag, foil tray, coffee cup), give the city's official answer with a link, and record the answers from the city's own chatbot, where it has one, and from at least three general AI models.
The briefCredit · unclaimed02
Contamination, council by council
An open table of English councils whose recycling goes to a sampled sorting plant, giving for each the share of its input samples that was non-target or non-recyclable material in 2025, with a confidence interval, the number of samples behind it and the council's collection system (co-mingled, twin-stream or separate), with code and a written method.
The briefCredit · unclaimed
Open 01 · the brief
Test chatbots on local rules
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI answers to "which bin does this go in?" for at least ten cities, scored against each city's own published rules. For each city, list 100 common items (pizza box, plastic bag, foil tray, coffee cup), give the city's official answer with a link, and record the answers from the city's own chatbot, where it has one, and from at least three general AI models.
Why it is still open
Cities have started answering residents with AI chatbots, such as Hayward, California's "Lottie Sortwell" (May 2026), without published accuracy figures. The closest study found, by Eliot Park, Abhi Kumar and Pranav Rajpurkar (arXiv, December 2025), tested vision models on photos of items with different levels of local guidance, not the text answers city chatbots give. Hobby assistants on GitHub that answer from official rules, such as RePilot for French cities (2025) and HallSort for one Hong Kong university (October 2026), do not compare systems across cities.
Built with what exists
Each city's A-to-Z recycling guide as the answer key; Meta's Muse Ember to pull accepted items out of the guides (see the Note "The jobs agents repeat"); Google's Gemini Decision 1, which returns probabilities for fixed-answer questions, as one of the systems tested (see the Note "A thousand questions, one pass"); Python and a spreadsheet for scoring.
First step
Pick three cities with published A-to-Z guides, write 50 questions for each with the official answers, and run them through two general AI models.
Done looks like
At least 1,000 questions across at least ten cities in at least two countries, each answer-key cell linked to the city's page. Accuracy is published for every system and city, with wrong answers split into "recyclable sent to trash" and "trash sent to recycling", and recycling staff in at least three cities confirm their answer keys.
Open 02 · the brief
Contamination, council by council
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of English councils whose recycling goes to a sampled sorting plant, giving for each the share of its input samples that was non-target or non-recyclable material in 2025, with a confidence interval, the number of samples behind it and the council's collection system (co-mingled, twin-stream or separate), with code and a written method.
Why it is still open
The government's May 2024 response on Simpler Recycling in England relies on three national averages from WRAP analysis: 4% contamination for separately collected dry recyclables, 9.5% for twin-stream and 13.5% for co-mingled. The Environment Agency's 2025 Materials Facilities file holds batch-level inputs and sample results from 171 facilities and names the supplying council where the operator could identify one, but it gives no per-council table. Defra's yearly local authority tables carry each council's estimated rejects in metric tons, which Defra says also counts material pulled out of residual waste and then rejected; that is a tonnage, not a sorted composition.
Built with what exists
The Environment Agency's yearly Materials Facilities files on data.gov.uk; Python with pandas and openpyxl; Muse Ember to read each council's collection system from its website (see the Note "The jobs agents repeat"); an agent that cites the page behind each council's coding, built with a kit such as Cursor's Agent SDK (see the Note "Findings that cite notes").
First step
Download the 2025 file, clean the council names (the file spells one "Leed City Council"), link samples to batches and batches to councils, and publish the sample count for each council.
Done looks like
Every council with at least 20 input samples in 2025 listed with its contamination share and a 95% confidence interval, each collection-system code linked to its source, averages by collection type set beside WRAP's 9.5% and 13.5%, and at least five councils confirming or correcting their rows.