What changed while the panels were meeting
The Gallatin Global Faculty Symposium ran on 27 and 28 July 2026 and treated food as a subject that refuses to stay inside one discipline: bagels leaving a city, Ottoman kitchens, wartime root vegetables in Singapore, hawker centres becoming community canteens in Shanghai, rock weathering on New York farmland. My paper, Food as Urban Infrastructure, closed the final panel on food disparities alongside Chen Chu's territorial analysis of rural–urban planning. The argument was straightforward: the movement from edible food to discarded material is neither automatic nor neutral, and food waste is therefore the most honest diagnostic a city has of its own governance.
Three independent 2026 data points now converge on that claim. First, the technology matured. ReFED's May 2026 study The Food Operating System, drawn from more than 40 operator and developer interviews, records waste reductions of 20% to 53% across commercial foodservice and manufacturing, and a single predictive-ordering pilot with a United States value grocer projecting USD 2.7 million in annual savings. Fresh retail is where the evidence is strongest, and it is exactly where the loss is largest.
Second, the constraint moved. Reporting on the same deployments in June found that technically correct ordering recommendations are routinely overridden by store managers who would rather carry spoilage than face an empty shelf. That override is not irrational. It is the correct response to an incentive structure that penalises a stockout visibly and a discarded case invisibly. Third, the public system produced its own verdict. New York's Independent Budget Office found that after a citywide mandate, an April 2025 enforcement launch, a pause three weeks later and a partial 2026 resumption, organics rose 68% year on year and still amounted to 2.4% of residential waste against an organic share of roughly 36%; total diversion sits near 17% against an estimated 75% potential.
Set that beside demand. Pantry and soup-kitchen demand in New York runs up to 80% above pre-pandemic levels; the Community Service Society finds roughly 22% of New Yorkers food insecure in a year. In Argentina, ODSA-UCA measured poverty at 30% in the first quarter of 2026, with the basic food basket rising 11.6% against an 8.6% wage index. Globally, UNEP counts 1.05 billion tonnes wasted in 2022, 60% of it in households.
A city that discards edible food while its pantry queues lengthen is not short of capacity, data or technology. It is short of an owner for the outcome.
The Three Distances
The framework I presented at Gallatin, and that Socradata now uses in operating work, replaces the familiar prevention–recovery–recycling hierarchy with a diagnostic question: what distance separates the actor who could prevent the loss from the actor who bears it? Food becomes waste across three measurable distances, and each has a different remedy.
Between what the system records and what the decision-maker can see. New York is data-rich and situational-awareness-poor: freight, food safety, commercial waste, hunger, labour and emissions are each observed by a different agency using a different definition, and firms hold operational data that never crosses an organisational boundary. Nobody is looking at the whole chain, so nobody can be held to it.
Between the actor who makes the decision and the actor who pays for it. The store manager overriding the forecast, the kitchen that overproduces against uncertain covers, the procurement contract rewarding volume and consistency rather than avoided loss: in each case waste is the cheaper private error. This is the distance that defeats working technology, and it is closed with contracts and measurement, not with better models.
Between the people who make the system circular and the institutions that hold decision rights. Kitchen staff, delivery workers, street vendors, community coordinators and, in Buenos Aires, the cooperatives of recuperadores urbanos are part of the infrastructure, not external to it. Registering them in a database is not inclusion. Inclusion is decision rights, safe conditions and a share of the value that circularity creates.
The spanning metric is the prevention ratio: avoided surplus divided by the sum of avoided, recovered and recycled volume. Cities and firms report diversion almost universally and prevention almost never, which is why a programme can post a 68% annual increase and still be governing badly. Diversion measures how well you dispose. Prevention measures how well you decide. Mourad's analysis of competing food-waste pathways made the political point a decade ago: prevention, recovery and recycling are not neutral technical options but rival claims on attention and investment, and the easiest one to celebrate is the last.
So what: a forecast that is correct and ignored is not a technology failure; it is a governance artefact, and it will reappear in every AI deployment a city or an operator attempts. Any organisation that cannot name who owns an avoided loss will keep buying accuracy it has no mechanism to act on.
Three regional operating patterns
The patterns below are composites from Socradata engagement work in the Southern Cone. None begins with a model. Each begins with a distance the organisation was already paying for.
Institutional catering, CABA — measurement at the tray, not at the bin. A Buenos Aires operator running school and corporate kitchens moved the measurement point upstream: plate waste weighed by menu line and service window rather than by total container, production redesigned against the resulting curve, and human sign-off retained on any menu change affecting nutritional adequacy. Preventable edible waste per meal served fell 31% in two terms; food purchased but not consumed fell 18%; the prevention ratio moved from 0.12 to 0.46. No forecasting model was purchased in the first six months.
Metropolitan recovery network, Buenos Aires — labour recognised as infrastructure. The city runs roughly 7,500 tonnes of waste per day against about 430 tonnes recovered daily across three separation plants, worked by 12 cooperatives and some 6,500 recuperadores, promoters and green-centre operators. A recovery programme designed with those cooperatives — shared route and volume data they co-own, payment tied to recovered quality rather than tonnage handled, and a seat in the rule-setting forum — lifted organics captured at source 2.4x in the participating comunas while cutting rejected material 27%. Research on the city's waste metabolism has long shown that a substantial share of its residues is managed beyond its own boundaries; a programme that ignores who performs that work optimises a fiction.
Fresh grocery, Argentina and Chile — closing the override. A regional chain had already deployed predictive ordering in produce and was seeing the documented pattern: accurate recommendations, high override rates, unchanged shrink. The fix was not a retrain. Store-level scorecards were rebalanced so that shrink and availability carried comparable weight, every override was logged with a reason code visible to the category team, and the model was re-tuned on the reasons rather than on the outcomes alone. Override fell from 41% to 9% of recommendations; produce shrink fell 24%; on-shelf availability held above 96%. The model had been right for eleven months.
From framework to municipal and operator practice
The sequence is deliberately organisational before technological. Begin with a small number of outcomes that cross agency and departmental boundaries — preventable edible waste in public institutions, affordable access in underserved neighbourhoods, the share of surplus recovered inside safe time windows, working conditions in circular-economy roles — and attach a baseline and a distributional measure to each. Average improvement is not enough if the benefit accrues to areas already well served.
Then map the operating system behind each outcome: actors, contracts, data, workflows, incentives, failure points. A school-meal waste programme involves procurement specifications, menu design, scheduling, portioning, safety rules, staffing and collection. A dashboard will not align those. Product governance will: a named service owner, a cross-functional team, a risk register, explicit decision rights, a recurring review. That is the difference between POC theater and an operating model.
So what: recovery is necessary and insufficient. A city that relies on surplus to address hunger has coupled two injustices — routine overproduction on one side, insecure access on the other — and called the coupling a solution. Prevention plus stable entitlements is the harder programme and the only durable one. From pilot to policy.
Governance
Give every failure point an owner, a baseline and a review cadence: procurement for contract incentives, kitchen management for production routines, logistics for cold-chain performance, agencies for regulatory coherence, platform operators for accessibility and data integrity. Keep humans in the loop where a machine decision changes what a resident or customer receives — and give that human authority, time and an escalation path, since confirmation without power is not oversight. Document model purpose, data sources, error rates, affected populations and appeal routes for any decision touching eligibility, enforcement or benefits. Interoperability or it doesn't scale: shared definitions of edible surplus, avoidable waste and recovery quality come before any integration.
KPIs
Prevention ratio above 0.40 within twelve months, reported beside diversion rather than behind it. Preventable edible waste per meal served, trending down at least 20%. Cost of food purchased but not consumed. Override rate on accepted recommendations under 10%, with reason codes. Share of surplus recovered inside safe time windows above 85%. Access to affordable nutritious food disaggregated by neighbourhood, not averaged. Compensation and safety in recovery and recycling roles. Participation of affected communities in decisions, counted as decisions influenced rather than meetings held.
12-month roadmap
0–90: define outcomes and shared definitions, select two institutional settings, run a participatory baseline with the workers and residents inside them, and measure the prevention ratio and override rate for the first time. 90–180: redesign the workflows those baselines expose, deploy a limited set of digital tools against a named commitment, publish an initial accountability register naming owners. 180–360: revise procurement so contracts reward avoided loss, scale shared services and definitions, formalise worker and community participation, and compare outcomes at neighbourhood level.
KPIs before APIs, on a plate.
Two days of interdisciplinary papers on food and the city are worth a quarter of sector reports, because food refuses the abstraction that smart-city discourse depends on. Historians, media scholars, planners and practitioners described the same object from seven directions, and it kept insisting on labour, memory, jurisdiction and the body. That is a useful discipline for anyone building urban technology: our systems recognise products far more readily than they recognise people.
The operational conclusion is unglamorous and, I think, correct for 2026. The frontier has moved past capability. Forecasting, computer vision, traceability and matching platforms all work well enough to put in a contract. What has not moved is the institutional machinery that decides whether a correct recommendation is acted on, who bears the cost when it is not, and whose labour absorbs the difference. A city can mandate, instrument, measure and still divert 2.4%; a grocer can buy an accurate model and override it for eleven months. Neither will be solved by the next model release.
This is the layer Socradata works in, and the entry point is deliberately modest: find the decisions your organisation already makes badly, measure the distance between the actor who decides and the actor who pays, and instrument that distance before buying anything. KPIs before APIs — because a city that can measure prevention will govern differently from one that can only count what it hauled away. The full working paper, Food as Urban Infrastructure: Waste, Care, and Human-Centered Governance in New York and Buenos Aires, extends this into a five-dimension governance-by-design framework and is available on request.
Measure the distance before you buy the model
Socradata runs Operational Diagnostics for municipalities, food-service groups, retailers and public institutions across Argentina and the region: a map of the decisions your operation already makes badly, a prevention-ratio and override-rate baseline, a governance and data-interoperability gap review, and a board-ready scorecard with a phased roadmap.
Sources and references
NYU Gallatin Global Faculty Symposia. (2026). Food and the City — New York 2026: Program. wp.nyu.edu
ReFED & The Spoon. (2026, May). The Food Operating System: How AI Is Being Deployed Across the Food System to Reduce Waste. refed.org
The Packer. (2026, June 15). AI Cuts Food Waste in the Produce Aisle, But Grocers Battle Cultural Overrides. thepacker.com
New York City Independent Budget Office, reported in Waste Today. (2026). NYC IBO releases curbside composting program analysis. wastetodaymagazine.com
United Nations Environment Programme & WRAP. (2024). Food Waste Index Report 2024. unep.org
New York City Council. Emergency Food in NYC (data dashboard). council.nyc.gov · Community Service Society of New York. Food Insecurity in New York. cssny.org
Gobierno de la Ciudad de Buenos Aires. Centros Verdes y Recuperadores Urbanos. buenosaires.gob.ar · ODSA-UCA. (2026). Informe de pobreza e indigencia, primer trimestre 2026.
Evans, D. (2012). Beyond the throwaway society. Sociology, 46(1), 41–56. · Foster, S. R., & Iaione, C. (2016). The city as a commons. Yale Law & Policy Review, 34, 281–349. · König, P. D. (2021). Citizen-centered data governance in the smart city: From ethics to accountability. Sustainable Cities and Society, 75, 103308. · López de Munain, D., Castelo, B., & Ruggerio, C. A. (2021). Social metabolism and material flow analysis applied to waste management: A study case of Autonomous City of Buenos Aires. Waste Management, 126, 843–852. · Morgan, K. (2009). Feeding the city. International Planning Studies, 14(4), 341–348. · Mourad, M. (2016). Recycling, recovering and preventing ‘food waste’. Journal of Cleaner Production, 126, 461–477. · Pierini, V. I., Mazzeo, N., Cazenave, M., & Semmartin, M. (2021). Waste generation and pro-environmental behaviors at household level: A citizen science study in Buenos Aires. Resources, Conservation and Recycling, 170, 105560. · Robeyns, I. (2005). The capability approach. Journal of Human Development, 6(1), 93–117.