Play the Planet asks people to build things: greenhouses, grow towers, planter boxes, and the occasional one of each for a neighbor who could not have built it alone. The obvious question about any of that is whether it helps. The less obvious question, and the one this piece is about, is how anyone would know.
We are publishing our arithmetic because a number nobody can check is not evidence, it is decoration. Everything below can be recalculated by a stranger from public data. Where we have made a judgment call, we say so. Where the method will overstate our results, we say that too, in its own section, before anybody else has to say it for us.
What we are measuring against
You cannot say a greenhouse closed part of a gap without a gap to measure. Ours comes from Feeding America’s Map the Meal Gap, which produces county-level estimates of food insecurity and, critically for us, publishes them in dollars and meals rather than calories.
That unit choice is not a convenience, it is the whole reason the method works. Food insecurity, in the official sense, is not a caloric deficit. It is inconsistent access to adequate food. A household can hit its calorie target on cheap shelf-stable starch and still be food insecure in every way that matters. If we measured a greenhouse in calories, we would be measuring the one dimension on which fresh vegetables perform worst, and we would conclude that growing food does not feed people. The dataset’s own units avoid that trap, so we use them.
Russell County, in four published numbers
Our pilot Holon is Russell County, Virginia. These are Feeding America’s figures for 2024, taken directly from the county sheet of the Map the Meal Gap data file, not recomputed by us:
| Food insecure persons | 5,410 |
| Food insecure children | 1,250 |
| Cost per meal | $3.50 |
| Weighted weekly dollars needed | $21.89 |
| Annual food budget shortfall | $3,593,000 |
More than one person in five in Russell County is food insecure. The shortfall figure is the amount of money that, spent on food at local prices, would close the gap for a year.
Three conversions, and only one of them is ours
Getting from a built greenhouse to a share of that gap takes three steps. Two of them belong to other people, which is exactly how we want it.
1. Pounds to meals: divide by 1.2
The Feeding America network converts donated food to meals at 1.2 pounds per meal, a figure derived from the USDA’s What We Eat In America dietary database. Every food bank in the country reports on it.
We adopt it for one reason: it makes our numbers comparable to everyone else’s. A number that can only be compared to itself tells you nothing about whether you are doing well. This is the only constant in the whole chain that we chose, and we chose the one already in use rather than inventing a better one.
2. Meals to dollars: multiply by the local cost per meal
Map the Meal Gap publishes a cost per meal for every county, computed as the average weekly food spending of food secure individuals divided by 21, being three meals a day for seven days. Russell County’s is $3.50. The county next door will differ, and that is the point: a meal grown in a place where food is expensive is worth more than the same meal somewhere cheap.
3. The denominator: shortfall divided by cost per meal
This is the step where it would have been easy to be clever and wrong.
Map the Meal Gap publishes both the annual dollar shortfall and the cost per meal. Dividing one by the other converts the county’s gap into meals:
annual meal gap = annual food budget shortfall / cost per meal
= $3,593,000 / $3.50
= approximately 1,030,000 meals per year
Russell County’s food insecurity gap is about 1,030,000 meals a year. That is the number everything else gets compared against.
We compute it this way deliberately. Both inputs are published columns, so the methodology underneath stays Feeding America’s. When they revise how the shortfall is modeled, our denominator follows automatically, without anyone here editing a constant or remembering that a constant exists. The alternative, rebuilding their model inside ours, would have left us maintaining a private copy of somebody else’s research that silently goes stale.
How we checked it
Before reading Feeding America’s methodology, we tried to reconstruct it from the data, on the theory that a formula you can rediscover is a formula you understand.
The obvious guess was that the annual shortfall is simply the food insecure population times the weekly dollars needed times fifty two weeks. That guess is wrong, and usefully so: it overshoots by a consistent amount. Across twelve counties the ratio of the published shortfall to our naive estimate came out at 0.5833 every time, varying only in the fourth decimal place, and that variation is explained entirely by the published shortfall being rounded to the nearest thousand dollars.
0.5833 is 7/12. Feeding America’s methodology page says so in as many words: the weekly figure is annualized by multiplying by 52 and again by 7/12, being the average number of months per year that food insecure households actually experience food insecurity, per the USDA. Food insecurity is episodic. Treating it as a condition people live in fifty two weeks a year would overstate the gap by seventy percent.
It has since survived a release it was not fitted to. The 2026 edition, carrying 2024 data, landed after this method was worked out. Running the same arithmetic on the new figures: 5,410 food insecure people times $21.89 a week times 52 times 7/12 is $3,592,161, against a published shortfall of $3,593,000, which is agreement inside Feeding America’s own thousand-dollar rounding. The 6.25 meals a week constant reproduced too, at 6.254 against last year’s 6.247. A formula that only fits the data it was derived from is a curve, not a method.
We mention this partly because it is a good constant to know and partly as a demonstration of the standard we are trying to hold: an independent derivation and the published documentation agreeing is worth considerably more than either alone.
One incidental finding fell out of the same exercise. Divide any county’s weekly dollars needed by its cost per meal and you get 6.25, everywhere. The weekly shortfall is a national estimate adjusted for local food prices, so the gap is a constant 6.25 meals per food insecure person per week across the entire country; only its dollar value is local. Annualized, that is 189.6 meals per person per year.
What one greenhouse actually does
Here is where we disappoint ourselves in public.
Take a four by six foot greenhouse, twenty four square feet, producing somewhere between twelve and twenty four pounds of mixed vegetables in a season. Being a greenhouse, and in this part of the country, it gets two plantings and two harvests a year. A third is possible with real work, but two is the claim we are confident nobody would argue with, so that is the one we make: twenty four to forty eight pounds a year. Run it through the chain:
- 24 to 48 pounds a year, divided by 1.2, is 20 to 40 meals
- times $3.50 is $70 to $140 of food at local prices
- against a 1,030,000 meal county gap, that is about two to four thousandths of one percent
Closing one percent of Russell County’s gap would take roughly 12,300 pounds of produce, which is somewhere between 260 and 510 greenhouses, each harvested twice a year. We are not going to build five hundred greenhouses in Russell County any time soon, and we are not going to pretend otherwise.
Which is why the county percentage is the wrong headline
A county-scale denominator under a household-scale intervention produces a number that rounds to zero. One greenhouse against 1,030,000 meals is two to four thousandths of one percent, and no amount of careful arithmetic is going to change that.
We are going to track and publish it anyway, and the reason is worth stating, because a rounding error is not the same thing as a zero.
Consider a single rooftop solar array in a rural county. Measured against the county’s total electricity demand it generates a meaningless fraction of a percent, and anyone wanting to dismiss it can do so with a straight face. But it is not going to be the only array on that chart forever. The useful thing about putting it on the chart while it is still negligible is that you establish the line early, and then you get to watch it stop being negligible. A measurement you only start taking once the number is flattering is not a measurement, it is an announcement.
So we report at both scales, deliberately. The household figure is what this intervention can honestly claim today. The county figure is the line we intend to watch stop rounding to zero. One greenhouse is a rounding error. Hundreds are not, and the arithmetic above says exactly where the crossover sits: somewhere between 260 and 510 builds puts Russell County at one percent. We would rather publish that number from the beginning, while it is embarrassing, than start publishing it at the point where it makes us look good.
The denominator that matches what we actually do is the household. At 189.6 meals per food insecure person per year, a two person household faces a gap of about 379 meals. One greenhouse at 20 to 40 meals a year covers something like five to ten percent of that household’s annual shortfall.
Five to ten percent of one family’s yearly food gap is a modest claim. It is also a true one, and it is the size of claim this intervention can actually support. “We closed eight percent of this family’s food gap” and “we closed 0.003% of Russell County” describe the identical event. We will be reporting the first one, alongside plain counts: how many builds, how many households, how many of those households told us they were struggling to put food on the table, and how many reported actually harvesting from the thing we helped them build.
The county gap stays on the page as context, because people deserve to know the scale of the problem next to the scale of the response. It is not the score.
The honest caveats, before someone else makes them for us
The 1.2 pound conversion flatters vegetables. It is derived across all food. Fresh produce is mostly water, so 1.2 pounds of lettuce is not a meal in any nutritional sense. Using the network standard keeps our numbers comparable to every food bank’s, which we think is worth more than a private correction factor would be, but it does systematically overstate what produce specifically contributes. If we ever publish a figure that depends on this being precise, we will have overreached.
A built greenhouse is not a harvested one. Square footage times a yield constant produces a number that moves in exact proportion to square footage and tells you nothing about whether a single thing was eaten. It is the square footage restated in friendlier units. This is why we collect a separate harvest report, and why anything we publish will keep observed quantities and estimated ones in different columns. A structure built in October grows nothing until spring, and a grow tower nobody waters grows nothing ever.
Our household food security data is self reported, and it has a twelve month memory. We use the Hunger Vital Sign, a validated two question screener drawn from the USDA’s Household Food Security Survey Module, which identifies households at risk of food insecurity with about 97% sensitivity. At risk is not the same as confirmed, and we will not quietly upgrade the one into the other. Both of its questions ask about the past twelve months, which means a household asked at build time gives a clean picture of the year before the build, and a clean picture of the year after requires waiting roughly twelve months. There is a one year lag built into the instrument. We would rather state that than publish a before and after comparison in month three that mostly measures the before.
Not everyone who builds is food insecure. A greenhouse in the yard of a household that was never struggling is a fine thing and it does not close any part of the gap. This is the entire reason we ask, and the reason the screening questions and the build records are kept together.
Growing food is not the only thing a greenhouse does, and the things it does besides are the ones our method cannot see. Knowing how to grow food is a durable change in a household’s position that no meal count captures. We are not going to argue that the unmeasured benefits are large, because that is what everyone says about the parts of their program that do not show up in the numbers. We are going to say that the number is a floor, and leave it there.
What happens when the data changes
It is already changing. The USDA has cancelled the Current Population Survey Food Security Supplement, the federal survey that has underpinned official food security statistics for decades, describing the reports as redundant and costly. Map the Meal Gap’s 2026 release, covering 2024, was unaffected. Future editions will not have that input, and Feeding America has said it is working with its network on alternative sources for 2027 and beyond.
So the ground under this method will move. We built it to survive that: the only number we own is 1.2, and the rest is two published columns and a division. If those columns survive under any name, our arithmetic survives with them. If they do not, we will publish what replaced them and why, in a post like this one.
One more thing worth saying plainly, since it bears on everything above. Roughly half of all food insecure people in the United States, about 25 million of them, have incomes too high to qualify for federal nutrition assistance. Any attempt to measure food insecurity by counting program enrollment will miss them. That is a large part of why we are measuring the gap rather than the caseload.
Why any of this matters
A project like this one can run for years on good intentions and photographs. The photographs are real, the intentions are real, and neither of them tells you whether the thing works. Without a denominator you are not doing community development, you are performing it.
The arithmetic above is conservative on purpose, and it produces unflattering numbers on purpose, because a measurement that can only return good news is not a measurement. We would rather publish five to ten percent of one household’s gap and have it be true than publish something impressive we could not defend to a hostile reader with a spreadsheet.
If you find an error in any of this, we want to hear about it. That is the other reason it is published.
Sources and further reading
- Feeding America, Map the Meal Gap: How We Got the Map Data (methodology, including the 52 and 7/12 annualization and the cost per meal derivation)
- Feeding America, Map the Meal Gap data by county
- Feeding America, Map the Meal Gap 2026 release
- Children’s HealthWatch, the Hunger Vital Sign two question food insecurity screener
- USDA Economic Research Service, Measurement of household food security
- USDA, What We Eat In America, the basis for the 1.2 pounds per meal conversion