Menu pricing, grounded: what 100 ingredients told me.

I price a menu against inputs that never sit still. The citrus that costs one thing in March costs another by June, the produce order swings while proteins hold steady, and the invoice pound is never the pound that reaches the plate. For years I did what plenty of operators I know do — held a price too long, then reprinted in a panic the week the invoice jumped. This spring I stopped guessing and did the work instead.

So here is the claim this piece makes, before its evidence: an independent kitchen’s costs are far more legible than they feel. Read each ingredient against its own public history and most of them hold still enough to print; the ones that swing, swing in ways you can name; the big jumps run one direction and come back on their own; and the cheapest week to buy is already on the record. The edge is not guessing where a price is going — it is sorting the prices you already have: which to print, which to float, when to buy, when to wait. What follows is that sort, ingredient by ingredient.

I joined our public wholesale-price history with open yield, seasonality, and co-movement data into a per-ingredient read for 100 ingredients, checked it against a 134-ingredient reference, and grounded every layer in a 61-study research base (the field report cites the 36 most load-bearing, each with a DOI). Two honesty notes before the numbers. This is a descriptive read, not a forecast — it says how an ingredient behaved against its own baseline, never what it will do next. And it leans on wholesale reference levels, which are not the delivered price on your invoice; it stands on peer-reviewed work without being peer-reviewed itself. Here is what fell out.

The meat order has a buying season — and it runs against the calendar

Start with the surprise, because it is the most useful thing here and the part you cannot get from a generic pricing article. The meat order — the line you lie awake over — has deep, readable buying windows in the wholesale record, and they rarely land where you would guess. Whole chicken has been cheapest in May, about 47% under its dearest month; striploin in September, about 40% under; ribeye in August, about 26% under; beef tenderloin in July, about 20% under. Not during the grilling rush you would expect to set the price — on the market's own calendar.

Here is why the meat line rewards this the most: the trough is deep and it freezes. Sturdy produce freezes too — a berry or green-bean low banks fine — but a delicate trough like lettuce is gone by the weekend, and an August ribeye low is a buy-and-hold play you can actually bank. These windows are the rare place where knowing the calendar turns into real money on the meat order. They are wholesale reference levels and a descriptive read of the record — not your delivered price, and not a forecast of next year — but the pattern is exactly where a freezer earns its keep.

And the calendar quirk is not a meat-only oddity. Of the 41 ingredients that carry both a named cheapest wholesale month and a harvest window, 27 have their cheapest month fall outside that window. Onion reads cheapest in December against a fall harvest, acorn squash in December, eggplant in January, pear in March — each low lands in a month the calendar would not point you to. The meat line pays the most for the insight, because a protein trough is deep and it freezes, but the habit generalizes: read the wholesale low off the record, not off the season you expect.

The margin is thin — but '90% fail' is a myth

The fear worth clearing first is the one that sets the panic level: the claim that 90% of restaurants fail in the first year is a myth. When Parsa and his co-authors went and measured it, first-year failure for independents landed near 26%, and the 90% figure traced back to a source that could not produce the data behind it (Parsa et al. 2005). The margins are genuinely thin — full-service median pre-tax margins run in the low single digits (National Restaurant Association 2025), so a few points of food cost is the whole game — but thin is not the same as doomed, and it is the reason the rest of this scores each ingredient instead of pricing on fear.

Print or float: 37 lock, 7 float, 37 withhold

So which prices do you print, and which do you keep loose? I scored all 100 ingredients into four postures. Thirty-seven earned a lock — steady enough against their own baseline to print a fixed price and leave it. Nineteen earned a cushion — print, but build in a buffer because they drift. Seven earned a float — they move too much to pin to a printed number, so they belong on the special board or a market line.

And for 37, the read withheld a posture — but that column is really two piles. For 27 there is no public wholesale series to read at all: most herbs, all the shellfish, the whole fish, a handful of produce. The instrument is blind there, not undecided — so bring your own invoices. The other 10 do have a series, and a loud, clean one: bands of 30% to 62% against their own baseline, wider than any float, so they are withheld for swinging too wide to anchor, not for a murky signal. Those belong on a market line too, same as the seven.

I want to name that last column plainly: the instrument says 'I don't know' 37 times out of 100. That honesty is the reason I trust the other 63.

Here are the seven, so the number is not abstract: broccoli, yellow squash, raspberry, green leaf lettuce, romaine, Brussels sprouts, and serrano — every one produce, each swinging more than 20% against its own baseline. Those are the ingredients to keep off a printed price and on a market line or the special board; the rest of the menu can hold its printed price still.

The instrument says “I don’t know” 37 times out of 100. That honesty is the reason I trust the other 63.

Here is the part that surprised me. You brace for the meat order — it is the one operators lie awake over. Nine of the 27 proteins scored steady enough to print a fixed price, so the pricey cuts hold tighter than their reputation. But proteins are also the category the instrument withholds on most: 16 of the 27, because a public wholesale reference wanders furthest from your real delivered cost exactly on meat. So the honest read cuts both ways — the meat order is not the wild card you fear, but it is the one place to trust your own invoices over any index, including this one. Score each ingredient on its own baseline, and price your proteins off what you actually pay.

Here is what one card looks like, so you know what the instrument hands back before you open it. Ask it about beef tenderloin and you get the whole decision in one place: print it — its price held inside a ±3.5% band across three-quarters of recent weeks; it trims to about 85% edible, so the edible pound costs about 1.18× the invoice pound; and its cheapest month in the record has been July, about 20% under its dearest. Posture, trim, and timing on one card. Now run the five ingredients that scare you most — start with the proteins — and let it tell you which ones it can call and which ones it honestly can't. That second list is worth as much as the first.

Notice that card answers two questions on two different clocks. The posture — print it — is scored on how tight the price held against its own baseline: that is how to price the dish. The cheapest month is a separate axis, the gap between a public wholesale reference in that month and the ingredient’s dearest one: that is when to buy it. The two are independent, and they routinely both fire on the same ingredient. Twenty-two of the 37 locks carry a live buy-timing lever alongside the fixed print, average seasonal spread about 29% — and the lever is not the locks’ alone: 54 of the 100 ingredients carry one, across every posture (22 locks, 17 cushions, 6 floats, 9 withholds). Even a withhold you cannot print a price for can still be bought on its season. The tenderloin above is one; so is every protein I lean on for the freeze play — ribeye held a 5.8% band yet its cheapest month ran about 26% under its dearest, striploin 3% and 40%, whole chicken 5.7% and 47%. And band width does not track the size of the seasonal buy: red potato held a 0.8% band and still ran about 26% cheapest-to-dearest. So a lock is a stable menu price, not a dormant ingredient — you print it once and still buy it on the season.

Price on the edible pound, not the invoice pound

The invoice lies to you, and not on purpose. You pay for the as-purchased pound; the guest eats the edible pound; the gap between them is a tax you pay whether or not you cost for it. In our read the tax ran from gentle to brutal: mushrooms lose the least, so the edible pound costs about 1.14 times the invoice pound, while citrus is the worst everyday category at 2.16 times — with one honest asterisk: citrus is the category where the surviving pound is juice, not trimmed flesh, so its 2.16× is really a juice-extraction cost, not a knife-trim one. Either way, cost a plate on the invoice pound and you have under-costed citrus by more than double.

Those are category averages, and the average hides a wide spread — citrus is the worst everyday category, but some proteins and shellfish trim even harder. Here is the tax cut by cut, on the edible pound:

A boneless breast barely trims; a whole crab quadruples on the way to the plate. In dollars, so it stings: say lime lands at $2.00 a pound at the door — the pound that reaches the plate costs about $5.70, because only about 35% survives to juice and zest. Price your citrus at the invoice $2.00 and you have hidden nearly two-thirds of its real cost. (Illustrative: the $2.00 stands in for your own case price; the multipliers are from the record.)

One caution before you act on that table: a steep trim tax tells you where cost is hiding, but only a lock is safe to bake into a printed price. The steepest-trim lock is lemon — 2.22× the invoice pound, and steady enough to print — so it is the cleanest permanent fix; whole chicken at 1.67× is the next. Lime (2.86×) and whole crab (4×) carry the heaviest trim in the table, but both are withholds — lime’s wholesale wanders about ±38%, whole crab has no clean series to read — so correct their cost on a market line, not a printed one.

This is not guesswork. The USDA's Food Buying Guide publishes as-purchased-to-served yield tables, and they work as a general purchase-yield reference, not only a school-kitchen tool (USDA Food and Nutrition Service 2024). And the loss is not mostly the prep cook's knife: the USDA's cooking-yield tables show that for cooked proteins, a large share of the loss from as-purchased to edible is moisture and fat driven off in the pan, not trim (Roseland et al. 2017). So there are really two haircuts, not one — the knife, which is the trim tax above, and the fire, which the trim table leaves out entirely.

Put both haircuts on one number. Cost per served ounce is 1 ÷ (edible-yield × cooked-yield), and cooked-yield is the multiplier the trim table leaves out. Walk the same cuts to the pass and the number moves: the card put beef tenderloin’s edible pound at ×1.18 — that is the knife alone; walk it through the fire and the served pound is ×1.57. Ribeye and pork shoulder both trim to about 75%, then cook down again, so the served pound runs ×1.78 and ×2.05, not the ×1.33 the trim table shows. Short rib lands at ×2.56. And the mushroom the figure calls the gentlest trim? It is mostly water — it cooks to about 45%, so on the served pound it is one of the worst at about ×2.5, not ×1.14. The gentle case at the knife is a brutal one on the plate.

Grains and dried beans run the trade the other way, which is why they are the cheapest way to stretch a protein on the menu: a pound of dry rice or quinoa absorbs water and yields about three pounds cooked, so the served pound costs about a third of the invoice pound — about ×0.33. Dry black beans run about ×0.42. You buy the low number and serve roughly three times the weight; here the fire gives weight back instead of taking it. (The served pound is 1 ÷ (edible-yield × cooked-yield); protein cooking yields are USDA, Roseland et al. 2017, and grain and bean yields are from the USDA Food Buying Guide — both cited below.)

Two moves fall out of this. First, rank your plates on contribution-margin dollars, not food-cost percentage — that is the menu-engineering point, and the lowest-food-cost item is not automatically your most profitable one (Kasavana and Smith 1982). Second, the trim you cost for is also trim you can cut: the widely quoted 14-to-1 return on food-waste work is a multi-sector average, and the restaurant-specific figure is about 7 to 1 (Champions 12.3 2019). Costing the edible pound and cutting the waste are the same project seen from two ends.

The swap that mirrors doesn't hedge

Here is the finding that cost me money before I understood it. When an ingredient runs hot, the instinct is to swap in a cheaper cousin — and usually the instinct is right. But not always, and the exceptions are the expensive part. In our read, 94% of the big moves overlapped with at least one other ingredient moving the same way in the same six-week window — but read that number with care. Big moves bunch into the same windows and nine in ten point up, so overlap is close to automatic: reshuffle the dates at random and you still land company beside about nine in ten moves, only a few points below the 94%. So common overlap is mostly a property of how crowded and one-directional the moves are, not a sign that any two ingredients are truly linked — which is exactly why you cannot assume a backup is independent and have to check the specific pair. So I checked all 94 named swaps in the read. Most hold up: 78 rarely move with the ingredient they back up, real hedges that give you somewhere to go when your primary spikes. Only 16 are mirrors that move in step — the ones that quietly double your exposure while you think you are covered. One caveat on the evidence: each verdict rests on how many of an ingredient’s handful of big moves its backup shared, and for eight of the swaps that rests on just one or two episodes — too thin to lean on, so I flag those rather than trust them. Read a single verdict as a reason to pull up the two charts, not as proof.

The economics say why you check rather than assume. Pindyck and Rotemberg showed that commodity prices co-move more than shared fundamentals explain — the co-movement is 'excess' (Pindyck and Rotemberg 1990). Two ingredients moving together tells you nothing about whether they are economically linked, so a chart, not a hunch, is what separates a hedge from a mirror. The backup that protects you is the one that moves on its own clock, the one whose chart looks nothing like your primary’s — and most substitutes clear that bar. Pick the one that does, and drop the roughly one in six that only resembles it.

Don't reprint on a spike — the median cleared in 77 days

The last finding is the one I reach for most. When an ingredient spikes, the median move in our read cleared in 77 days, and three of four had cleared by about day 105. For the median move, the spike self-resolves inside a quarter. Reprint the menu the week the invoice jumps and you have paid the menu cost (Mankiw 1985) to chase a number that was coming back on its own. And you are reacting to a wholesale reference, not the plate: farm establishments received about 15.9 cents of each food dollar in 2023 (USDA Economic Research Service 2023), and the rest is labor, freight, packaging, the walk from the field to your kitchen — so when a wholesale number jumps, your plate cost moves less than the headline does.

And the big moves are lopsided, though it pays to read the lopsidedness carefully. Of the 432 big moves in the read, 393 point up and only 39 point down — about ten to one. Most of that is real: a sharp, sustained departure from an ingredient’s own normal is usually a supply shock, and supply shocks push up, not down. Some of it is the yardstick — a percent above normal has no ceiling but stops at −100% below, so the raw ratio flatters the gap. Up-moves also run somewhat bigger, a median 85% above normal against 33% below, though measured on a like-for-like scale that gap is about half again, not the two-and-a-half times the raw percentages suggest. What the data does not show is asymmetric persistence — the slow ‘feather’ down a rocket-up-feather-down pattern would predict: up-move episodes and down-move episodes last about the same, a median 77 days against 70. So on the input side relief is not slow, only rarer, and it arrives as drift rather than a matching downward spike.

One caveat on that 77: it is a pooled median across all 432 notable moves in the read — the middle half landed between 50 and 105 days — not a per-ingredient promise. A single ingredient’s own spikes ride on far fewer episodes, and a given one can sit well out in the tail; two of the 432 ran past a year.

This is why regular prices are sticky, and why the stickiness is rational. Bils and Klenow found the median price spell lasts about 4.3 months including sales (Bils and Klenow 2004). Strip the temporary sales out and Nakamura and Steinsson found regular prices change about half as often, closer to every 8 to 11 months (Nakamura and Steinsson 2008). Firms that Blinder's team interviewed reported repricing once or twice a year, with the physical cost of the change ranking low (Blinder et al. 1998). Prices are sticky because repricing has real friction — the research and the floor agree.

So the move is a temporary lever, not a reprint. Pull the item to a special, tighten the portion, run a substitution, and put a recheck on the calendar near day 105. Two honesty notes on yourself. The first is about your own menu, not the market: our input read reverts about as fast as it spikes, but that says nothing about whether you will bring a printed price back down once you have raised it. Peltzman found output prices across 242 markets rise faster than they fall in more than two of every three (Peltzman 2000) — the rocket up and the feather down is a pattern in what sellers do, and your menu is not exempt. So when the input finally eases, watch that the printed price actually follows it down. And do not assume the guest will not notice: a review of 160 studies found no single food-price elasticity, with food away from home among the most responsive categories (Andreyeva et al. 2010). The printed price is a promise. Keep it steady, and move the temporary levers underneath it.

Step back and the six reads are one habit. The panic I used to feel came from treating every input as a guess about the future. It never was. The citrus, the ribeye, the mushroom, the cheaper cousin, the jump on the invoice — each is a thing the record already sorts, once you read it against its own past instead of the headline. Print what holds. Float what swings. Buy on the low the calendar hides. Wait out the jump that clears. You do not need to know where prices are going; you need to know which prices you already have — and that, unlike the future, is on the record.

Every number here traces to the drawer beside it — our field report, and the peer-reviewed work it stands on. This is a descriptive read of public data as of July 16, 2026, not a forecast of your costs. Check my math.