The Cost Index · Monthly Dispatch · Edition of 2026-07-09 · 19 min read · By Don Goldstein

The Restaurant Cost Index — July 2026. Zero of 81 cleared the gate — hold.

Name: Muntin Restaurant Cost Index — July 2026 edition (read as of 2026-07-06)
Published: 2026-07-09
License: https://creativecommons.org/publicdomain/zero/1.0/

Read as of July 6: the 16-staple wholesale basket sits −5.0% against baseline, 24 of 81 items up , and none cleared the noise gate . The licensed move is hold. A dated read, not a month average — and wholesale levels, never your delivered price.

This is the first monthly print: the prior data print is dated 2026-06-18 (the 2026-07-06 weekly shares this edition's data ), and this window runs 2026-06-08 through 2026-07-06 .

−5.0%

the weighted basket vs baseline · read as of 2026-07-06

24 above baseline · 21 flat · 36 below
0 of 81 cleared the false-discovery gate
month window since 2026-06-08

As of July 6, 2026, the 16-staple wholesale basket reads −5.0% against its tracked baseline; the loudest raw move, green beans at +125.4%, did not clear the noise gate — the licensed read is hold. These are wholesale levels, not delivered prices. If green beans on last week's invoice made you reach for the menu file, this is the print that says put it down.

The gate held everything back, and that is the finding

Of the 81 ingredients on the July panel , zero cleared the false-discovery gate . Not green beans at +125.4% . Not blueberries at +42.3% . Not romaine, reading −55.9% the other way. The verdict column splits flat 47, easing 16, structural 15, insufficient 3 ; the bias column reads hold 63, re-price 15, watch 3 . And still the gate-cleared re-price list prints empty, and so does the watch list .

Here is the machinery under that. Running 81 tests at once guarantees loud accidents — reads near +40% (asparagus, blueberries) are what that many noisy series produce by chance . So every read passes a false-discovery correction across the whole panel before an action verb is allowed to ride on it . That is all "licensed" means on this page: the gate licenses verbs, and nothing else here is allowed to tell you to act. Fifteen items carry a structural verdict with a re-price bias this month ; all fifteen failed that gate. The script's own fallback line for this state is the honest one: "The loudest raw move is green beans at +125.4% , but nothing cleared the noise gate — the licensed read is hold."

I want to be plain about what this is not. It is not a sleepy month at the terminal; 24 up, 36 down, 21 flat is plenty of motion. It is the Index declining to manufacture a story out of motion its own history already accounts for. Do nothing, and know why, is an operator decision.

1

81 ingredients read

24 up, 36 down, 21 flat — each against its own tracked baseline window.
2

Verdicts assigned

flat 47 · easing 16 · structural 15 · insufficient 3.
3

The false-discovery gate

81 simultaneous tests; at that scale, reads near +40% (asparagus, blueberries) are expected noise. Cleared: 0 of 81. The 15 structural reads — green beans at +125.4% the loudest among them — all carry gated: false.
4

The licensed action list

re-price: 0 entries · watch: 0 entries. The licensed read is hold.

The panel's 81 reads, walked through the false-discovery gate for the 2026-07-06 data: zero cleared, and the action list prints empty — hold.

How the gate works

Benjamini–Yekutieli false-discovery correction across the whole 81-item panel (cost-null-gate.js, methodology v1.3.0) — an action verb may ride only on a read that survived it. See the Cost Index methodology.

The basket reads −5.0%, and the month-over-month line stays blank

The weighted Restaurant Basket sits −5.0% against its baseline, direction down , with 63.6% directional agreement among contributors and 16 of 16 declared ingredients reporting on full coverage . The basket's data anchor is 2026-05-01 .

That anchor is why no month-over-month basket move prints here. The comparison machinery ran — the prior data print, dated 2026-06-18 — but the anchor hasn't refreshed since that print , so there is no new basket move to report, and I'd rather show a blank line than a manufactured delta. Per-ingredient month-over-month deltas are blank for a related reason: they compute only for gated items , and nothing gated.

Below baseline24 above · 21 flat

Weighted basket vs baseline — a displacement, not a share

Basket 16 of 16 contributing · 63.6% agreement · anchor 2026-05-01

−5.0%

The tick is the baseline; the bar is the basket’s gap below it, drawn against a 10% scale.

Where the panel sits as of 2026-07-06: 36 of 81 reads below baseline, and a weighted basket displaced −5.0% from its baseline on the 2026-05-01 anchor.

Source: the dispatch payload behind these rings

Muntin Cost Index dispatch payload, read as of 2026-07-06 — basket weights v2026-Q2, methodology v1.3.0. Frozen edition snapshot: week-2026-07-06.json (CC0).

The biggest individual pulls, stated as forces and not as a sum: chicken breast (boneless) reads −14.0% at weight 0.17 , worth −2.38 basket points — the heaviest single force on the board, with a pressure read behind it naming feed grain (corn) on a 6–9 week lead (association, not cause). The counterweight is vegetable oil at +29.3% on weight 0.12 , worth +3.51 points of push, with no pressure read attached . Behind those two: tomato −33.3%, onion +25.6%, pork shoulder −19.8%, butter −15.1%, eggs −15.0% . One driver read worth naming: imported seafood at +1.9% , leading 16 seafood lines .

Three arc shapes: given back, still climbing, gone silent

Percent-against-baseline is a snapshot, not a move, so this section walks the dated reads instead. All three loud risers carry low-confidence reads , so I describe levels and gaps and leave the arrow alone.

Green beans read +125.4% against their tracked window: median $40.75 a carton, range $36.50–$44.75 , as of July 6 . The arc (Chicago wholesale series, 18 reads ): $33.50 on June 9, a dip to $25.00 on June 15, a peak of $45.50 on June 25, then $40.00 at the window edge on July 6 . The shape is the story: the end sits +19% over the start but −12% off the peak . Whatever happened here peaked inside the window and is already digesting.

Blueberries read +42.3% , median $35.94, range $30.69–$39.94 . The Boston arc (18 reads ): $28.00 on June 9, a trough of $12.50 on June 25, then $37.00 on July 6 . The price nearly tripled in the 11 days from trough to window edge, and the window ends on the peak — still climbing at last read. Asparagus wears the same shape: +39.7% , median $49.75, range $44.37–$55.13 ; $23.50 on June 9, a trough of $18.25 on June 11, $44.00 on July 6 — a 2.4× climb off the trough , ending on its peak.

Cantaloupe's +34.2% is a different case: the Atlanta series carries only 8 reads and goes silent on June 18, mid-window, at $36.13 (from $26.00 on June 9) . The panel's verdict is "insufficient" , and its reason prints verbatim: "not enough history to tell a spike from a real trend — treat as real" .

On the down side, romaine reads −55.9% : $88.00 on June 9 straight down to $36.25 on July 6 , a monotone −59% unwind across the whole window, and the panel marks it seasonal . Lime reads −53.1% : $34.00 down to $22.50 by June 29, then flat at that floor through July 6 . Zucchini (−51.9%) and eggplant (−39.1%) round out the fallers. Green beans, blueberries, and asparagus all wear the same "structural" stamp ; three different pictures on the glass.

Green beans · +125.4% vs its tracked window

open 06-09 $33.50trough 06-15 $25.00peak 06-25 $45.50last 07-06 $40.00median $40.75/carton · range $36.50–$44.75

Blueberries · +42.3% vs its tracked window

open 06-09 $28.00trough 06-25 $12.50last = peak 07-06 $37.00median $35.94/carton · range $30.69–$39.94seasonal July median $18.70 — this window reads nearly double it

Asparagus · +39.7% vs its tracked window

open 06-09 $23.50trough 06-11 $18.25last = peak 07-06 $44.00median $49.75/carton · range $44.37–$55.13

Romaine lettuce · −55.9% vs its tracked window

open = peak 06-09 $88.00last = trough 07-06 $36.25median $36.25/carton · range $27.12–$45.38

Lime · −53.1% vs its tracked window

open = peak 06-09 $34.00trough 06-29 $22.50last 07-06 $22.50

Five arcs, 2026-06-09 to 2026-07-06 — three shapes under one window, drawn without a basket line because the basket's 2026-05-01 anchor has not refreshed since the prior edition.

Source: the June arcs, read by read

USDA AMS terminal-market reports (Chicago for green beans; Boston for blueberries, asparagus, romaine, and lime), wholesale basis, dated reads 2026-06-09 → 2026-07-06 — via the frozen edition snapshot, week-2026-07-06.json (CC0). Each line ends at its last committed read; a break in a line is a gap in the source, and nothing interpolates.

One blueberry, three instruments, three true readings

The panel calls blueberries +42.3% , structural. The seasonal file says July is historically blueberry's cheapest month — the established July median is $18.70 , against a current median of $35.94 , nearly double the norm. So this read is counter-seasonal, not the calendar doing its usual work. The lock/float classifier buckets blueberry float at ±24.8% — a band too wide to commit a contract against. And the gate declines to license any action on it: gated false . No contradiction anywhere in that stack. The panel measures the gap, the calendar measures the norm, the classifier measures the risk, the gate measures the proof — every layer is telling the truth about different things.

Romaine: the fall is measured, the leading read points back up

The measured story is finished: −55.9% , a straight unwind from $88.00 to $36.25 , and the calendar's established August takes another −18.2% off romaine (July median $31.80 to August median $26.00) . But the pressure layer reads building on a 0–3 week lead from shipment volume: lettuce shipments −8% , CA/AZ drought +6% , plus a desert-to-coast transition flag (association, not cause). The fall already happened; the leading indicators point at the rebound. I hold both in view without netting them: one number looks back at a fall already banked, the other is a lead window.

Withheld for opposite sins

The lock/float classifier refuses a posture on 73 of 100 ingredients, and the refusals teach more than the grants. Eggs are withheld as volatile: a ±65.2% band, with a committed history spanning an $8.51 high and a $0.49 low . Button mushrooms are withheld as flat: every committed read prints exactly $14.75 , zero up-weeks and zero down-weeks across 149 tested weeks . And chicken breast, whose ±5.2% half-width would pass the ±8% lock bar on width alone, is demoted to cushion because its proven coverage floor is 0.581 against the 0.60 honesty bar . Chaos, deadness, unproven confidence — the classifier prices three separate failure modes.

What to do this month

Hold prices. That is the licensed action. The loudest raw gap on the board is green beans at +125.4% , and gaps that loud are what that item's own history produces by chance; the gate exists to catch exactly that. Nothing in this print licenses a re-price, and nothing licenses opening a vendor negotiation off these pages.

Two qualifiers ride along. Every level here is a wholesale reference, not your delivered price. And every number is a first print: the revision ledger's median revision magnitude is 14.0%, its 90th percentile is 53.3% , and the record's largest single correction moved zucchini from $9.50 to $26.50 — a +179.0% revision . Read the levels between the produce drop and pre-shift, keep the printed ranges on hand for your own invoices, and leave the menu where it stands. You're not behind for holding. Hold is the decision the whole panel earned, all 81 items of it — a decision, not a shrug.

The full board

Every shippable ingredient on the panel — all 81 — grouped the way the hub groups them and sorted loudest-first inside each group. The summary row is the scan: direction, name, gap, dollar median. Open any row for the month detail: the dollar band and percentile where the classifier has one, the dated month endpoints, the elevation count where one exists, the seasonal position where the history is deep enough, and the pressure read where one exists. No row below cleared the gate this month — every read on this board is a snapshot, not a move , and its pressure reads are associations with lead windows, not causes. Each row deep-links (this page + #board-<ingredient>).

Where every row's numbers come from

The frozen edition snapshot — week-2026-07-06.json + .csv (CC0) — plus data/cost-lockfloat.json (percentile bands), data/seasonality.json (typical-month bands, ≥2 years), and data/cost-pressure.json (lead reads, stamped 2026-06-08).

Beef — 4 items · 1 above / 0 flat / 3 below

Ribeye−3.1%$11.52/lb

$11.52/lb sits at the 97th percentile of its range; band $11.12–$11.92.

Opened $12.30 on 2026-06-09; closed $11.52 on 2026-07-02; peak $13.25 (06-23); trough $10.49 (06-26) — 18 reads, usda-lmr, wholesale.

Above its typical Jul band ($10.75–$11.37, 3 yrs).

Pressure read: building, led by cattle-on-feed placements on a 16–26 week lead.

The gate held everything back, and that is the finding

The basket reads −5.0%, and the month-over-month line stays blank

Three arc shapes: given back, still climbing, gone silent

One blueberry, three instruments, three true readings

Romaine: the fall is measured, the leading read points back up

Withheld for opposite sins

What to do this month

The full board

Looking ahead — as far as the record permits

What's building, what's easing

The August calendar

Lock or float

The one-print reach

The record

Methodology & provenance

Quick answers