This is Corner's store study in full. The short version is on one page, and there is a 78-page PDF.
The study site, from the north-east. The corner of Barton Street and Dewitt Road, April 2025. Site photography, the study’s own archive.
Craftpine · Corner · Study
520 Barton
An unattended-retail feasibility study.
A corner property in Stoney Creek, Hamilton, and the venue question inside Corner, a 76-day program. Can unattended, machine-served retail carry itself with no staff, in a small structure on a high-exposure lot?
Craftpine · TorontoJuly 2026 · This edition October 2026
Abstract
This study examines 520 Barton Street as a site for unattended, machine-served retail. It covers the site’s exposure and zoning and the concepts considered. It then sets out the unattended fresh-drink category’s public evidence base, the regulatory classification question in two jurisdictions, and a buy-or-build assessment of drink automation. It also covers the Corner software platform built alongside, and the unit economics at total capital.
The method was kill criteria set before spend. The verdict is kill. Unattended fresh-drink retail fails on demand per machine in the one case anyone has measured. The cost of proving otherwise exceeds the return available, as better-funded operators established before this program began. The corner remains a strong asset. Venue capital spent was approximately zero.
The study in one number per chapter.
The verdict rests on the demand ceiling of chapter 03 and the capital arithmetic of chapter 07, not on the site or its circulation.Numbers as each chapter states them, from the “At a glance” list on its opening page.
Findings, by chapter
01The asset is the corner. Everything else in this study tests what it can carry.
02Eight concepts entered the funnel and none survived. The program’s kill criteria were set before any capital was spent.
03The unattended fresh-drink category has exactly one measured throughput figure in its public record. This chapter is built around it.
04On paper, an unattended fresh-drink kiosk in Ontario needs a trained human standing next to it. In practice, two cities’ own open data show the paper is not how the category is policed.
05The machine question entered the study as a prediction: a doctrine-compliant drink robot exists at or under $20,000 CAD landed. The research falsified it.
06Over the program’s 76 days, one founder built Corner’s software platform and tested it to destruction.
07The zero-labour drink store clears its running costs at 3.8 cups a day. The problem is paying back the capital, at demand anyone has measured.
08A walking customer cannot reach the rear structure without crossing the lot’s vehicle circulation. That caps what the structure can hold. It is the narrowest finding in this study, a bound on one 240 square foot structure, and it is not the reason the program closed.
09Unattended fresh-drink retail fails on demand per machine in the one case anyone has measured, and the cost of proving otherwise exceeds the return available. Several far better-funded operators established that before this program began.
The asset is the corner. Everything else in this study tests what it can carry.
In this chapter
1.1Exposure
1.2Context
At a glance
90 × 160 ft
the lot, with frontage on both streets
21,000+
vehicles a day at the junction, as published
5,364 sq ft
the building, two storeys, built 1947
No. 520 from the north, its twin bays facing Barton Street, April 2025. Site photography, the study’s own archive.
Figure 1.1A Hamilton Street Railway bus on route 1A at Eastgate Square, Stoney Creek, September 2026. Five HSR routes run the Barton corridor (section 1.1).Photo: BrackishStowaway, CC BY 4.0, via Wikimedia Commons.
520 Barton Street sits at the signalized junction of Barton Street and Dewitt Road in Stoney Creek, Hamilton. The lot is 90 by 160 feet, with frontage on both streets. The building is 5,364 square feet over two storeys, built in 1947 of steel and masonry, with three vehicle bays.1
The parcel is zoned C2, Neighbourhood Commercial. Its permitted uses include motor vehicle service, restaurant, medical clinic, financial establishment, and veterinary practice.1 That wide list of permitted uses is part of why the parcel matters.
Stoney Creek, on the QEW at the west end of Lake Ontario.
Figure 1.2The site in its region. The Queen Elizabeth Way follows the lake from Toronto through Hamilton toward Niagara, and passes two kilometres from the site, via Fruitland Road. Approximate scale.Base map: Natural Earth, public domain. Drawn for this study.
One corner, two frontages: the exposure geometry.
Figure 1.3The lot fronts both streets of a signalized junction, shown here with the through-traffic and transit routes that produce its exposure. The building sits against the lot’s west line, long axis north–south, with the parking on the Dewitt side. It has twin bays on the Barton face and a third at the south end of the east face. The billboard across the intersection faces east and west over Barton’s traffic. Schematic, approximate scale.Geometry: 2007 Ontario Land Surveyor site plan of the parcel; property facts as published on the property’s website; aerial photography, April 2025 (orientation from the source files’ camera-heading EXIF). Traffic figure as published; see note 2.
Exposure
The junction carries 21,000 or more vehicles per day, as published for the property.2 Five HSR bus routes run the corridor: 1, 2, 3, 4, and 10.1 The QEW is two kilometres away via Fruitland Road, and the Red Hill Valley Parkway is three kilometres away.1Fig. 1.4 sets these distances on one scale.
Two schools within a walk, two highways within three kilometres.
Figure 1.4The corner’s reach on one scale, from the 400 and 800 metre walking rings of Fig. 1.8 out to the two highway links.Property facts and highway distances as published; see note 1. Walking rings as the crow flies. Traffic figure as published; see note 2.
Third parties price this exposure. The digital billboard across the intersection, at 521 Barton, is marketed at 21,034 daily impressions.3 Its two faces point east and west down the Barton corridor, at exactly the traffic the counts describe.
Figure 1.5Barton eastbound toward the Dewitt junction, April 2025, with the 521 Barton digital board beyond the junction at left and the subject corner at right. The board’s faces point down the corridor, at the same traffic the counts describe.Site photography, the study’s own archive.
Figure 1.6The corner from the north-east, April 2025, with a queue on Barton at the signal. Townhouse and detached residential fabric lies on three sides.Site photography, the study’s own archive.
Context
The surrounding fabric is residential on three sides, with detached houses along Dewitt and the side streets and townhouse blocks to the west. Across Barton to the north runs an industrial and commercial strip. Two schools sit within walking distance, a secondary school and an elementary school.
Figure 1.7The junction from above Barton, looking south along Dewitt Road, April 2025. The corner is left of centre, with townhouse blocks to its west, detached houses along Dewitt and the side streets, and the escarpment on the horizon.Site photography, the study’s own archive.
From the street, the building shows twin bays on the Barton face and a third at the south end of the east face. A parking apron wraps its east and south sides, and the lot has three curb cuts, one on Barton and two on Dewitt (Fig. 1.3).
The subject of this study is the small structure at the rear of the lot, 12 by 20 feet.
Notes
Property facts (lot, building, zoning and permitted uses, transit routes, highway distances) as published on the property’s website, accessed 2026-07-14 and 2026-07-15.
21,000+ daily vehicles: as published for the property, reported here as published, not independently measured. A City of Hamilton open-data traffic count for this junction was searched for and not found, and no estimate is substituted.
21,034 daily impressions: the billboard vendor’s listed figure for the 521 Barton Street digital board (Vendo Media). The vendor’s per-board listing page was not directly fetchable at the 2026-07-15 access attempt; the figure is reported as listed by the vendor.
Chapter 02
The concept space
Eight concepts entered the funnel and none survived. The program’s kill criteria were set before any capital was spent.
In this chapter
2.1What each concept met
2.2The kill criteria
At a glance
8
concepts entered the funnel
0
survived
7
numbered kill criteria on the fresh-drink format, set before capital
An independent convenience store in Toronto. The program began as software for neighbourhood retail, then narrowed to what one site could carry. Photo: Chicken4War, CC BY-SA 4.0, via Wikimedia Commons.
The program, Corner, ran from late April to mid July 2026. It began as a software platform for neighbourhood retail, narrowed to what the site itself could carry, and ended on the day the last candidate use was falsified. Each concept below is listed with what killed it, because the kill reasons are the study’s findings.
Eight concepts entered. What killed each one, and when.
CPG micro-retailParked. Payment processing takes about 15% of a $2.50 basket, at any volume.Apr 30 → Jul 13
Staffed bubble teaRejected the same day it was proposed. It set the rule of no staff in store operations.Jul 13
Unattended bubble teaPivoted. Tapioca needs a heated hold, the exact burden an unattended format cannot carry lightly.Jul 13 → 14
Unattended smoothieKilled. The one measured figure in the category is 6.8 cups a day, against a 12-cup floor. Capital of $26K to $37.6K pays back in 2.2 to 3.3 years, past the 24-month line. Better-funded operators ran it first and lost.Jul 14
Prepackaged cold drinksPassed every format test and was never falsified. It closed with the program and is the most revivable line.Jul 13 → 14
Venue-placed unitsClosed. Where these machines survive, venue service fees carry them, not cup margins.Jul 14
Private training venueKilled on paper, on the change-of-use permit and assembly-occupancy obligations.Jul 14
Micro officeConsidered, then closed with the venue lines. The record holds no dated entry for it, and none is invented.undated → Jul 14
Figure 2.1The concept record. Every exit landed on July 13 or 14, when the evidence arrived together. One line, in blue, passed its own format tests and closed with the program.Program ledger and tracker, close-out record, July 2026.
What each concept met
CPG micro-retail. The founding concept: consumer packaged goods on a shelf, transparent platform pricing, and checkout by phone. Payment processing on a $2.50 convenience basket takes roughly 15 percent of revenue, at any volume, so the concept was parked. The platform shipped and works. The format waits for a higher-ticket application.
Staffed bubble tea. It was recommended to the program and rejected the same day, and the rejection became standing doctrine: no human staff in store operations, ever. A format that only works with a shift worker is a different business from the one under study.
Unattended bubble tea, then smoothie. These were the core candidates. Bubble tea fell first, on a food-safety mechanic. Tapioca requires a heated hold, and a heated hold of a time-and-temperature-controlled food is exactly the obligation an unattended format is least able to meet. The pivot target was a smoothie format built on individually quick-frozen fruit, which deletes that burden.
The food-safety mechanic between the two core candidates.
Figure 2.2Left: tapioca pearls at the bottom of a cup of bubble tea. Tapioca needs a heated hold, the burden an unattended format is least able to carry. Right: frozen berries. The smoothie format was built on individually quick-frozen fruit, which deletes that burden.Photos: left, sam651030, CC0; right, I saw_that, public domain; both via Wikimedia Commons.
The smoothie format died on July 14 anyway, on three findings (Fig. 2.3). Measured category demand sits below the model’s floor (chapter 03). Total capital does not pay back inside the 24-month kill line (chapter 07). Better-funded operators had already run the experiment and lost (Appendix D).
The fresh-drink line: two pivots in two days, then three findings against the last candidate.
Figure 2.3Each pivot removed the obstacle that ended the concept before it, and the smoothie then fell on three separate findings.Dates: program ledger, July 2026. Findings: chapters 03 and 07, Appendix D.
Figure 2.4Prepackaged cold drinks, unattended: beverage vending machines in Wadabori Park, Suginami, Tokyo, May 2024.Photo: Ximonic (Simo Räsänen), CC BY-SA 4.0, via Wikimedia Commons.
Prepackaged premium cold drinks. The low-risk format passed every regulatory and machine test in this study. It closed with the program and was never falsified. Of everything in the funnel, it remains the most revivable.
Venue-placed units, and selling the machines. Both were floated on the final day and closed by the category evidence. Where unattended drink machines survive, venue service fees carry them rather than cup margins. Operating placed units is a real business, but it is a services business with a sales motion, and the studio was not going to bolt that on to rescue one store concept.
The format at incumbent scale, in both placements.
Figure 2.5Left: a street-facing Żabka Nano checkout-free store in Katowice, Poland. Right: a venue-placed unit inside the Warszawa Gdańska rail station concourse, 2025. Europe’s largest autonomous-store chain runs both placements. The venue-placed unit is the shape the category evidence favours.Photos: left, Jonathan Deamer, CC BY 4.0; right, Cybularny, CC0; both via Wikimedia Commons.
Private training venue. This was the last non-retail reuse, and it failed on paper before it failed on geometry (Fig. 2.6). A change-of-use permit applies even without construction. A fitness use classifies as assembly occupancy under the building code, and a 240 square foot unplumbed structure cannot meet the washroom and barrier-free obligations that classification brings without disproportionate spend.1 Client access would still cross the vehicle plane, the paved area the lot’s vehicles use, every hour (chapter 08). The local market for hourly training space is also already served.
Why the private training venue failed on paper.
Figure 2.6A fitness use brings a permit and a building-code class whose obligations the unplumbed structure cannot meet without disproportionate spend.City of Hamilton Building Permit By-law 15-058; Ontario Building Code Group A-2, as assessed in the program’s close-out record, July 2026.
Micro office. Considered, then closed with the venue lines. The record gives it no date.
The kill criteria
The fresh-drink format carried seven numbered criteria, set while the machine was still a spreadsheet line. Five are worth reading in full, because they did the killing or would have.
Table 2.1The kill criteria, verbatim from the program record. Read each row as: if this happens, do that.
#
If this happens
Then
K1
Counted demand below 6 cups/day after the minimum sample; a go requires a counted base case of 12 or more
Re-site or re-product. Do not buy the machine.
K2
No doctrine-compliant machine at or under $20K CAD landed, and no DIY build at or under $15K in parts
Wait or descope. The rules are not negotiable to make a purchase happen.
K5
Machine availability under 95% in the first 60 days, any health-unit stop order, or more than two verified made-wrong incidents
Halt scale-out. Re-engineer before a second store.
K6
Winter net below zero and trailing payback projecting past 24 months
Format kill. The structure reverts to storage.
K7
The public-health unit classifies the operation as a food service premise requiring an on-site certified food handler, and no low-risk menu redesign escapes the classification
Format kill for fresh drinks at this site. Doctrine forbids staffing around it.
Research falsified K2 before any order was placed. No compliant machine lands under $20K (chapter 05). K1 was never run, because the program closed before the September cup count. K7 was never answered, and chapter 04 sets out the public evidence on the classification question.
What the kill criteria tested, and what closed the program.
Figure 2.7The two facts that closed the program sat outside the criteria set in advance, and both were public the whole time.Program record and Table 2.1; close-out record, July 14, 2026; chapters 03, 04, 05, 07 and 08, Appendix D.
What closed the program was not on this list at all. It was the category’s own published record and the total-capital arithmetic, both of them free and public the whole time. Walk-up circulation was not on the list either. Chapter 08 records it as a bound on one structure, and it moved none of the arithmetic. The method note in Appendix C carries the amendment that follows: read the category’s record, and price what it costs to prove a model, before building the model.
Notes
City of Hamilton Building Permit By-law 15-058 (change-of-use permit requirement); Ontario Building Code Group A-2 assembly classification and its washroom and barrier-free obligations, as assessed in the program’s close-out record, July 2026.
Chapter 03
Category analysis: unattended fresh-drink retail
The unattended fresh-drink category has exactly one measured throughput figure in its public record. This chapter is built around it.
In this chapter
3.1Throughput: the one real number
3.2Where the units live
3.3The economics the record supports
3.4The patent shadow
3.5The adjacent evidence
At a glance
6.8
cups a day: the category’s one measured throughput figure
12
cups a day: what this study’s cheapest case needs
~US$420,000
of confirmed capital per placed unit
Chicago, from the 16th Street bridge. The category’s best-documented operator is a Chicago company. Photo: John Picken, CC BY 2.0, via Wikimedia Commons.
The best-documented operator in the category is Zenblen, a Chicago company running robotic smoothie kiosks. This study presents it as the category’s best evidence, not as a target. It is the operator that has survived, iterated, raised institutional money, and placed units, and its public record is the most favourable data available on how the category performs.
Throughput: the one real number
In a 2024 interview, Zenblen’s founder stated the company “served over 2500 smoothies” from its field unit in 2023.1 Over a full year that is roughly 6.8 cups a day.
The unit sits inside a members-only innovation hub, among a captive and friendly population of habitual users. The hub’s own blog places the kiosk on site by May 2022, the year before the counted period, so the 6.8 reads as a mature rate rather than a ramp-up undercount.2 No one in the category has published a later or better figure, in any climate or venue class.
The demand gap: the category’s best measured unit serves 6.8 cups a day. This study’s cheapest case needs 12.
Figure 3.1The floor case, the minimum the smallest capital stack needs, sits 75 percent above the best measured mature rate anywhere in the category’s public record.6.8/day: 2,500 drinks in calendar 2023, operator interview (Ottomate News); scenarios: this study, chapter 07.
Where the units live
Zenblen lists eleven kiosks, in university buildings, hospital towers, a members-only innovation hub, a residential building’s gym, and a pilates studio.3 A check of each venue’s own published hours and access rules yields a clean structural finding: every unit sits inside an access-controlled or hours-limited building. None is street-facing, and none can be reached by a member of the public walking up off a sidewalk, which is the use an unattended machine nominally exists for.3
What the category leader’s placement looks like.
Figure 3.2A Zenblen kiosk in a Chicago building lobby, beside the security turnstiles: indoors, access-controlled and procurement-placed. The full company record is Appendix D.Photo: Polsky Center for Entrepreneurship and Innovation, University of Chicago (2025).
Placement follows procurement rather than retail siting. At two universities the kiosks arrived through the incumbent campus vending contractor, a division of Compass Group. In the campus paper it is the contractor’s regional director, not the startup, who speaks for the expansion decision.4 A third campus lists the kiosk as a line item in the vending contractor’s procurement bid.5 The category leader’s distribution runs on the vending industry’s existing rails (Fig. 3.3).
Where the eleven kiosks sit, and how the campus units got there.
Figure 3.3All eleven listed units sit inside access-controlled or hours-limited buildings, and on the three campuses in the record the kiosk runs through the vending contractor’s procurement.zenblen.com/find-a-kiosk and each venue’s published hours, accessed 2026-07-14; The Daily Northwestern, 2026-05-26; College of DuPage vending solicitation, June 2025.
The economics the record supports
On stage in 2026, the founder described two revenue lines: “we get revenue both from our location partners in terms of monthly service fee and we also get retail revenue from each smoothie we sold.”6 No dollar figure has been disclosed for either line. The venue pays for the amenity. The retail price point at one campus is US$8 per smoothie, per the student paper.4
The same pitch claims over 550 percent contribution margin, a six-month payback per kiosk, and 99 percent reliability. All three are the operator’s own unaudited statements with no disclosed dollar basis, and are labelled here accordingly.6
Two revenue lines: the disclosed one is small, and the undisclosed one carries the business.
Revenue line 1 · retail cups
US$8 a cup
one campus venue, 2026
The one published count, at a different kiosk, is about 6.8 cups a day. At this price that is about US$54 a day of retail revenue, before costs.
Revenue line 2 · venue service fee
a monthly fee
paid by the location partner, amount undisclosed
The venue pays for the amenity. This line, not the cups, is what carries the business.
Figure 3.4The category’s revenue structure, as its best-documented operator describes it. The measured operators are venue-amenity businesses with a retail line, not standalone stores. The operator’s own unaudited pitch claims (“over 550% contribution margin,” “six-month payback,” “99%+ reliability”) disclose no dollar basis and are labelled in the text.Operator’s own statements (pitch, 2026; interview, 2024); SEC Form D filings, 2023 and 2026; campus press, 2026. Self-reported figures labelled.
The capital side is better documented. Two SEC Form D filings (notices of exempt securities offerings) confirm US$2,117,117 sold in the 2023 round and US$2,480,647 in the 2025–26 round. That is a floor of US$4.6 million across the two, against eleven placed units at the time of this audit.7
In the same 2026 pitch, the founder described a US$6 million Series A as “closing.” No filing or press confirmation existed at this study’s access dates.6 The ratio the public record supports is roughly US$420,000 of confirmed capital per placed unit (Fig. 3.5).
The capital the record confirms: a US$4.6 million floor across eleven placed units.
Figure 3.5Split across the eleven units, the confirmed floor comes to roughly US$420,000 each, and the US$6 million Series A had no filing or press confirmation at the access dates.SEC EDGAR, Zenblen, Inc. (CIK 0001993331), Form D filings of 2023-09-18 and 2026-01-21; Series A as described in the 2026 pitch. All accessed 2026-07-14.
The patent shadow
The blend mechanism itself is not open ground. F’real Foods holds two live United States patents on in-cup blending with an elevator assembly and removable spindle, expiring 2033 and 2037.8 A third patent often casually attributed to f’real, US10682942B2, in fact belongs to Tikiz Franchising and covers mobile-kiosk dispensing hardware. The correction matters because the misattribution circulates.8 Zenblen’s own application was allowed but not yet granted at this study’s access date.9
The patent shadow runs from the study date to 2037.
Figure 3.6At the study date both of f’real’s live in-cup blending patents had years to run, and f’real had already won a verdict on earlier ones.patents.google.com, accessed 2026-07-14; verdict per Law360 (2019-05-03) and Morris Nichols (2021-07-21), accessed 2026-10-02; Zenblen’s application rechecked 2026-10-02.
Whether any given built machine infringes the live claims is a freedom-to-operate question that the public record cannot resolve. The record does establish that a new entrant’s cheapest mechanical path runs near claims in force until 2037, held by a patentee that has won an infringement verdict on its earlier blender patents.8
The patent shadow, in the patentee’s own drawings.
Figure 3.7Sheets from US 9,579,615 B2, “Blender with elevator assembly and removable spindle,” granted to f’real Foods on February 28, 2017, one of the two live patents above. From left: the machine with a cup in its holder (Fig. 2A), and the elevator assembly at three positions (Figs. 10 to 12).US 9,579,615 B2, sheets 3 and 11 to 13. United States Patent and Trademark Office.
The adjacent evidence
The two nearest comparators point the same way. teaBOT, the automated tea kiosk company, remains active after a decade. Its chief executive also serves as Zenblen’s CTO, per his own public profile, and appears on Zenblen’s first Form D but not its second.10
Blendid, the robotic smoothie company, raised venture capital for an eight-by-eight-foot robot-arm format. Its FY2025 annual report disclosed substantial doubt about continuing as a going concern if revenue targets are not met, thirteen employees, and a US$308,000 obsolete-inventory write-off. At the end of 2024 it stopped selling its original kiosk for two new products, a smaller, cheaper kiosk and a back-of-house unit, to be sold rather than leased.11
What Blendid changed at the end of 2024, and what its FY2025 report disclosed.
Figure 3.8The original robot-arm kiosk gave way to two new products to be sold rather than leased, and the FY2025 annual report disclosed going-concern doubt if revenue targets are not met.6d bytes inc. (Blendid) FY2025 annual report, filed 2026-03-02, SEC EDGAR CIK 1830909, accessed 2026-07-14.
The direction of travel among the category’s established operators is uniform. Machines are getting smaller and cheaper, and operators place them in venues on service fees instead of on streets to earn cup margins.
Figure 3.9The Jamba by Blendid co-branded kiosk inside a Walmart in Dixon, California: the eight-by-eight-foot robot-arm format Blendid stopped selling at the end of 2024.Photo: Blendid, via The Spoon.
What the category’s public record supports: unattended fresh-drink machines survive as venue amenities on service-fee economics, inside access-controlled buildings, placed through vending-industry procurement.
What no public record supports: a street-facing unattended fresh-drink store carrying itself on cup margins. Nobody has published a case, and the one measured throughput figure is below every capital stack’s viability line in this study.
Notes
Ottomate News, “Zenblen brings robo-crafted smoothies” (founder interview), ottomate.news/p/zenblen-brings-robo-crafted-smoothies, accessed 2026-07-14. The 2,500-smoothie figure is the operator’s own; 6.8/day is the naive full-year division.
mHUB blog, “Hardtech startup Zenblen inspires healthy consumption,” published 2022-05-23, mhub.org, accessed 2026-07-14; IIT Institute of Design article, published 2020-11-05, iit.edu, accessed 2026-07-14.
zenblen.com/find-a-kiosk, accessed 2026-07-14; each venue’s access reality checked against the venue’s own published hours pages (rush.edu, lib.uchicago.edu, mhubchicago.com, rentnemachicago.com, onyxpilates.com, and others), all accessed 2026-07-14.
The Daily Northwestern, “Second Zenblen smoothie machine installed in Main Library,” 2026-05-26, accessed 2026-07-14. Canteen (Compass Group) senior regional director quoted on the expansion decision; US$8 price point per a student quoted in the same article.
College of DuPage vending-services procurement solicitation, June 2025, cod.edu (PDF), indexed via search snippet, accessed 2026-07-14.
mHUB HardTech Summit pitch (video uploaded 2026-03-25), youtube.com/watch?v=wW6zQRT78pA at 41:28–44:18, auto-caption transcription, accessed 2026-07-14. All figures in that segment are the operator’s own unaudited statements.
SEC EDGAR, Zenblen, Inc. (CIK 0001993331), Form D filed 2023-09-18 (US$2,117,117 sold of a US$2,840,275 offering, 34 investors) and Form D filed 2026-01-21 (US$2,480,647 sold, offering indefinite, 12 investors), both accessed 2026-07-14. Commonly cited broker figures (US$2.84M, US$3.55M) understate the SEC-confirmed total.
US9579615B2 (expires 2033) and US10299628B2 (expires 2037), assignee F’real Foods LLC; US10682942B2, assignee Tikiz Franchising, LLC. patents.google.com, all accessed 2026-07-14. The verdict: f’real Foods and Rich Products v. Hamilton Beach Brands and Hershey Creamery (D. Del.), a jury verdict of May 2019 on three earlier f’real patents, affirmed by the Federal Circuit on 2021-07-21 (Law360, 2019-05-03; Morris Nichols, 2021-07-21), accessed 2026-10-02.
US20220160162A1, Zenblen, Inc., legal-events table showing the application allowed, with the notice of allowance not yet mailed, as its last indexed event (2026-04-12). patents.google.com, accessed 2026-07-14. Rechecked 2026-10-02: a later entry records the notice as mailed on 2026-04-14, and the application is still listed as pending.
LinkedIn public profile headline (“CEO @ teaBOT + CTO @ Zenblen”), ca.linkedin.com/in/rehmanmerali, accessed 2026-07-14; presence on the 2023 Form D and absence from the 2026 Form D per note 7.
6d bytes inc. (Blendid) FY2025 annual report (year ended 2025-10-31, filed 2026-03-02), SEC EDGAR CIK 1830909, accessed 2026-07-14. Going-concern language, headcount, write-offs, and the product transition are the filing’s own disclosures.
Chapter 04
Regulatory posture
On paper, an unattended fresh-drink kiosk in Ontario needs a trained human standing next to it. In practice, two cities’ own open data show the paper is not how the category is policed.
In this chapter
4.1Ontario, on paper
4.2Toronto, in practice
4.3Chicago, in practice
At a glance
104,790
inspection records in Toronto’s DineSafe dataset
0
records for the unattended espresso operator, across nearly six years
26
logged queries against Chicago’s open data
The Ontario Legislative Building at Queen’s Park, Toronto. O. Reg. 493/17, the province’s Food Premises regulation, is made under the Health Protection and Promotion Act. Photo: Benson Kua, CC BY-SA 2.0, via Wikimedia Commons.
Ontario, on paper
Ontario’s Food Premises regulation, O. Reg. 493/17, defines a “food service premise” as any food premise where meals or meal portions are prepared for immediate consumption. The definition is agnostic to who, or what, does the preparing.1 A “food handler,” by contrast, is defined only as a person, and the regulation contemplates no machine substitute.1 Vending machines appear in the text as a species of equipment, not as an alternative premises class.1
Two sections then pull in opposite directions. Section 17, an equipment-sanitation rule, plainly contemplates lawful unattended operation. A machine that auto-mixes water needs a piped potable supply, and the operator’s name and phone number must be posted “if an employee of the operator is not in full-time attendance.”1
Section 32 points the other way. It requires every operator of a food service premise to keep at least one trained food handler or supervisor on the premise “during every hour in which the premise is operating.”1 A fresh drink blended from dairy and frozen fruit is a time-and-temperature-controlled food, which forecloses the regulation’s low-risk exemption, section 3.1, on its face.1
The classification question is therefore exact. Is the kiosk a food service premise, which section 32 staffs, or equipment inside a premise, which section 17 sanitizes? The regulation’s text does not resolve it.
Read as a food service premise, the kiosk needs a trained person on the premise every operating hour. Read as equipment, it may lawfully run unattended.
Figure 4.1O. Reg. 493/17 read against the kiosk, one branch for each reading. The text does not say which branch applies, and this study assumes Ontario’s text will be enforced as written.O. Reg. 493/17, ss. 1, 3.1, 17 and 32; Ontario Ministry of Health, Food Premises Reference Document (2019), p. 18. Notes 1 and 2.
The Ministry’s own reference document acknowledges that many vending machines “operate automatically without employee assistance.” It affirms that all other duties still apply, and it leaves section 32 untouched.2
Toronto, in practice
Figure 4.2Toronto City Hall and Nathan Phillips Square. The DineSafe dataset is the City of Toronto’s own open data.Photo: Taxiarchos228, CC BY 3.0, via Wikimedia Commons.
Toronto’s DineSafe open dataset held 104,790 inspection records at access.3 RC Coffee has operated unattended robotic espresso kiosks, fresh dairy and cashless, since 2020, and under its own Robo Cafe brand since June 2021. At access it listed twelve or more locations, major hospitals among them. The dataset holds no record of it under any searchable name.3
Caffeo, a directly comparable unattended fresh-dairy espresso kiosk that opened in September 2025, is listed. It passed a Food Premises inspection on September 12, 2025.3 So the same city treats the same class of operation both ways, and Toronto’s enforcement of the category is unsettled.
Figure 4.3The two Toronto operators of the comparison. Left, RC Coffee’s Robo Cafe at its 2021 launch: the unattended fresh-dairy kiosk with no DineSafe record under any searchable name. A DineSafe placard is visible in the neighbouring storefront’s window. Right, Caffeo at 405 Richmond Street West: the directly comparable kiosk that is listed and passed a Food Premises inspection on September 12, 2025.Photos: RC Coffee / TechTrex via Newsfile (press release); Dustin Fuhs, 6ix Retail.
Chicago, in practice
Figure 4.4Chicago City Hall on LaSalle Street. The city’s code routes automatic food vending through a retail-food-establishment license.Photo: ajay_suresh, CC BY 2.0, via Wikimedia Commons.
Chicago’s municipal code anticipates this exact business model. Section 4-8-032 defines an automatic-food-vending-machine operation and routes it through a retail-food-establishment license. The route includes health-department approval of the machine type and inspection of the commissary and servicing facilities.4 On paper, then, an operator’s dispensing locations carry licenses.
The study queried the city’s open-data API directly at all nine of Zenblen’s Chicago venue addresses, across both the food-inspections and business-licenses datasets and over multiple spelling variants. The queries returned zero records.5 The datasets are not blind at those addresses. The same queries return the co-located food operators, individually licensed and inspected, from campus dining and a chain bakery to the vending contractor’s own micro-markets (Fig. 4.5).5
Two of the three unattended operators are missing from their cities’ own food records. One is listed, and the conventional vendors beside them appear as usual.
Toronto · DineSafe open data, 104,790 records
RC Coffeeunattended robotic espresso · fresh dairy · since 2020 · 12+ locations0 recordsacross nearly 6 years of operation
Caffeosame class of operation · opened September 2025listed · passedFood Premises inspection, 2025-09-12
Chicago · food inspections + business licenses, 26 logged queries
Zenblensmoothie kiosks · fresh dairy · queried at all 9 venue addresses, 2021 to 20260 recordsboth datasets, every spelling variant
Co-located food vendorscafés, a chain bakery, campus dining, the vending contractor, same addresseslicensed · inspectedindividually listed in the same datasets
Figure 4.5Two jurisdictions’ own open data, queried directly, with methods in the notes. The datasets are not blind, since the conventional food vendors co-located at the same addresses appear normally. One downtown “shared kitchen user” license exists for the category leader, at an address unconnected to any venue, renewed through January 2027.City of Toronto DineSafe open dataset; City of Chicago Food Inspections (4ijn-s7e5) and Business Licenses (r5kz-chrr) via the Socrata API, 26 logged queries, accessed 2026-07-14.
One Zenblen license does exist. It is a shared-kitchen-user license at a downtown address unconnected to any venue, renewed through January 2027.5 After five years of operation there is no dispensing-location record, though the code requires one.
Illinois’ statutory language is also softer than Ontario’s. The state food-handling act requires food service establishments to be under a certified manager’s “operational supervision.” That wording plausibly permits one manager across several unattended sites, where Ontario’s section 32 says a person on the premise, every operating hour.6 The federal model code’s own working group put the strict reading plainly: an unattended operation selling anything beyond prepackaged, non-hazardous food “would be required to have a person-in-charge at all hours of operation.”7
What each text asks of an unattended kiosk, beside what each city’s own records show.
Figure 4.6For each place, the arrow runs from what its text asks to what its city’s own records show. In blue, the one text this study assumes will be enforced as written.O. Reg. 493/17, s. 32; Chicago Municipal Code §4-8-032; 410 ILCS 625/3; City of Toronto DineSafe and City of Chicago open data. Notes 1 and 3 to 6.
The finding is regulatory lag, not wrongdoing. Enforcement practice in two large cities has not caught up to a category their own codes already describe.
Neither city’s practice is evidence that Ontario’s stricter text will not be applied in Ontario. Toronto has already inspected one unattended kiosk. The study therefore assumes Ontario’s text will be enforced as written.
Notes
O. Reg. 493/17 (Food Premises), under the Health Protection and Promotion Act: s.1 definitions, s.3.1 exemption, s.17, s.32. ontario.ca/laws/regulation/170493 and canlii.org, accessed 2026-07-14. Statutory text cross-checked two ways (rendered e-Laws fetch; text extraction of the Ministry reference document, which reproduces the definitions verbatim).
Ontario Ministry of Health, Food Premises Reference Document (2019), p. 18, on s.17, accessed 2026-07-14.
City of Toronto DineSafe open dataset, 104,790 records at access, parsed directly, 2026-07-14. RC Coffee locations per rccoffee.com, accessed 2026-07-14; its first kiosk per a Newsfile release of 2020-10-07. RC Coffee searched again on 2026-10-02 in the current dataset and the historical files for 2001 to 2023: no record. Caffeo (405 Richmond St W) inspection pass dated 2025-09-12 in the same dataset.
Chicago Municipal Code §4-8-032 (automatic food vending machine operations, licensed through retail food establishments), accessed via the code’s indexed text, 2026-07-14.
City of Chicago open data via the Socrata API: Food Inspections dataset 4ijn-s7e5 and Business Licenses dataset r5kz-chrr; 26 queries with full raw responses archived, 2026-07-14. Co-located examples include license 2796844 (chain bakery at a hospital address) and licenses 2877822 and 2882530 (the vending contractor’s café and micro-market at a hospital tower). The shared-kitchen-user license is 2762826, current renewal expiring 2027-01-15.
410 ILCS 625/3 (Food Handling Regulation Enforcement Act), ilga.gov, accessed 2026-07-14. Illinois adopts the FDA Food Code (2022) as the Illinois Food Code, dph.illinois.gov, accessed 2026-07-14.
Conference for Food Protection, “Unattended Food Establishments” issue paper (2014 packet), foodprotect.org, accessed 2026-07-14.
Chapter 05
Automation assessment
The machine question entered the study as a prediction: a doctrine-compliant drink robot exists at or under $20,000 CAD landed. The research falsified it.
In this chapter
5.1The landed-cost arithmetic
5.2The payment filter
5.3The component build
5.4Reliability is the bar
At a glance
$26,500–28,400
a robot-arm machine, landed in Ontario, CAD
30–40%
the understatement in the sticker
4–6 months
from purchase order to legally operable
Container cranes at the Centerm terminal, Port of Vancouver, 2022. A machine priced free-on-board overseas still carries ocean freight, duty, tax, brokerage and an electrical field evaluation before it can legally operate in Ontario. Photo: Dietmar Rabich, CC BY-SA 4.0, via Wikimedia Commons.
The landed-cost arithmetic
Robot-arm bubble-tea machines list on overseas marketplaces at US$12,800 to $15,900. That converts to roughly $18,000 to $22,500 CAD and appears to clear the study’s $20,000 ceiling.1 Once the machine is landed in Ontario, it does not.
A worked mid-range example at US$14,350 free-on-board is $20,305 CAD before the boat. Ocean freight adds $600 to $1,500. Import duty adds zero to about $1,700, depending on an unresolved tariff-heading question. Tax and brokerage add about $1,300. None of the researched Chinese units carries a Canadian electrical approval, so an Electrical Safety Authority field evaluation adds an estimated $2,000 to $6,000 more.2
The worked landed estimate is roughly $26,500 to $28,400 CAD, so the sticker understates the cost by 30 to 40 percent. The realistic timeline from purchase order to legally operable is four to six months.2 The ladder in Fig. 5.1 sets the landed machine beside the other options priced, and Fig. 5.2 walks the example one cost at a time.
The machine cost ladder, cheapest to dearest. Only the component build lands compliant under the $20K ceiling.
Figure 5.1Sticker rungs are what listings imply, and the landed rung is what a unit costs in Ontario, legally operable. The dashed line is the study’s $20K kill threshold (criterion K2).Vendor listings and worked landed-cost example, prices researched 2026-07-13/14, FX 1 USD = 1.415 CAD; full source list in Appendix B.
From sticker to legally operable: what the listing leaves out.
Figure 5.2The robot-arm rung of the ladder, opened up one cost at a time for the worked mid-range example. The listing prices the machine at the factory gate. Four costs and four to six months stand between that price and a unit legally operable in Ontario.Worked landed-cost example, note 2; FX 1 USD = 1.415 CAD (note 1).
The payment filter
One binary filter comes before price: who is the merchant of record, the business in whose name the shopper’s card is charged. The studio’s standing rule is that the operator owns the payment flow. Most overseas vending hardware is built for sealed domestic payment rails and bakes the reader into the machine. There is no legitimate way to intercept a sealed proprietary payment path without the vendor’s cooperation.3
The compliant integration shape is a documented bus or dispense-trigger interface, which is how third-party vending processors integrate.3 Of everything researched, exactly one vendor claims third-party point-of-sale openness in its marketing, and the claim is unverified. The rest are unconfirmed or lean toward sealed.1 A component-level build passes the filter by construction, because every interface in it is the operator’s own.
A sealed reader fails the payment filter, and a documented interface passes it.
Figure 5.3The two integration shapes of this section, from the shopper’s card to the filter’s verdict. Most overseas vending hardware takes the upper path, and a component build takes the lower by construction.Notes 1 and 3.
The payment filter, in two machines.
Figure 5.4Left: a Coca-Cola vending machine at Xietang Station, Suzhou, China, its reader built into the cabinet, the shape most overseas hardware takes. Right: a third-party card and mobile-payment reader fitted to a soda machine, the documented-interface path third-party vending processors use.Photos: left, Shwangtianyuan, CC BY-SA 4.0; right, Tony Webster, CC BY 2.0; both via Wikimedia Commons.
The component build
The reference design prices a fresh-drink machine as a set of individually certified commercial food-service parts. The set is a merchandiser fridge, a chest freezer, food-grade dosing pumps, a hot-hold unit, a cup sealer, stainless prep surfaces, an enclosure, and the actuation electronics an edge computer drives directly (Fig. 5.5). Hardware totals roughly $8,400 to $20,000 CAD, with a midpoint near $14,000, priced line by line in Table 5.1.4
The build has two structural advantages beyond price. First, its parts already hold recognized food-equipment certifications individually, a materially cleaner compliance story than a sealed imported box with none. Second, its integration surface is plain relays and drivers, with no vendor protocol to reverse-engineer.4
The component build as one machine: certified parts, the operator’s own interfaces.
Figure 5.5The subsystems of the reference design, grouped by the job each does. The edge computer drives the actuation electronics directly, through plain relays and drivers. The least mature line, in blue, is the frozen-fruit dosing mechanism, which is custom fabrication.Reference design and bill of materials, note 4.
Table 5.1The component build, priced line by line. The least mature line is the frozen-fruit dosing mechanism, custom fabrication with no off-the-shelf part found.
Subsystem
Range, CAD
Dairy/perishables fridge
$1,500–3,000
IQF fruit chest freezer
$1,610–2,500
Chassis / food-zone enclosure *
$1,500–4,000
Peristaltic pumps × 4 *
$600–1,800
Contingency (~10%)
$700–1,700
Water line-in + drain fittings *
$500–1,500
NSF stainless prep tables × 2
$420–1,800
Diaphragm pump (purée) *
$400–900
IQF dosing mechanism (custom) *
$300–800
Actuation electronics *
$300–800
Cup sealer
$260–400
Cup dispenser *
$150–400
Tapioca hot-hold unit *
$150–350
Ice machine
$0 · chilled, not iced, by design
Subtotal, hardware only
~$8,400–20,000
* Vendor capability confirmed, price quote-only; range estimated against the comparable market. The midpoint of the subtotal is roughly $14,000.Reference design bill of materials, sources accessed 2026-07-14.
Reliability is the bar
The category’s corporate history says a demonstration machine is the easy part. Machines with no human backstop depend on uptime, restock cadence, and a permit officer’s classification.
Bobacino raised about US$3.5 million for a robotic boba bar, showed it at trade shows, and closed without shipping a unit. Its chief executive resigned in November 2023, and by January 2024 it was reported shut down with other companies of its venture studio.5 Tea Industries, the first robot bubble-tea shop in British Columbia, opened in March 2024 and closed in 2025.5
Bobacino never shipped its bar, and Tea Industries closed its shop within two years of opening.
Figure 5.6Left: Bobacino’s demonstration bar on the floor of the World Tea Conference + Expo, Las Vegas, March 2022. Right: the Tea Industries storefront on No. 3 Road, Richmond, British Columbia, in December 2025, its sign still up over papered windows and lease notices. The shop had opened in March 2024.Photos: left, Aaron Kiel, World Tea News; right, Marco Ovies, Dished (Daily Hive).
Cafe X closed its downtown San Francisco robot-coffee locations in January 2020 and retreated to airports. Crunchbase News reported that its founder wanted the company to be “less robot, more old-school coffee.” The pandemic closed the airport bars that year. Cafe X reopened at San Francisco’s airport in 2021, where it runs three today.5
Figure 5.7Cafe X’s robot-arm coffee bar, the downtown format it later closed. Company-by-company records, with what each raised and how each ended, are in Appendix D.Photo: Cafe X press imagery.
Briggo, which reported US$26 million of stock and note sales across a decade, exited by acquisition into a coffee incumbent.5Fig. 5.8 puts all four companies on one time axis.
Since 2020, two of these four companies have closed, one was acquired, and one retreated to airports.
Figure 5.8The four companies of this section on one time axis, with the dates the text gives. A line that starts at the left edge began before 2020. Where the text gives only a year, the point within it follows the lifelines of Appendix D.Note 5 and the caption of Fig. 5.6; lifelines as in Fig. D.2.
At the study’s $15–20K tier, build, do not buy. No researched machine lands compliant under $20K. The component build is the only path that clears cost, payment ownership, and the certification story at once.
The machine was never the binding constraint, however. Chapters 03 and 07, and the category record of Appendix D, are where the format failed.
Notes
Overseas marketplace listings for robot-arm and elevator-style bubble-tea machines (alibaba.com showrooms), and vendor pages for the named platforms (Smoodi US$14,999 outright; a smoothie-vending OEM at US$9,500 FOB; a Taiwanese vendor claiming point-of-sale API openness), all accessed 2026-07-14. FX at 1 USD = 1.415 CAD, Bank of Canada-range market rate for the week of 2026-07-07.
Worked landed-cost example: LCL ocean freight $80–200 per cubic metre; duty scenarios under tariff headings 8476.89 (free) versus 8476.21 (unconfirmed) with no Canada–China surtax applying to this equipment class; 5% GST on the duty-paid value; brokerage $175–230; an Electrical Safety Authority field evaluation under SPE-1000, no published price, estimated $2,000–6,000. Sources: tariffnumber.com, ghy.com, esasafe.com (fees revised 2026-04-01), all accessed 2026-07-14. The researched Chinese units carry CE/CB/ISO marks only, none a Canadian electrical approval.
Multi-drop bus (MDB) as the documented integration path for third-party vending payment (nayax.com glossary, accessed 2026-07-14); dominance of sealed domestic mobile-payment rails in the exporting market (oceanpayment.com, accessed 2026-07-14).
Component reference design and bill of materials: commercial merchandiser fridge, chest freezer (gofoodservice.com), NSF-certified peristaltic dosing pumps (dispensingcomponents.com), food-contact diaphragm pumps (graco.com, arozone.com), pearl warmer, semi-automatic cup sealer (lollicupstore.com), NSF prep tables (katom.com), all accessed 2026-07-14. NSF/ANSI 18 covers component-level dispensing equipment (webstore.ansi.org, accessed 2026-07-14).
Tea Industries closure, Daily Hive; Cafe X retreat, The Spoon, Axios and Crunchbase News (January 2020); Briggo acquisition, Daily Coffee News (November 2020); accessed 2026-07-14. Bobacino, SEC Forms 1-SA and 1-U and Hubtas (2024-01-30); Briggo, SEC Forms D (2012–2019); Cafe X at SFO, The Spoon (2021-01-21) and cafexapp.com; accessed 2026-10-02.
Chapter 06
The platform build
Over the program’s 76 days, one founder built Corner’s software platform and tested it to destruction.
In this chapter
6.1Money state machines, built for missing webhooks
6.2The live transaction
6.3The edge experiment
At a glance
173
commits in 76 days, one founder
19 · 71
adversarial scenarios and checks; all held
$1.99
a live purchase on production rails, July 9
The bench unit, a single-board computer behind a small touchscreen. Photographed on the bench, July 2026.
Corner’s platform is a production-grade payments and inventory system. At the close-out commit it counted 173 commits, made between April 30 and July 14, 2026, along with 57 API route handlers, 20 database tables with row-level security on every one, 48 schema migrations and 44 automated regression suites.1
Seventy-six days of commits, one founder.
Figure 6.1Commits per day, April 30 to July 14, 2026: 173 in all. The bursts are build days.git history of the platform repository, counted per author date through the close-out commit.
The platform ships four ways to move money: a code-confirmed shopper checkout, a walk-in register for cash and external cards, tap-to-pay through a card processor’s terminal rails, and an unattended charge-on-exit loop for a store with nobody in it (Fig. 6.2).
What the 76 days built: five counts, and four ways to move money through one operation.
Figure 6.2The platform at the close-out commit. All four payment paths reach the same money operation, the subject of the next section.Repository tree and git history at the close-out commit, July 14, 2026 (note 1).
Money state machines, built for missing webhooks
The platform assumes throughout that the network is unreliable and that event delivery is optional. Every money mutation carries a mandatory idempotency key, a tag that lets a repeated request take effect only once, and runs through a reserve-then-complete envelope. Every webhook, the event notice a card processor sends, is deduplicated by event identity.
A reconciliation job converges any stuck payment or refund against the processor’s truth, through the same idempotent operations. Webhook delivery is therefore a latency optimization rather than a correctness dependency (Fig. 6.3). Refunds on the terminal path run in two phases and are confirmed by event. An atomic claim means two concurrent refund attempts cannot both fire.
One money operation, three ways in. A lost webhook can delay a payment’s final state but cannot corrupt it.
Figure 6.3A request, a webhook and the reconciliation job all arrive at the same idempotent operation. The store’s device never holds payment authority, and the cloud is the ledger of record.The platform as built, close-out commit, July 14, 2026.
That design was then attacked on purpose, with nineteen adversarial scenarios and 71 checks. They covered oversell races, idempotency storms, duplicate-webhook floods, cron-versus-webhook finalize races, decline storms with inventory restore, out-of-order refund events and reconciliation cap saturation. All nineteen held.
The destruction pass surfaced exactly one real defect, an idempotency key that leaked across two transactions. It was fixed, and the pass re-ran clean.2
Table 6.1Nineteen adversarial scenarios against the money loop. All held, and scenario S found the pass’s one real defect.
#
Scenario
Result
Local concurrency5 scenarios, 23 checks
A
Oversell race
held
B
Idempotency storm
held
C
Rate-limit burst
held
D
Concurrent refund
held
E
Checkout key storm
held
Live test-mode payment rails14 scenarios, 48 checks
F
Refund-initiate storm
held
G
Cron ∥ webhook race
held
H
Duplicate webhooks
held
I
Persist-race adopt
held
J
Decline storm
held
K
Webhook-drop converge
held
L
Cron-only backstop
held
M
Rate-limit isolation
held
N
Reconcile cap
held
O
Cross-path collision
held
P
Pending refund arc
held
Q
Out-of-order events
held
R
Two-cron race
held
S
Key reuse across transactions
held · found the one real defect (fixed, re-run clean)
The platform’s chaos suites: 23 local concurrency checks; 48 checks against live test-mode payment rails, June 2026.
The live transaction
The loop was proven with money. The physical bench unit is a single-board computer behind a small touchscreen, driving a real lock. On July 9, 2026, it rang up a live $1.99 purchase by phone wallet against production infrastructure. The charge landed on the studio’s own merchant account and was receipted with no personal data. It was then refunded by hand from the processor’s dashboard.
The bench unit, from three angles.
Figure 6.4Checkout, with a contactless “tap to pay” beside the phone-handoff QR, then the menu grid, then the checkout again from further back. The physical bench unit, on the workbench, is a single-board computer behind an 800 by 480 touchscreen, driving a real lock. The screens run a demo fresh-drink menu.Photographed on the bench, July 2026; demo configuration. The kiosk software’s wordmark is retouched out of the screens. The on-screen QR was staged for the shoot and now leads to this study.
The shopper’s side: the phone checkout, and the July 9 charge.
Figure 6.5The phone checkout the kiosk’s QR opens, backed by the payment processor’s own elements; on a real phone the wallet sheet sits above the card form. Then the actual July 9 live purchase, $1.99, as it appeared in the cardholder’s banking app.Phone checkout: capture of the live surface, test rails, July 15, 2026. Bank app capture: July 9, 2026, cropped to the one charge.
The refund exposed a gap, which the program recorded but did not close. Nothing ingests a refund started from the processor’s dashboard, so the platform’s own ledger still reads the sale as succeeded (Fig. 6.6).3
The July 9 sale: the processor shows the refund, and the platform’s ledger still reads succeeded.
Figure 6.6The live purchase, step by step, in the two places that record it. The refund was made outside the platform, from the processor’s dashboard, and nothing carries it back in.Live-mode verification on the bench unit against production, July 9, 2026 (note 3).
The edge experiment
Figure 6.7The $250-class edge computer: an NVIDIA Jetson-class board the voice spike ran on, with the language and speech models co-resident under the heatsink. Beside it is the single-board computer driving the kiosk screen. The cloud keeps payment authority.Photographed on the bench, July 2026.
A parallel spike, a short exploratory build, asked whether store intelligence could live in the store. A $250-class edge computer with 8 GB of memory holds a 3-billion-parameter language model and a speech-to-text model in memory at once. In the common case a voice command completes end to end in about 2.6 seconds, under the program’s 4-second bar.4
How a voice command runs on the edge computer. Unambiguous commands skip the model entirely.
Figure 6.8One command from speech to applied action, all on the edge computer. Payment authority never leaves the cloud.Edge hardware spike, June 3–9, 2026, measured on-device (note 4).
Two software levers did the work. A deterministic fast path skips the model entirely for unambiguous commands. Grammar-constrained decoding against an indexed stock list cut the model’s proposal latency from 8.8 to 2.9 seconds.4 The edge device never holds payment authority, and the cloud stays the ledger of record.
On a $250-class edge computer, software alone cut the model from 8.8 seconds to 2.9 and brought a common voice command to 2.6 end to end, under the 4-second bar.
Figure 6.9The latency levers, then end-to-end time by language. Mandarin lands around 3.7 seconds; Tamil, at roughly 5.9, is the documented miss.Edge hardware spike, June 2026, measured on-device; per-language results as recorded, including the documented Tamil miss.
The platform, the money state machines, the chaos suites, and the edge pattern are assets of the studio, independent of any particular corner.
Notes
Counts audited at the program’s close-out commit, 2026-07-14: commits, route handlers, tables with row-level security, migrations, and regression suites counted directly from the repository tree and git history.
Chaos suites: 23 checks across five local concurrency scenarios; 48 checks across fourteen adversarial scenarios against live test-mode payment rails, all fourteen held, one defect found (cross-transaction idempotency-key leak), fixed and re-verified, June 2026.
Live-mode verification, 2026-07-09: $1.99 wallet purchase on the bench unit against production, charged and subsequently refunded on the studio’s own merchant account.
Edge spike, June 3–9, 2026, on an 8 GB Jetson-class edge computer: 3B-parameter instruct model plus whisper-class speech-to-text co-resident; English end-to-end voice ~2.6 s, Mandarin ~3.7 s, Tamil ~5.9 s (the documented miss); grammar-constrained decode 8.8 s → 2.9 s.
Chapter 07
Unit economics
The zero-labour drink store clears its running costs at 3.8 cups a day. The problem is paying back the capital, at demand anyone has measured.
In this chapter
7.1The per-cup model
7.2Volume scenarios
7.3Capital, fully totalled
7.4Then apply chapter 03
At a glance
$3.87
contribution per cup, at 30 percent cost of goods
3.8
cups a day to clear running costs
27.0 mo
payback of the smallest real stack at 12 cups a day, against a 24-month line
Juices and smoothies on ice at a stall in the Mercat de la Boqueria, Barcelona, 2009. The cup is this chapter’s unit: one $6.25 ticket, priced at 30 percent cost of goods. Photo: Bert Kaufmann, CC BY 2.0, via Wikimedia Commons.
The per-cup model
The modelled ticket is $6.25 pre-tax, mid-ladder among local drink prices. Its $5.49 floor sits below every comparable found (Fig. 7.1).1
The modelled $6.25 ticket sits mid-ladder among local prices, and its $5.49 floor sits below every comparable found.
Figure 7.1The model’s two prices against the Ontario bubble-tea range and its premium lines.Local price ladder, chain menus in the trade area, surveyed July 13–14, 2026 (note 1).
At 30 percent cost of goods, one cup contributes $3.87 after processing fees. Processing takes 8.1 percent of revenue on a single-cup ticket and falls as baskets grow.2
Table 7.1Where a $6.25 cup goes. CAD, single-cup ticket, 30% COGS.
Line
$
Cup price, pre-tax
6.250
Charged incl. 13% HST (remitted)
7.063
Processing (2.9% + $0.30 on gross)
0.505
Cost of goods (30%)
1.875
Contribution per cup
3.870
The same cup, drawn to scale.
Figure 7.2One single-cup ticket at 30 percent cost of goods. The shopper pays $7.06 with HST, which the store remits. Processing takes $0.51 on the gross and cost of goods $1.88, and the $3.87 left pays the fixed costs and, after them, the capital.Table 7.1; note 2.
Fixed costs run $450 a month, for power (refrigeration and climate included), connectivity, insurance and consumable service. The model charges no rent, and an operator that pays rent should add it.
No labour is charged either, because the format bans it. The check cuts the other way. One loaded minimum-wage part-timer, at roughly $2,800 a month, exceeds the entire floor-case revenue line.2 Break-even is $450 divided by $3.87, which comes to 116 cups a month, or 3.8 cups a day.
Volume scenarios
Table 7.2 runs the per-cup model at three daily volumes and three levels of cost of goods, over a 30.4-day month.
Table 7.2Monthly net by volume and COGS. CAD, 30.4 days/month.
Cups/day
28% COGS
30% COGS
32% COGS
12 (floor)
1,007
962
916
30 (base)
3,193
3,079
2,965
45 (stretch)
5,015
4,844
4,673
Monthly net at 30 percent cost of goods, from break-even at 3.8 cups a day to $4,844 at 45.
Figure 7.3Table 7.2 drawn across demand at 30 percent cost of goods. The whiskers mark the 28 and 32 percent cases, which sit so close to the line that the dots mostly hide them. The dashed line is the monthly cost of one loaded part-timer.Model: contribution $3.87/cup, fixed $450/mo, 30.4 days/mo (note 2).
Capital, fully totalled
The machine-only frame flatters the format. A $15K machine at the 12-cup floor pays back in 15.6 months, inside the study’s 18-month comfort line.
The machine needs a structure around it. The program’s bill of materials prices that buildout to the fastener: $17,638 for the walk-in-vestibule configuration and up to $23,720 for the serving-window configuration, permits and contingency included.3 Total capital for the smallest real configuration comes to about $26,000. With the component build at the top of its range, it reaches $37,600 (Fig. 7.4).
The real stacks add a $17,638 structure to the machine, and at 12 cups a day they take 27.0 and 39.1 months to pay back.
Figure 7.4The four capital stacks of Table 7.3, by what they hold, with the months each takes to pay back at the 12-cup floor.Buildout bill of materials, July 2026 (note 3); component build $8,400–20,000 (chapter 05); payback, Table 7.3.
Months to pay back four capital stacks, across demand. The curves flatten only well to the right of anything ever measured.
At the 12-cup floor, only the machine-only frame clears 18 months. The vestibule stacks run 2.2 to 3.3 years and the platform tier 5.2. In the shaded band, at or below measured demand, no total-capital stack pays back inside six years.
Figure 7.5The payback surface, with the category’s measured best and the study’s thresholds marked. The shaded band is demand at or below the measured line. Curves are undefined where net margin approaches zero.Model: this chapter’s tables (contribution $3.87/cup, fixed $450/mo). Reference lines: chapter 03 (6.8 measured); the program’s kill criteria (12 floor, 24-month line).
Table 7.3Payback in months, 30% COGS, total capital.
Stack
@ 12/day
@ 30/day
@ 45/day
$15K machine only (the flattering frame)
15.6
4.9
3.1
$26K vestibule + component build, low
27.0
8.4
5.4
$37.6K vestibule + component build, high
39.1
12.2
7.8
$60K platform tier
62.4
19.5
12.4
The program’s own kill line was 24 months. Winter makes the picture worse. At 60 percent cold-month volume with no hot menu line, the floor case’s annualized net falls by about a quarter.2
Then apply chapter 03
The 12-cup columns above are the floor case, 12 cups a day, every day. The category’s best measured operator serves 6.8. It is mature, it sits in a captive venue, and its machine class has a decade of iteration behind it.
The floor case is therefore not conservative. It requires beating the category’s best documented steady state by 75 percent, from a cold start. The 30- and 45-cup columns, where the larger stacks become rational, have no measured comparable anywhere in the public record (Fig. 7.6).
Running costs clear at 3.8 cups a day, the floor case assumes 12, and the category’s best measured unit serves 6.8.
Figure 7.6The chapter’s demand numbers on one axis of cups a day: what the margins need, what has been measured, and what the model’s cases assume.Break-even: this chapter. Measured best: chapter 03. Floor, base and stretch cases: Table 7.2.
The model fails on demand, not margins. Left of the measured line, no total-capital stack pays back inside six years, and the counted evidence that would move the model right of it does not exist.
Notes
Local price ladder surveyed 2026-07-13/14 across chain menus in the trade area: the modelled $5.49 floor undercuts every found comparable line; $6.25 sits between mid-ladder chain prices. Ontario bubble-tea range roughly $5–8, premium lines to $8–12.
The program’s unit-economics model, recomputed independently at close-out from first inputs (ticket, HST, processing 2.9% + $0.30, COGS 28/30/32%, fixed $450/mo, 30.4 days/mo), July 2026. Winter sensitivity modelled at 60% volume (no hot line) and 80% (hot line) for November through March.
Buildout bill of materials, July 2026: walk-in vestibule configuration $17,638 all-in (60A panel variant); serving-window configuration to $23,720 (100A variant); both include a 12% contingency on physical goods and $2,040 of permits and fees.
Chapter 08
Site circulation
A walking customer cannot reach the rear structure without crossing the lot’s vehicle circulation. That caps what the structure can hold. It is the narrowest finding in this study, a bound on one 240 square foot structure, and it is not the reason the program closed.
In this chapter
8.1What the geometry caps
At a glance
240 sq ft
the rear structure, 12 by 20 feet
3
curb cuts: one on Barton, two on Dewitt
0
walking routes that stay off the vehicle plane
The lot from above, April 2025. The finding of this chapter is about the paved plane around the building. Site photography, the study’s own archive.
The lot’s geometry fixes its vehicle circulation. Dewitt carries the entries: two curb cuts, the first just south of the corner and the second further down. A third cut opens on Barton (Fig. 8.1).1
The vehicle plane wraps the building on three sides, and every walking route to the rear structure crosses it.
Figure 8.1The circulation finding. The shaded vehicle plane covers the north apron in front of the bays, the east corridor and the rear lot, entered from the two Dewitt cuts and the Barton cut. The strip west of the building is a gully with no vehicle path. Walk-up approaches are dashed and conflict points marked.Site-plan analysis from the 2007 Ontario Land Surveyor site plan of the parcel, the study’s aerial photography (April 2025) and ground photography (December 2024 and April 2025), and site observation; schematic, approximate scale.
Figure 8.2The vehicle plane at ground level, from the rear lot looking north, April 2025. This is the plane a walking customer must cross.Site photography, the study’s own archive.
The strip west of the building is a gully with no vehicle path. Everything on wheels therefore wraps the building’s north, east and south sides: the apron in front of the twin Barton-facing bays, the east corridor, and the rear lot.
Figure 8.3The lot from the north-east, April 2025. The north apron and the east corridor wrap the building, and the rear structure is the grey shed at left. Everything between it and either frontage is vehicle plane.Site photography, the study’s own archive.
Figure 8.4Two faces of the building, December 2024. Left, the twin Barton-facing bays, seen from the north apron. Right, the east face with its third bay at the south end, and vehicles parked against it on the plane a walking customer must cross.Site photography, the study’s own archive.
The rear structure stands at the south-west of the lot, on the far side of that vehicle plane from both sidewalks. Trace any pedestrian path from the Barton sidewalk or the Dewitt sidewalk to the rear structure, and it crosses the vehicle circulation area. The gully rules out a third approach. Every route the geometry offers puts a walking customer and a vehicle in the same plane (Fig. 8.5).
Two approaches cross the vehicle plane, and the gully closes the third.
Figure 8.5The walk-up finding of Fig. 8.1, route by route: where a walk to the rear structure can start, and what it crosses on the way.Site-plan analysis, as Fig. 8.1.
Figure 8.6The subject itself, December 2024. The 12 by 20 foot rear structure stands at the lot’s south-west, beyond the parking row and the rear apron. At right are the south end of the building’s east face and its third bay. Everything between the structure and either sidewalk is the vehicle plane of Fig. 8.1.Site photography, the study’s own archive.
What the geometry caps
The consequence is a cap on the structure’s uses. The structure itself is not ruled out. In the lot’s current configuration, rear-structure uses that require the public to walk in are out. Uses that do not, such as storage, service functions or equipment, are unaffected.
A customer-facing use sited in that structure would need a walking route clear of vehicles, or a different structure. Both are engineerable.
The cap removes walk-in uses and leaves the rest of the structure’s uses alone.
Figure 8.7What the circulation finding rules out, what it leaves, and the two ways a customer-facing use could get past it.Site-plan analysis, as Fig. 8.1.
The cap decided nothing. Chapters 03 and 07 and Appendix D hold what closed the program: measured category demand below the model’s floor, total capital that does not pay back inside the kill line, and better-funded operators who had already run the experiment and lost.
Notes
Analysis of the study site’s own plan geometry: lot and frontage dimensions from a 2007 Ontario Land Surveyor site plan of the parcel and as published on the property’s website; bay positions and rear-structure position from the study’s aerial photography, April 2025, orientation fixed by the source files’ camera-heading EXIF, and its ground photography of December 2024 and April 2025. The two Dewitt entrances and the gully are from site observation. The walk-up finding is the studio’s own site-plan analysis.
Chapter 09
Verdict, and what transfers
Unattended fresh-drink retail fails on demand per machine in the one case anyone has measured, and the cost of proving otherwise exceeds the return available. Several far better-funded operators established that before this program began.
In this chapter
9.1How it closed
9.2What transfers
9.3What was never proved
9.4What would reopen the question
At a glance
Kill
the verdict, in full
≈ $0
venue capital consumed by the 76-day program
3
conditions recorded at close that would reopen the question
The corner and its neighbourhood, the escarpment on the horizon, April 2025. Site photography, the study’s own archive.
Verdict, in full: kill. The cost to prove this model, across prototyping, development, marketing and selling it, exceeds what it returns at any demand anyone has measured.1
The smallest real capital stack is about $26,000 to $37,600. It pays back in 2.2 to 3.3 years against a 24-month kill line, and it needs 12 cups a day to do it. The category’s one measured throughput figure is 6.8, from a mature unit in a captive indoor venue. That operator’s disclosed economics are a venue service fee alongside the retail line, rather than cup margin alone.
The category is contested, and its distribution is owned. The leader’s placement runs on the vending industry’s existing procurement rails, the region already carries operators, and the better-funded entrants ran this experiment first and lost.
The site is not the reason. Its exposure is real and independently priced, and its zoning is permissive. The rear structure’s walk-up cap is a bound on one 240 square foot structure, and an engineerable one. The viable expressions of the work are venue-placed units, on the service-fee economics the category runs on, and the platform itself.
Measured demand sits below what the model needs, and even at the demand it needs, the capital misses the kill line.
Figure 9.1Top, the category’s one measured throughput figure against the 12 cups a day the smallest real capital stack needs. Bottom, that stack’s payback at 12 cups a day against the program’s 24-month kill line.Chapter 03 (6.8 a day, measured); chapter 07, Table 7.3 (payback at 12 a day for the $26K low and $37.6K high stacks).
How it closed
Each concept in chapter 02 died against a named criterion or a falsified premise, or closed with the program. The evidence that closed the remainder was public and free the whole time. Operators with more money had already written the category’s own record, and Appendix D holds the roster.
An entrant with US$26 million of stock and note sales on file, hundreds of times what this program would ever commit, was acquired by a coffee incumbent. A US$3.5 million crowdfunded entrant closed without ever advancing past prototypes and trade-show demos. A robot bubble-tea shop opened in March 2024 and closed in 2025.
The demand ceiling of chapter 03 and the capital arithmetic of chapter 07 then said the same thing in this program’s own numbers. The program’s founder closed all venue exploration the same day the last reuse of the site in chapter 02 was falsified. The circulation finding of chapter 08 landed the same week. It caps one structure, and it moved none of the arithmetic.
The 76-day program consumed approximately zero venue capital. Nothing physical was bought for the structure, no permit was filed, and no machine was ordered. The spend was bench hardware and domain renewals.
Three lines of evidence closed the program on July 14, 2026: demand, total capital and the category record. The circulation finding of the same week was not one of them.
Figure 9.2The three findings under “the evidence” closed every venue line. The dashed boxes are the two other findings that the close-out entry of Table A.1 records, neither of them binding. Below the rule, the venue capital the program spent.Close-out record of the program, July 14, 2026; Appendix A, Table A.1.
What transfers
The platform. A production payments and inventory system with four money paths, idempotent state machines, reconciliation convergence, receipts, and a 71-check adversarial test record. It charged a live purchase on production rails, and it is venue-independent by construction.
The edge pattern. A validated recipe for on-device store intelligence: a small language model and speech-to-text co-resident on a $250-class computer, sub-4-second voice in the common case, with the cloud as ledger of record. The same architecture serves any deployment where compute must live on site.
The method. Kill criteria before capital, primary-source verification of load-bearing claims, and the one amendment this program earned: read the category’s own record, and price what it costs to prove the model, before building the model. Appendix C states it in full. It is the studio’s standing operating procedure.
What was never proved
The market. No store ran on the platform and no shopper used it in the wild. The platform was never opened to the public, and the program closed before the cup count ran. The close-out state is launch-ready engineering with no market evidence. The program ruled out each venue as a place to test the platform, but it never tested the platform as a business.
Three things transfer from the program. The market was never proved.
Figure 9.3What the close-out record carries forward, and what it records as never proved.Close-out record of the program, July 14, 2026.
What would reopen the question
Three conditions were recorded at close (Fig. 9.4). The first is an inbound placement request from a venue. No outbound motion reopens the file.
The second is counted demand, measured somewhere real, that clears the floor case with the total-capital arithmetic closing inside the 24-month line. That is the first gate now. No parcel reopens the file on its shape, because the shape of a parcel was never what failed.
The third is a need for the unattended-retail stack as a hardware testbed, in which case the store becomes the harness rather than the business. Absent one of these, the verdict stands.
Three conditions recorded at close would reopen the question. A better parcel is not one of them.
Figure 9.4The revival conditions filed at close. Counted demand is the first gate now.Close-out record of the program, July 14, 2026.
Figure 9.5The corner, looking west along Barton Street, April 2025.Site photography, the study’s own archive.
Notes
Close-out record of the program, July 14, 2026: verdict, capital accounting, transferring assets, and revival conditions as filed in the program ledger. The same record carries what was never proved: zero stores onboarded across the program’s life, the platform never opened to the public, and the cup-count protocol written but never run. Category figures in this chapter are sourced in chapters 03 and 07 and Appendix D.
Appendix A
Timeline
Every dated entry below is anchored to the program’s ledger, its git history, or a filed record.
Seventy-six days, April 30 to July 14, 2026.
Figure A.1The program’s dated milestones from Table A.1, on one line.Program ledger and git history.
Table A.1The program, in date order.
Date
Event
2026-04-30
Founding commit of the software platform.
2026-05-13
The walk-in register ships: cash and card ring-up, end-of-day closeout.
2026-05-25
Strategy pivots register-first; two days later the posture hardens to no partial rollout, test to destruction before any real user.
2026-05-28
The payments spine lands in one long day: processor onboarding, terminal simulator, admin authentication with an append-only audit log, and the first chaos harness.
2026-06-03
to 06-09
Edge hardware spike: a 3B language model and speech-to-text co-resident on an 8 GB edge computer; voice under the 4-second bar for the common case. Validated, then deliberately parked.
2026-06-10
The bulletproofing capstone: two-phase terminal refunds, the reconciliation backstop, and the 48-check destruction pass on live test-mode rails. All scenarios held.
2026-06-12
All money paths verified live on deployed infrastructure, test mode, end to end.
2026-06-30
The unattended loop ships: session, charge-on-exit, inventory decrement, receipt.
2026-07-06
First full bench proof on physical hardware: a real session rung up and closed by a single-board computer driving a real lock, still test mode.
2026-07-09
First live dollar: a $1.99 phone-wallet purchase charged on production rails and refunded by hand from the processor’s dashboard.
2026-07-13
The studio’s doctrine is written down. The drink program is specified to the fastener: hypothesis, seven kill criteria, market calendar, machine assessment, buildout bill of materials. The staffed variant is rejected the same day.
2026-07-14
The evidence closes the question: the measured demand ceiling (chapter 03), the landed-cost falsification (chapter 05), the total-capital arithmetic against the 24-month kill line (chapter 07) and the record of better-funded operators who lost (Appendix D). All venue lines are closed and the record is filed. The circulation finding (chapter 08) lands the same week, a bound on one structure and not a cause.
Appendix B
Sources
Every external source cited in the study, grouped by chapter, with access dates. A figure reported “as listed” is the listing party’s own number. Where such a source could not be fetched directly at access time, the entry says so.
Half the entries document the category: chapter 03 and Appendix D.
Figure B.1One square for each numbered entry below, in list order. The filled rows, 29 of the 58 entries, are the sources on the category itself. Sixteen entries, eleven of them in Appendix D, are dated 2026-10-02, when the category record was checked again after the program closed.Counted from the entries of this appendix.
The site (chapter 01)
The property’s website: lot and building dimensions, zoning and permitted uses, transit routes, highway distances, and the 21,000+ daily vehicle figure, reported as published. Accessed 2026-07-14 and 2026-07-15.
Vendo Media, 521 Barton Street digital billboard: 21,034 daily impressions, as listed by the vendor. Per-board page not directly fetchable 2026-07-15; vendomedia.co consulted the same day.
City of Hamilton open data was searched for a junction traffic count (open.hamilton.ca, ArcGIS Hub API), 2026-07-15; no citable dataset was found and no estimate is substituted.
Site plan of the parcel by an Ontario Land Surveyor, 2007: lot and frontage dimensions and the building’s position, used for the plans of chapters 01 and 08.
The concept space (chapter 02)
City of Hamilton Building Permit By-law 15-058 (change-of-use permits); Ontario Building Code, Group A-2 assembly occupancy and its washroom and barrier-free requirements. As assessed in the program’s close-out record, July 2026.
SEC EDGAR, Zenblen, Inc., CIK 0001993331: Form D filed 2023-09-18; Form D filed 2026-01-21; two Sydecar-administered feeder SPV filings. All accessed 2026-07-14.
mHUB HardTech Summit pitch video (uploaded 2026-03-25): youtube.com/watch?v=wW6zQRT78pA, 41:28 to 44:18, auto-captions. Accessed 2026-07-14.
UChicago Polsky Center, 2025-06-11: polsky.uchicago.edu. Accessed 2026-07-14.
LinkedIn public profile: ca.linkedin.com/in/rehmanmerali. Accessed 2026-07-14.
6d bytes inc. (Blendid), FY2025 annual report, SEC EDGAR CIK 1830909, filed 2026-03-02. Accessed 2026-07-14.
Built In Chicago company and job listings (labelled broker data where used). Accessed 2026-07-14.
Regulatory posture (chapter 04)
O. Reg. 493/17 (Food Premises): ontario.ca/laws/regulation/170493 and canlii.org. Accessed 2026-07-14.
Ontario Ministry of Health, Food Premises Reference Document, 2019, p. 18. Accessed 2026-07-14.
City of Toronto DineSafe open dataset (104,790 records at access), parsed directly. Accessed 2026-07-14. DineSafe historical data, 2001 to 2023, and the current dataset as updated 2026-10-01, searched again for RC Coffee. Accessed 2026-10-02.
rccoffee.com (locations, the Robo Cafe brand since 2021). Accessed 2026-07-14. Newsfile release of 2020-10-07 (the first kiosk). Accessed 2026-10-02.
City of Chicago Socrata API: Food Inspections 4ijn-s7e5; Business Licenses r5kz-chrr. 26 queries, raw responses archived. Accessed 2026-07-14.
Chicago Municipal Code §4-8-032 (automatic food vending machine operations), indexed text. Accessed 2026-07-14.
Illinois Department of Public Health, retail food program (FDA Food Code 2022 adoption): dph.illinois.gov. Accessed 2026-07-14.
410 ILCS 625/3: ilga.gov. Accessed 2026-07-14.
Conference for Food Protection, Unattended Food Establishments issue paper (2014 packet): foodprotect.org. Accessed 2026-07-14.
Automation assessment (chapter 05)
Entrepreneur (Bobacino funding coverage), accessed 2026-07-14. SEC EDGAR, Future Pearl Labs, Inc. (Bobacino), CIK 1830113: Form 1-SA filed 2022-09-28 and Form 1-U filed 2023-12-07; Hubtas newsletter, 2024-01-30. Accessed 2026-10-02.
FX: 1 USD = 1.415 CAD, market range for the week of 2026-07-07, exchangerates.org.uk. Accessed 2026-07-14.
Platform, economics, circulation (chapters 06 to 08)
The program’s own records: repository git history and close-out audit (2026-07-14), chaos-suite results, the unit-economics model, the buildout bill of materials, and the close-out ledger. Internal records; figures reproduced in the chapters are the audited counts.
Local drink-price ladder: chain menu prices surveyed in the trade area, 2026-07-13/14.
Site photography: the study’s own archive, ground photography December 2024 to April 2025, aerials April 2025. All images resized and stripped of metadata for publication.
Bench photography and screen captures (chapter 06): the studio’s own hardware and surfaces, documented July 2026, after the production deployment was parked. On-screen QR codes in these frames resolve, deliberately, to this study rather than to a live checkout session, and capture-tool overlays were removed in processing. The kiosk software’s wordmark was retouched out of the screens. No production endpoint, payment session, or chargeable code appears in the published record. The July 9 live dollar is documented at Fig. 6.5.
Third-party pictures are credited under each figure and listed in full in the picture credits at the end of the study. Photographs under a free licence, from Wikimedia Commons and Flickr, are used on that licence’s terms. Press photographs, marked as such in the picture credits, are reproduced at review scale for the purpose of review, each with its photographer or publisher named under it. The drawings of Fig. 3.7 are United States Patent and Trademark Office records. The maps are drawn from OpenStreetMap data (ODbL) and Natural Earth (public domain). A rights holder’s removal request to hello@craftpine.com is handled by deletion.
The category record (Appendix D)
Entries draw on the sources listed above for chapters 03 to 05 and on these, checked in October 2026; all accessed 2026-10-02.
Bobacino: SEC EDGAR, Future Pearl Labs, Inc., CIK 1830113, Form 1-K for 2021 (filed 2022-05-02: incorporation, the Wavemaker Labs studio), Form 1-SA filed 2022-09-28 (offering proceeds) and Form 1-U filed 2023-12-07; The Spoon, 2020-11-17 (the prototype); Hubtas newsletter, 2024-01-30.
Tea Industries: Richmond News, 2024-03-26 (soft opening); Noms Magazine, 2024-03-25 (staff and timing) and 2025-12-13 (closure); Foodgressing, 2024-03-26 (the robot’s sequence); Daily Hive, 2025-12-22.
Briggo: SEC EDGAR, Briggo, Inc., CIK 1539135, Forms D and D/A filed 2012-01-11 to 2019-08-23 (seven offerings of stock and notes, US$26,030,283 reported sold); Austin American-Statesman, 2011-12-02 (the founder) and 2020-10-30; CultureMap Austin, 2020-10-30 (five active kiosks, the rebrand); FER, 2018-05-01 (founding year); Texas business records via opencorporates.com (Briggo, Inc. renamed BC Wind-Down, Inc., 2020-05-05). Accounts differ. The Statesman gives a 2009 founding and about US$10 million raised, and the filings give 2011 as the year Briggo, Inc. was incorporated. Daily Coffee News gave US$19 million, the sum of the filings made through 2018; a 2019 amendment and a 2019 offering bring the total to US$26.0 million. The 2017 stock was paid for partly by converting bridge notes, so the total can count some money twice.
f’real Foods: Rich Products release, PR Newswire, 2012-12-12 (founding, locations, the acquisition); f’real release, 2023-09-25 (the first blender, Tulsa, 2003); the 2019 verdict as in chapter 03, note 8.
Blendid: SEC EDGAR, 6d bytes inc., CIK 1830909, Form 1-SA filed 2026-07-14 (paid-in capital at 2026-04-30, the going-concern wording), Form 1-K for FY2025 filed 2026-03-02 (balance sheet, going concern, staff, write-off, products) and for FY2022 filed 2023-02-28 (first kiosks, pilots, kiosks deployed); Entrepreneur, 2022-09-02 (kiosks in contract).
Artly: SEC EDGAR, Blue Hill Tech, Inc., CIK 1936104, Form D filed 2022-08-08, Form C-U filed 2024-04-23 and 2026-05-04, Form C filed 2026-06-22 with FY2025 financial statements and the auditor’s report, and Form C/A filed 2026-09-18; nomsmagazine.com (MUJI Robson Street, Vancouver, 2024-01-15); blogto.com (MUJI, 20 Dundas Street West, Toronto, March 2024); tocityscapes.com (the Toronto counter, 2025-09-03); Google Maps listing, Artly Coffee, 20 Dundas St W, Toronto (3.7 of 5 from 12 ratings).
teaBOT: ycombinator.com/companies/teabot; TechCrunch, 2015-07-23; U of T Engineering News, 2016-05-31 (locations, revenue-share kiosks); teabot.com/how-it-works (the office subscription).
Cafe X: SEC EDGAR, Cafe X Technologies, Inc., CIK 1748775, Form D filed 2018-08-06 (US$9,420,709 sold; incorporated 2014) and Form C-U filed 2023-05-01 (US$319,548); TechCrunch, 2017-01-30; The Spoon, 2020-08-11 and 2021-01-21; Crunchbase News, 2020-01-07; cafexapp.com/locations.
RC Coffee: Newsfile releases of 2020-10-07 (the first kiosk) and 2021-07-20 (the Robo Cafe brand; KioSoft Technologies as parent); rccoffee.com/locations (ten Robo Cafes); City of Toronto DineSafe historical data, 2001 to 2023, and the current dataset as updated 2026-10-01, searched by RC Coffee’s names and addresses (no record).
Caffeo: 6ix Retail, 2025-10-21 (founding, opening, the Series A and five-site plan); Retail Insider, 2026-06-18 (the relaunched flagship); City of Toronto DineSafe dataset as updated 2026-10-01 (passes of 2025-09-12 and 2026-09-03).
Smoodi: SEC EDGAR, smoodi, Inc., CIK 1956086, Form D filed 2022-12-02 (US$4,974,505 sold); TechCrunch, 2022-12-02 (the Series A, Dot Foods); getsmoodi.com, June 2026 (lease and purchase prices, locations).
Appendix C
Method note
The studio keeps three standing commitments.
Kill criteria before capital. Every format under study carries numbered, numeric, dated criteria with the exit action written in advance. This study’s K2 was falsified by research before a purchase order existed.
Falsifiable gates, amended by their misses. The amendment this study earned is to read the category’s own record, and to price what it costs to prove a model, before building the model. The evidence that closed this program was public and free the whole time. It was written by operators carrying more capital than this studio will commit.
A better site, a better machine or a better menu does not rescue a model that cannot pay back at any demand anyone has measured. So the category record and the total-capital arithmetic now run first, before format work, before machines and before any site. Site constraints, circulation among them, are checked inside the site review where they belong.
The category record and the cost to prove the model now come before any format, machine or site.
Figure C.1The amended order of the studio’s gates, with the category record and the total-capital arithmetic run first.The studio’s operating procedure, with the amendment this study earned.
The operating model. The studio runs as two AI surfaces around one founder. One authors strategy and task specifications, and the other executes against the real system and writes back evidence. Every task ends in a written report, and the reports are the record. Load-bearing claims are verified against primary sources before they are trusted, and the confirmed, refuted, and excluded counts are kept.
Two AI surfaces around one founder, and a written report at the end of every task.
Figure C.2How a task moves through the studio, and the check every load-bearing claim passes before it is trusted.The studio’s operating procedure.
Appendix D
The category record
Every company this study leaned on, in one place: what each built, what it raised, and where it stood in October 2026.
Facts below follow the study’s sourcing rule. Capital figures are amounts documented in SEC filings, and none is an estimate. Each fate is dated in its entry. Each company is shown with a photograph of what it built, credited under it and again in the picture credits at the back.
The record was checked again in October 2026, after the program closed, and Artly was added then. Cafe X, which the July record had out of retail, is shown operating, since it reopened at San Francisco’s airport in 2021. Briggo’s capital, a press figure of US$19 million in July, is now the US$26.0 million total of its filings. None of these changes the verdict.
Where the category was tried: twelve companies in eight metropolitan areas.
Figure D.1Every operator in the record, by home city and status in October 2026. The San Francisco Bay Area holds three, and Toronto, the nearest of them to the study site, holds three more.Home cities and fates as in the entries below. Base map: Natural Earth, public domain. Drawn for this study.
Thirty years of the category, company by company.
Figure D.2Each line runs from founding, or first unit, to the company’s fate as recorded in its entry, or to October 2026 for a company still operating. A dashed line is life under an acquirer.The entries below; dates as documented there.
The durable outcomes are machine sales, venue placements on service fees, and machines simpler than a robot. No operator has published a street-facing unattended fresh-drink store that carries itself on cup margins (chapters 03 and 07).
The largest documented sum, about US$36.0 million, went to Blendid, now under going-concern doubt.
Figure D.3Capital each company documented in SEC filings, as given in its entry, keyed by its status in October 2026 as on the map. The measures differ by filing, as labelled. Blendid’s is capital paid in, and Briggo’s is stock and notes reported sold, a total that can count some money twice (Appendix B).The Capital line of each entry below.
The largest footprints belong to f’real and Smoodi, whose machines are simpler than a robot.
Figure D.4Units in place, on a logarithmic scale, as each entry states them. Units and dates differ by company, as labelled. The entries give no count for Blendid, teaBOT or Tea Industries, and Bobacino never advanced past prototypes.Counts and dates as stated in the entries below.
Closed
BobacinoclosedFigure D.5Bobacino’s arm at work for a trade-show audience, World Tea Conference + Expo, Las Vegas Convention Center, March 22, 2022.Photo: K.M. Cannon, Las Vegas Review-Journal.
Home
Los Angeles
Founded
2020
Capital
about US$3.5M crowdfunded (SEC Regulation A+, 2021 and 2022)
Bobacino was incorporated in March 2020 inside Wavemaker Labs, a robotics venture studio, and unveiled a robotic boba bar prototype that November. It raised about US$3.5 million from crowd investors and never advanced past prototypes and trade-show demos. Its chief executive resigned in November 2023, and by January 2024 it was reported shut down, along with other companies from the studio.
Tea Industriesclosed 2025Figure D.6Tea Industries at its soft opening on No. 3 Road, Richmond, BC, March 2024: the robot’s glass cabinet at left, the ordering counter at right.Photo: Valerie Leung, Richmond News.
Home
Richmond, BC
Opened
March 2024
Capital
not disclosed
Tea Industries was British Columbia’s first robot bubble-tea shop, and the Canadian precedent closest to this study’s format. A robotic arm behind glass brewed, shook, and sealed each order in about two minutes. Staff were still on hand to prepare ingredients, help customers, and clear faults.
It opened in March 2024. By December 2025 its windows were papered over and the unit was up for lease (Fig. 5.6). Local press reported the closure without a post-mortem, and no public account of why has been found.
Acquired
Briggoacquired 2020Figure D.7A Briggo Coffee Haus set into the wall at Austin-Bergstrom International Airport, October 2019: the machine behind the window at left, the pickup cubby in the middle.Photo: LunchboxLarry, CC BY 2.0, via Flickr.
Home
Austin
Founded
2008
Capital
US$26.0M in stock and notes reported sold (SEC Form D, 2012–2019)
Briggo raised hundreds of times the capital this program would ever commit. Founded by a semiconductor engineer, it placed its Coffee Haus kiosks with employers, grocers, and airports across a decade of raises. Costa Coffee, a Coca-Cola subsidiary, acquired it in 2020 on undisclosed terms and rebranded its five active kiosks as Costa Coffee BaristaBots.
f’real Foodsacquired 2012Figure D.8An f’real blending bar in a truck-stop store in Ontario, December 2014: the in-cup blender at left and the freezer of cups at right, where milkshakes are priced at $3.29.Photo: Cory Doctorow, CC BY-SA 2.0, via Flickr.
Home
Emeryville, CA
Founded
1995
Capital
not disclosed
The category’s one durable success is f’real, and it is not a robot. Its self-serve in-cup blender for real-milk shakes was first installed in a store in Tulsa in 2003. It was in more than 9,000 locations by December 2012, when Rich Products acquired the company.
Its live patents on in-cup blending run to 2037 (chapter 03). It enforces its patents: in 2019 a Delaware jury found that Hamilton Beach and Hershey Creamery had infringed three earlier ones, and the verdict was affirmed on appeal in 2021.
Under going-concern doubt
Blendid6d bytes inc.going-concern doubtFigure D.9Blendid’s first San Francisco kiosk, in the Market Cafe of the University of San Francisco, on its test run, March 22, 2019. The menu screen prices its Strawberry and Cream smoothie at $6.Photo: Paul Chinn, San Francisco Chronicle.
Home
Sunnyvale, CA
Founded
2015
Capital
about US$36.0M paid in by April 2026 (SEC Form 1-SA), raised in part by Regulation CF and A+ crowdfunding
Blendid built an eight-by-eight-foot robot-arm smoothie kiosk, launched in 2019 and piloted at Walmart and on campuses. A claimed pipeline of hundreds of kiosks did not materialize. At the end of 2024 the company stopped selling the original kiosk and turned to two new products, a smaller, cheaper kiosk and a back-of-house unit, to be sold rather than leased.
Its FY2025 annual report discloses thirteen employees, a US$308,000 obsolete-inventory write-off, and substantial doubt about continuing as a going concern if revenue targets are not met. Its July 2026 semiannual report adds a failure to raise capital as a second condition of that doubt.
ArtlyBlue Hill Tech, Inc.going-concern doubtFigure D.10Artly’s arm behind glass at the coffee bar in MUJI at the Atrium on Bay, Toronto, August 2025, between the espresso machine and the grinder.Photo: Marvin Job, TO Cityscapes.
Home
Seattle
Founded
2020
Capital
US$10.3M documented (SEC Form D, 2022; Regulation CF, 2023–24 and 2026)
Artly’s robot arm was trained by imitating human baristas. Most Artly Coffee locations are company-operated. Its robots also serve enterprise customers and host stores, MUJI among them in Vancouver and Toronto from 2024.
Its June 2026 offering statement reports FY2025 revenue of US$2.2 million, a net loss of US$2.8 million, more than forty robots deployed in the United States, and an auditor’s going-concern paragraph. The same statement calls the robot closer to a premium vending solution than a coffee shop.
The Toronto counter held a Google rating of 3.7 of 5, from 12 ratings, in October 2026. Its reviews describe slow drinks and staff stepping in to refill the robot.
Operating
ZenblenactiveFigure D.11A Zenblen kiosk on the University of Chicago campus, April 2025: touchscreen, card reader and pickup window in one wood-grain cabinet.Photo: Institute of Design, Illinois Institute of Technology.
Home
Chicago
Founded
2018
Capital
US$4.6M documented (SEC Form D, 2023 and 2026)
Zenblen is the category’s best-documented operator and the subject of this study’s chapter 03. It had eleven kiosks at audit, placed through foodservice procurement on venue-fee-plus-retail economics. Every one was inside an access-controlled or hours-limited building, and none was street-facing. One unit’s 2,500 smoothies in calendar 2023 is the only measured per-unit throughput figure the category has published.
teaBOTactiveFigure D.12A teaBOT kiosk at Northeastern University, Boston, November 2019: the clear cone under the tea canisters, the hot-water spout, a custom blend at $2.99 on the screen.Photo: Martin Lewison, CC BY 2.0, via Flickr.
Home
Toronto
Founded
2013
Capital
Y Combinator S15; venture amounts undisclosed
teaBOT’s robotic kiosk blends custom loose-leaf tea in under thirty seconds. By 2016 the company had placed kiosks in cafés, libraries, and campuses. It survived its decade by moving from revenue-share kiosks in cafés and libraries into offices, where it places and services the machine for a flat monthly fee. Its chief executive also lists himself as Zenblen’s CTO (chapter 03), one sign of how small the category’s talent pool is.
Cafe XactiveFigure D.13Cafe X’s first robot bar, inside the Metreon, San Francisco, March 2018: the arm between two espresso machines, the cup towers behind it.Photo: Ed Johnson, CC0, via Flickr.
Home
San Francisco
Founded
2014
Capital
US$9.7M documented (SEC Form D, 2018; Regulation CF, 2023)
Cafe X put an industrial robot arm behind glass in downtown San Francisco in January 2017 and closed all three city locations in January 2020. Crunchbase News reported that its founder wanted it to be less robot, more old-school coffee. The pandemic closed its airport bars within months, and it reopened at San Francisco’s airport in 2021. In 2026 it runs three bars there and sells its machines to other operators, with placements listed from Dubai to Berlin.
RC CoffeeactiveFigure D.14RC Coffee’s Robo Cafe kiosk in the food court of Toronto General Hospital, July 2022: order screens and the pickup hatch in the window.Photo: handout, via Global News.
Home
Toronto
First kiosk
2020
Capital
not disclosed; a unit of KioSoft Technologies, an unattended-payments company
RC Coffee runs Toronto’s unattended espresso network. Its first kiosk opened in fall 2020 with a local roaster, and its own Robo Cafe brand followed in June 2021. It listed twelve or more locations in July 2026 and ten in October 2026, hospital sites among them. Its kiosks are cashless and use fresh dairy. DineSafe holds no record of them in any year of operation, which is half of chapter 04’s enforcement finding.
CaffeoactiveFigure D.15Caffeo’s arm setting down an iced latte inside the Richmond Street West store, 2025.Photo: Caffeo, via 6ix Retail.
Home
Toronto
Founded
2023
Capital
not disclosed
Caffeo’s robot arm works espresso equipment behind glass on Richmond Street West, where the store opened in September 2025. It passed a DineSafe Food Premises inspection on September 12, 2025, in its first weeks of trading, and another on September 3, 2026. It is the other half of chapter 04’s finding.
A Series A it targeted for the end of 2025 has not been announced, and one of the five sites it planned for the end of 2026 is open. The flagship relaunched in June 2026 and keeps operating.
SmoodiactiveFigure D.16A Smoodi self-serve blending station, pre-portioned frozen fruit cups in the glass-front freezer beside the blender, February 2026.Photo: יאיראבני, CC BY-SA 4.0, via Wikimedia Commons.
Home
Boston
Founded
2018
Capital
US$4.97M Series A (SEC Form D, 2022)
Smoodi tests the cheaper thesis. It has no robot arm, only a self-cleaning blender over a freezer of pre-portioned frozen fruit cups, leased from US$299 a month or bought from US$14,999. It closed a Series A in December 2022, has national foodservice distribution, and claimed 300-plus locations by 2026. Its machine is the US$14,999 rung in Fig. 5.1.
Picture credits
Every photograph, map and reproduced drawing in this study, in the order it appears. Photographs under a free licence are credited by title, author, source and licence, and are used on that licence’s terms; changes are noted. Press photographs are reproduced at review scale for the purpose of review. The study’s own pictures carry no third-party rights.
CoverThe study site from the north-east, April 2025.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Chapter 01No. 520 from the north, April 2025.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 1.2The site on the QEW at the west end of Lake Ontario.Natural Earth. Public domain, naturalearthdata.com/. Lakes, rivers, roads and the international boundary at 1:10 million, redrawn.
Figure 1.5Barton eastbound toward the Dewitt junction.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 1.6The corner from the north-east.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 1.7The junction looking south along Dewitt Road.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 1.8The junction and its neighbourhood.OpenStreetMap contributors. ODbL 1.0, openstreetmap.org/copyright. Roads, schools, land use and building footprints extracted October 2026 and redrawn.
Figure 3.7Drawings from US 9,579,615 B2.“Blender with elevator assembly and removable spindle, US 9,579,615 B2 (February 28, 2017)”, United States Patent and Trademark Office (patentee: f’real Foods, LLC). Public record, image-ppubs.uspto.gov/dirsearch-public/print/downloadPdf/9579615. Sheet headers removed; four drawings set side by side at one height.
Figure 5.6Bobacino’s demonstration bar, Las Vegas; the closed Tea Industries storefront, Richmond, BC.Aaron Kiel, for World Tea News. Press photograph, review scale, worldteanews.com/whats-brewing/fomo-heres-what-you-missed-world-tea-expo-2022-photo-essay. Cropped to the machine; the exhibitors beside it are cut out. Marco Ovies, Dished (Daily Hive). Press photograph, review scale, dailyhive.com/vancouver/tea-industries-richmond. On the lease signs, a realtor’s name, telephone number and e-mail address, a brokerage telephone number, and the poster’s contact lines and QR code retouched blank; resized.
Chapter 06The bench unit, July 2026.Craftpine. Own work. The kiosk software’s wordmark is retouched out of the screen.
Figure 6.4The bench unit, three views.Craftpine. Own work. The kiosk software’s wordmark is retouched out of the screens; the on-screen QR, staged for the shoot, leads to this study.
Figure 6.5The phone checkout and the July 9 charge.Craftpine. Own work. Cropped to the one transaction; no personal data.
Figure 6.7The edge computer on the bench.Craftpine. Own work.
Chapter 08The lot from above, April 2025.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 8.2The vehicle plane at ground level.The study’s site archive. Own work. Ground photography of the site, December 2024 to April 2025; signage, vehicle markings and licence plates retouched blank; resized, metadata stripped.
Figure 8.3The lot from the north-east.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 8.4Two faces of the building.The study’s site archive. Own work. Ground photography of the site, December 2024 to April 2025; signage, vehicle markings and licence plates retouched blank; resized, metadata stripped.
Figure 8.6The rear structure.The study’s site archive. Own work. Ground photography of the site, December 2024 to April 2025; signage, vehicle markings and licence plates retouched blank; resized, metadata stripped.
Chapter 09The corner and its neighbourhood, April 2025.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure 9.5The corner, looking west along Barton Street.The study’s site archive. Own work. Aerial photography of the site, April 2025; signage and vehicle markings retouched blank; colour-graded, resized, metadata stripped.
Figure D.1The category’s operators by home city.Natural Earth. Public domain, naturalearthdata.com/. Land, lakes and boundaries at 1:50 million, redrawn in an Albers equal-area projection.
Figure D.7A Briggo Coffee Haus at Austin-Bergstrom International Airport.“IMG_20191029_040541”, LunchboxLarry (Flickr). CC BY 2.0, flickr.com/photos/squirrel02/48979141078/. Cropped to 4:3 and resized.
Figure D.8An f’real blending bar in a truck-stop store, Ontario.“Coin-op milkshake robot, truck-stop, Ontario, Canada”, Cory Doctorow. CC BY-SA 2.0, flickr.com/photos/doctorow/16113241132/. Resized.
Figure D.10Artly’s coffee bar in MUJI at the Atrium on Bay, Toronto.Marvin Job, TO Cityscapes. Press photograph, review scale, tocityscapes.com/jarvis-the-robot-barista/. Cropped to 4:3.
Figure D.12A teaBOT kiosk at Northeastern University, Boston.“teaBOT at Northeaster University”, Martin Lewison. CC BY 2.0, flickr.com/photos/milst1/49557048233/. Cropped to 4:3 and resized.
Figure D.13Cafe X’s first robot bar, the Metreon, San Francisco.“Coffeebot 3000”, Ed Johnson. CC0, flickr.com/photos/edwardotis/25842108757/. Cropped to 4:3 and resized.