BRENNEN

Industries / Food, farming and forestry / Food and beverage processing / Soft drink bottling

Quality control

Quality control is the testing that keeps every batch safe and the same: checking the water, sugar, juice and carbon dioxide that go in, finding contamination, and pulling bad batches off the shelves.

7

Milestones achieved, from 1916 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1916

    Caffeine ruled an added ingredient

    US officials seized barrels and kegs of Coca-Cola syrup shipped from Atlanta to Chattanooga under the 1906 food and drugs act, alleging that its caffeine was an added harmful ingredient. The company argued that a product made to its own formula and sold under its own name could not contain an "added" ingredient. On 22 May 1916 the US Supreme Court ruled that the caffeine was added, because a formula and a brand name do not take a food outside the law, and sent the question of harm back for trial. The court's record gives the syrup as 52.64% sugar and 42.63% water, mixed at about 1 ounce of syrup to 7 ounces of carbonated water.

    Lessona drink is judged by what is put into it, not by its own recipe.

    law.cornell.edu

    Credit

    Justice Charles Evans Hughes, US Supreme Court; the US government, which brought the case

  2. 1984

    A spoiler survives pasteurising

    After pasteurised apple juice spoiled in Germany in 1982, G. Cerny, W. Hennlich and K. Poralla isolated the cause, a bacterium later named Alicyclobacillus acidoterrestris, whose spores survive the heat used to pasteurise acid juices. It makes no gas and does not swell the pack, so the spoilage shows only as a medicinal or smoky taste, mostly from a compound called guaiacol, once the juice is opened. It does not make people ill. The International Federation of Fruit Juice Producers (IFU) later issued a standard test for it, IFU Method No. 12, and a 2022 review says the trade accepts at most 100 spores per millilitre of raw material.

    Lessonan acid, pasteurised juice can still spoil, so plants test their raw juice for heat-resistant spores.

    pmc.ncbi.nlm.nih.gov

    Credit

    G. Cerny, W. Hennlich and K. Poralla (isolation); IFU working group on microbiology (test method)

  3. 1999

    Bad gas in Belgian Coke

    In June 1999 schoolchildren in Belgium fell ill after drinking Coca-Cola, and 14 million cases of Coca-Cola drinks were recalled in five European countries, The Wall Street Journal reported. The bottler, Coca-Cola Enterprises, traced an off-odour at its Antwerp plant to carbon dioxide holding traces of carbonyl sulphide and hydrogen sulphide, and said the plant had neither tested that batch nor asked the gas supplier for a certificate of analysis; the supplier, Aga, said its retained samples were clean. A case-control study found that regular Coca-Cola explained the first cases at the first school, and that mass sociogenic illness, symptoms spread by anxiety, could explain most of the others. The same year the International Society of Beverage Technologists (ISBT) adopted voluntary guidelines for drinks-grade carbon dioxide, with limits and test methods for 20 parameters, after earlier recalls and a 1997 taste study had shown that the old limit for carbonyl sulphide, 0.5 parts per million, was too loose.

    Lessoncarbon dioxide is an ingredient, so each delivery needs a certificate and a test.

    sites.pitt.eduebi.ac.ukfood-safety.com

    Credit

    Coca-Cola Enterprises (recall); A. Gallay and colleagues (study); ISBT carbon dioxide subcommittee, chaired by Stefanie Ringo of Cargill (guidelines)

  4. 2001

    Juice plants must kill germs

    In October 1996 unpasteurised apple juice from Odwalla's plant at Dinuba, California, infected more than 65 people in the western US and British Columbia with E. coli O157:H7, a bacterium that can cause kidney failure; one child died. FDA inspectors found poor sanitising and decayed fruit accepted from suppliers. Odwalla began pasteurising, and in 1998 it pleaded guilty to 16 federal charges and paid a $1.5 million fine. In January 2001 the US Food and Drug Administration (FDA) required juice makers to run a hazard analysis and critical control point (HACCP) plan, a written set of checks at each step where a hazard is controlled, and to treat juice so as to cut the target germ 100,000-fold, a 5-log reduction. Large plants had to comply from January 2002 and the smallest from January 2004.

    foodsafetynews.comgovinfo.gov

    Credit

    US FDA; the Seattle-King County health department, which linked the first cases to the juice

  5. 2004

    Parliament backs pesticide tests

    In August 2003 the Centre for Science and Environment (CSE), a research group in New Delhi, reported pesticide residues in 36 samples of 12 soft drink brands sold around Delhi. Coca-Cola and PepsiCo, which made them, rejected the findings at a joint press conference. India's Parliament set up a joint committee chaired by Sharad Pawar, which reported on 4 February 2004 that CSE's findings were correct for the samples it tested, that the government's laboratories had also found residues, and that India's food law did not deal with pesticides in soft drinks; it asked for standards.

    Lessonthe makers' own assurances did not settle the question; tests by outside laboratories did.

    downtoearth.org.inexchange4media.com

    Credit

    Centre for Science and Environment; the Joint Parliamentary Committee

  6. 2006

    Benzene designed out of drinks

    Benzene, a chemical that causes cancer, can form at parts per billion in drinks that hold both benzoate preservatives and vitamin C, faster in heat and light. The soft drink industry told the FDA of this in 1990, and many makers reformulated then. After new reports in November 2005, FDA chemists tested 199 soft drinks and other beverages through May 2007 and found 10 above 5 parts per billion, the US limit for drinking water. The makers reformulated them, and retests found less than 1.5 parts per billion. The International Council of Beverages Associations and the American Beverage Association issued guidance on avoiding it.

    Lessonwhat happens to a drink in a hot shop after it leaves the plant counts, so shelf-life tests for these recipes are run warm and under light.

    fda.govebi.ac.uk

    Credit

    Patricia Nyman and colleagues, FDA Center for Food Safety and Applied Nutrition

  7. 2025

    Chlorate from the water plant

    In January 2025 Coca-Cola Europacific Partners recalled drinks made at its plant in Ghent, Belgium, first in Belgium, the Netherlands and Luxembourg and then, for tall cans of Coke, Coke Zero, Diet Coke and Sprite Zero sold to cafés and six-packs of Appletiser, in the UK. The cause was high chlorate, a by-product of the chlorine-based disinfectants used to treat water, which the company traced to one container in the plant's water treatment. The UK Food Standards Agency judged the risk low but said harm could not be ruled out for people sensitive to chlorate.

    Lessonthe water treatment that keeps a drink safe can add a contaminant of its own.

    thechemicalengineer.com

    Credit

    Coca-Cola Europacific Partners (recall); UK Food Standards Agency (risk assessment)

Watch for

  1. India orders 'ORS' drinks removed

    ORS, oral rehydration salts, is a medicine made to a World Health Organization formula for people losing water through diarrhoea. In October 2025 India's food regulator, the Food Safety and Standards Authority of India (FSSAI), barred fruit-based, ready-to-drink and other food products from using "ORS" in their names, calling it misleading, and on 31 October 2025 the Delhi High Court dismissed a challenge by Dr. Reddy's Laboratories. In November 2025 FSSAI said such drinks were still on sale in shops and online, and ordered its officers to inspect, remove them and report what they had done.

    Achieved whenFSSAI publishes its officers' inspection results, or an outside survey of shops and online listings finds no drinks still sold under "ORS" names.

    scconline.comnewsable.asianetnews.com

    Claimed by

    FSSAI, November 2025

Open

  1. 01

    Spot reprinted date codes

    An open set of photos of date codes on drink cans and bottles, genuine and erased-then-reprinted, and a model that flags a reprinted code from one phone photo, for food inspectors and shop buyers.

    The brief
    Credit · unclaimed
  2. 02

    Paint shed by bottle caps

    A bench test that measures how many paint particles metal crown caps and screw caps shed into a drink, compared across cap makers, storage times and cleaning steps, with a tumbler to copy the rubbing of caps against each other in storage and hoppers.

    The brief
    Credit · unclaimed

Open 01 · the brief

Spot reprinted date codes

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open set of photos of date codes on drink cans and bottles, genuine and erased-then-reprinted, and a model that flags a reprinted code from one phone photo, for food inspectors and shop buyers.

Why it is still open

On 31 March 2026 Delhi Police found a unit that used chemical thinners to erase the dates on cans of Thums Up, Sprite, Limca and Coca-Cola and reprinted fresh ones with a coding printer, and in August 2026 Maharashtra's food regulator raided a Navi Mumbai warehouse that relabelled expired PepsiCo, Nestlé, Coca-Cola and Unilever products for export. Published work reads date codes, for example Seker and Ahn's 2022 expiration-date framework and its open dataset, and several GitHub projects read expiry dates from photos, but no dataset or test found labels reprinted codes.

theprint.indailysabah.comhuggingface.co

Built with what exists

Phone cameras and a USB microscope camera for close-ups of ink edges; a handheld inkjet coder of the kind sold for marking cartons, used only on emptied cans bought for the project; isopropyl alcohol to erase codes; Label Studio for labelling; an open detector such as Ultralytics YOLO, or a small vision-language model fine-tuned on one graphics card (see the Note "One GPU to fine-tune").

  1. First step

    Buy 50 cans of five brands, photograph their codes, erase and reprint the codes on half of them once emptied, and check whether close-up photos already show the difference by eye.

  2. Done looks like

    At least 3,000 labelled photos from at least 30 products, published under an open licence. On 500 codes held out from training, the model flags at least 90% of reprints with no more than 5% false alarms on genuine codes, reported separately for each printing method.

Open 02 · the brief

Paint shed by bottle caps

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A bench test that measures how many paint particles metal crown caps and screw caps shed into a drink, compared across cap makers, storage times and cleaning steps, with a tumbler to copy the rubbing of caps against each other in storage and hoppers.

Why it is still open

France's food safety agency, ANSES, found more microplastics in glass-bottled colas, lemonades and beers than in plastic bottles or cans, and traced the particles in one soft drink brand to paint on the outside of its caps. Blowing air on the caps and rinsing them before capping cut that brand's count from 287 to 87 particles per litre. Only one brand was tested, and counts varied strongly even within one batch.

foodpackagingforum.org

Built with what exists

A laboratory tumbler or shaker, glass-fibre filters on a vacuum filtration set, a stereo microscope with a camera, time on an infrared (FTIR) microscope at a university core laboratory to identify each particle's polymer, and an open counting script in OpenCV checked against hand counts.

  1. First step

    Buy 100 crown caps from two suppliers, tumble half of them for an hour in a closed jar, rinse every cap into filtered water, and count the particles on the filters under a microscope.

  2. Done looks like

    Particle counts for at least 20 cap types from at least five makers, before and after at least three cleaning methods, with five repeats each and blank controls, all images and counts public. A second laboratory repeats five of the tests and gets counts within 30% of the first.