Sources and notes

Every figure, and the document behind it.Where we could only reach the claim, we say so.

The rule is that a number on this site has a document behind it, and the document is the primary one: the regulator, the company, the court, the paper. Not a summary of it, and not somebody’s account of it.

Three things can be true of a figure here. Most have a document we have read, and it is named below. Some are claims made by a named party where we could reach the claim but not the data underneath it, and those are marked not verified. A few move over time, a citation count for instance, in which case we name the live source and treat the number on the page as a snapshot. Not verified means we could not reach the underlying document. It does not mean we doubt the claim.

Where we do doubt something, we say that separately, and we correct it here rather than quietly on the page it came from. There is one correction below.

A cyanotoxin at a drinking-water intake

Supports the left-hand column of the sampling-gap figure on the home page.

Ohio Administrative Code 3745-90-03, Harmful algal blooms, monitoring, effective 1 November 2022
The fortnightly cadence. Cyanobacteria screening at least once every two weeks through the bloom season, raw-water microcystins on the alternate week, and routine analysis no later than five days from collection.
Oregon Administrative Rule 333-061-0540, Cyanotoxin monitoring
Routine raw-water cyanotoxin monitoring at least once every two weeks between 1 May and 31 October, escalating to weekly on a result at or above 0.3 µg/L.
US EPA 815-R-15-010, Recommendations for Public Water Systems to Manage Cyanotoxins in Drinking Water, 2015
The five-day analysis window. Samples generally should be analysed within five days of collection.
City of Toledo, water quality page
The utility at the centre of the best-known cyanotoxin incident in the developed world states that microcystin samples are collected weekly and analysed on Tuesday.
Ma, Guerra Maldonado, Zamyadi, Dorner & Prévost (2023), Environmental Monitoring and Assessment 195:1042
The dynamics. A 93-fold change in cyanobacterial signal within 24 hours at a full-scale treatment plant intake, including a rise from 0.6 to 55.8 RFU across four hours, and the finding that afternoon peaks were unlikely ever to be measured by morning grab sampling. On 10 August the probe registered alert-level while that morning’s grab sample returned a result below the vigilance threshold.
Chen, Zhu, Wang, Feng & Xue (2023), Hydrobiologia 850:123
Wind-driven accumulation. At 2.1 m/s about half the Microcystis biomass of the whole water column accumulated in a one-centimetre surface layer, with surface transport at roughly 1.5 to 5 percent of wind speed.
Ohio EPA, Developing a Harmful Algal Bloom (HAB) Treatment Optimization Protocol, version 2.2, August 2022
Pre-oxidation. Do not apply chlorine ahead of filtration, because any dose of chlorine is expected to lyse cyanobacteria cells. Also the dose ranges: 3 to 10 mg/L of PAC in normal operation, 10 to 40 mg/L during an event, and feeder capacity of 40 to 50 mg/L strongly recommended.
Ohio EPA, Guidance for Public Water Systems Developing a Cyanotoxin General Plan
The stock recommendation, quoted exactly: A minimum of two weeks of PAC storage dosed at 40 mg/L or greater should be kept on site or be readily available from the manufacturer/supplier to meet demand. Note the wording is should. This is guidance, not a mandate, and the site says “recommends”.
Greenstein, Zamyadi, Glover, Adams, Rosenfeldt & Wert (2020), Toxins 12(5):335
Delayed release. Partially oxidised cells continue releasing intracellular microcystin after the oxidant is quenched, beginning 8 to 33 hours later in natural bloom samples.
World Health Organization (2020), Cyanobacterial toxins: microcystins, background document for the Guidelines for Drinking-water Quality
The 1 µg/L provisional guideline value for microcystin-LR, and the observation that concentration gradients over depth and area can change within less than an hour.
Water Services (Drinking Water Standards for New Zealand) Regulations 2022
New Zealand’s maximum acceptable value for microcystins is the same 1 µg/L, and unlike the US health advisories it is legally binding.
Taumata Arowai, Drinking Water Quality Assurance Rules 2022, and Small to Medium Networks Acceptable Solution 2025
Both regulate cyanobacteria detection, risk categorisation and response planning. Neither imposes a chemical stocking requirement or a treatment-readiness requirement. This is the basis for the statement that New Zealand has no equivalent of the Ohio recommendation.
Taumata Arowai, Drinking Water Regulation Report 2024
Most council surface water sources continue to be assessed as medium or high risk of cyanobacteria.
Watercare Services, 14 July 2021
The Waikato plant near Tuakau treats up to 50 million litres a day. The 28 tonne figure is arithmetic from that capacity: 50 ML/day at 40 mg/L for fourteen days.
H2O Systems, Ardmore water treatment plant powdered activated carbon dosing plant
Twin dosing trains, 15 m³ powder storage each, dose rates from 25 to 480 kg an hour, built to Zone 22 explosive dust requirements. The replacement plant was operational within six months of contract award.
Watercare, Ardmore information leaflet, 2024
Ardmore supplies around 60 percent of Auckland’s drinking water.
Waikato District Council, Tuakau and Pōkeno Drinking Water Safety Plan; Porter & Rose, Water New Zealand, on the Hamilton plant; Wellington Water on Te Mārua
All three run activated carbon in the filter bed, as biological or granular activated carbon, rather than powdered carbon dosing. Filter-bed carbon has standing capacity that degrades slowly and cannot be surged during an event.
James Cumming & Sons, Australian activated carbon
The comparison. An Australian manufacturer publishes both a general lead time and an event lead time: For Domestic orders, we aim to be able to facilitate delivery within 2 weeks from date of Order, and In the case of Algal taste and odour events, we can often have a truck dispatched within two days.
McCarty and others (2016), MMWR 65(35):926; Steffen and others (2017), Environmental Science & Technology
Toledo, August 2014. Peak microcystin in drinking water of 3.19 µg/L, 440,552 residents across 108,301 households in four counties, a do-not-drink advisory issued 2 August and lifted 4 August.
Bingham, Sinha & Lupi (2015), Economic Benefits of Reducing Harmful Algal Blooms in Lake Erie, for the International Joint Commission
The economic loss of the 2014 Lake Erie event, estimated at about US$65 million.
Novak Consulting Group after-action report, July 2018; Adams and others (2025), Toxins 17(2):53; City of Salem
Salem, Oregon, 2018. Advisories ran 29 May to 3 July, a 33-day period affecting about 200,000 residents, with the National Guard deployed for bulk water distribution and a subsequent ozone facility costing more than US$50 million.
Not verified

No New Zealand supplier, utility or regulator publishes a PAC lead time, a stock tonnage or a delivery term. We searched supplier sites, council asset management plans, drinking water safety plans, long-term plans, Taumata Arowai’s rules and the Hinekōrako public register. The statement that one permanent municipal PAC dosing plant exists in New Zealand is therefore a statement about the published record, not a survey of the country. No public dataset lists which plants have PAC feeders, and the Hinekōrako register cannot be filtered by treatment process.

Acticarb states it holds stock in Auckland and Tauranga but publishes no quantity. Ixom lists PAC for New Zealand in 20 kg bags with the origin given as Australia, China or Sri Lanka, and does not state whether bulk quantities are held locally. Activated Carbon NZ in Taupō describes itself as the only New Zealand producer; its annual tonnage is not published.

The reading of Ardmore’s silos as about a day at full rate assumes a PAC bulk density of 0.25 to 0.5 tonnes per cubic metre. The grade used at Ardmore is not published.

We could not obtain the measured toxin concentrations in Salem’s treated water in 2018, or the concentrations and population for the Carroll Township, Ohio advisory in 2013.

Two things a technical reader should know. NOAA already operates a five-day toxin forecast for western Lake Erie, and Toledo has run phycocyanin sensors at fifteen-minute intervals at its intake. The gap described here is real; being first to close it is not a claim that survives in that geography. And we found no New Zealand drinking-water cyanotoxin advisory in the public record. New Zealand exposure here is prospective, not historical.

Ethylene inside a reefer container

Supports the right-hand column of the sampling-gap figure on the home page.

Carrier Transicold, PrimeLine microprocessor and DataCORDER technical publication
The container controller records supply and return air at selectable intervals of 15, 30, 60 or 120 minutes, holding two years of data at one-hour logging. Optional inputs include up to three USDA pulp probes, fitted for cold-treatment compliance rather than routine quality assurance.
Thermo King, Magnum Plus technical specification
Temperature logs default to a one-hour interval, user changeable, with 15,000 logs of memory. Control sensors are PT1000 Class A, sensor accuracy 0.15 °C.
Zespri, Handling and Quality Guide
One TempTale logger per container in the New Zealand leg, placed on a pallet near the door, described as capturing the warmest air temperature in the container. Wireless loggers on approximately ten percent of pallets, in the pallet core. Storage specification is flesh temperature 0.0 °C plus or minus 0.5, at 92 to 95 percent relative humidity.
Carrier, XtendFRESH controlled atmosphere product documentation
The unit removes ethylene with a self-regenerating activated carbon scrubber and carries sensors for oxygen and carbon dioxide. It has no ethylene sensor. It removes ethylene without ever measuring the quantity it is removing.
Maersk Remote Container Management; MSC iReefer; Emerson cargo monitoring catalogue; Sensitech real-time monitors
Checked for ethylene sensing. Temperature, humidity, oxygen, carbon dioxide, light, position and power status are all available. No product in any of these lines carries an ethylene sensor.
Mayston (2010), Kellogg Rural Leadership Programme, kiwifruit supply chain, quoting Zespri’s coolstore service agreement
The onshore obligation. The contractor agrees to monitor the ethylene levels of the Kiwifruit throughout the storage period and must notify Zespri of readings above acceptable levels. This obligation has no counterpart once the fruit is in a container.
UC Davis Postharvest Center, kiwifruit produce facts (Crisosto & Kader)
The threshold. As little as 5 to 10 parts per billion of ethylene will induce softening, and concentrations should be kept below 20 ppb in storage. Kiwifruit at 0 °C produces less than 0.1 µL/kg per hour; ripe fruit at 20 °C produces 50 to 100.
Massey University research repository, ethylene production by Botrytis cinerea and infected kiwifruit
At 0 °C no ethylene was produced by wounded or healthy fruit; only infected fruit produced it. With four infected fruit in a tray the softening rate of the healthy fruit was faster than with one.
Lukasse, Staal & Wissink (2019), International Congress of Refrigeration, article 54
Within-container heterogeneity, measured. Six 40-foot containers instrumented with 31 loggers each inside cartons: warmest-to-coldest cargo temperature spread of 2.3 to 2.9 °C across the trip, rising to 2.9 to 3.6 °C during peak ambient heat, with the warmest spots at the door end and against the ceiling.
New Zealand Kiwifruit Growers Incorporated, Kiwifruit Industry Performance Report 2025
22,582 pallets condition-checked in 2024, representing 7.5 percent of pallets shipped. Cost of quality in 2024/25 of about NZ$292.3 million across SunGold, Green and the organic pools, covering fruit loss, repacking, condition checking, claims and discounts. Claims by variety in 2024: Green conventional 3.8 percent, Gold conventional 2.1 percent, Green organic 3.4 percent.
Zespri, Kiwiflier issue 484, July 2026
The 22 percent figure, quoted exactly: In 2025, claims on fruit sold from November reached as high as 22% - more than one in five pieces of fruit failed to meet our standards. This is a late-season figure in a discussion of the China market, not a season-wide rate. The season-wide claims rates are the NZKGI figures above.
Zespri, Pool Policy Manual, 2026/27 season
The Sales Forecast is the allocation of New Zealand sourced kiwifruit to market by class, variety and growing method, against a crop estimate. None of those variables is how long the fruit will last.
Frontiers in Plant Science 10:888 (2019)
The counterweight, included because it cuts against the simple version of the story. Kiwifruit also softens at low temperature without ethylene, and Zespri’s offshore defect reporting is dominated by storage breakdown disorder, which is a chilling disorder. A warm excursion is one loss mechanism among several, and not necessarily the largest.
Not verified

The recording interval of the TempTale and wireless loggers as configured by Zespri is not published, and the manufacturers do not publish sampling intervals on public product pages.

We have no sourced figure for the value of a single rejected container, and none is used. A per-container value can be derived from export value and tonnage but the trays and tonnes per container are assumptions, so we do not state one.

We could not establish whether reefer setpoints can be changed remotely mid-ocean on any carrier’s public documentation, and we found no kiwifruit-specific marine cargo insurance claim data.

An earlier version of this argument said the damage passes between two logger reads. That was wrong and has been removed. Reefer controllers log continuously at 15 to 120 minute intervals and there is no temporal gap for a thermal excursion to hide in. The gap is the variable, not the cadence.

Nature versus maths

Supports the regulator figure and the counting-states argument on the home page.

W. Ross Ashby, An Introduction to Cybernetics, Chapman & Hall, London, 1956, §11/7, p.207
The law of requisite variety. The operational form at §11/6 reads: Only variety in R’s moves can force down the variety in the outcomes. Pinpoints are read directly off the printed page.
R. C. Conant & W. R. Ashby (1970), International Journal of Systems Science 1(2):89–97
Every good regulator of a system must be a model of that system. The abstract states that any regulator that is maximally both successful and simple must be isomorphic with the system being regulated. The result is narrower than it is usually quoted as being: it concerns regulators that are optimal and simple, under stated assumptions, and optimal is defined as minimising the entropy of outcomes.
Not verified

The page number for the formal theorem in Conant and Ashby is unconfirmed. The copy read is a re-typeset edition whose pagination does not correspond to the 1970 journal. The quotations themselves are verbatim.

Whether the proof has been criticised or qualified in the citing literature has not been checked. A fifty-six-year-old theorem in a small field has very likely been examined, and if it has been qualified we would rather cite the qualification than be shown it.

Requisite variety is used here as a frame, not as authority. Whether any particular system possesses variety adequate to the living system it manages is an empirical claim about that system, and nothing on this site asserts it.

The cards on the home page

Six shapes of the same problem. Five of the six carry a qualitative statement of cost rather than a figure, because we do not have a figure we can source for them. The sixth carried a number, and it was wrong.

Honig, Miron, Lehrer, Jackoby, Zachut, Zinou, Portnick & Moallem (2012), Journal of Dairy Science 95(7):3736–3742
DOI 10.3168/jds.2011-5054. High-yielding cows under a hot and humid climate, given five or eight cooling sessions a day. Milk yield was 9.6 percent higher in the eight-session group, 40.1 against 36.6 kg a day, and dry matter intake 9.3 percent higher, 27.0 against 24.7 kg a day. Same animals, same barn, same equipment: the difference is how hard the cooling is run. This is the figure now on the heat-stressed dairy card.
Corrected, 30 July 2026

That card previously read ~30% daily milk lost, summer versus winter. We traced the figure, and it does not support the claim. It has been replaced.

Where it came from. Altyeb, Absy, Ghanem, Sharawy & Hassan (2025), BMC Veterinary Research, DOI 10.1186/s12917-025-05039-6. One commercial Holstein farm in Ismailia Governorate, Egypt, over 2023 and 2024. Summer THI 79.0, winter THI 59.1. Daily milk yield 31.2 kg in summer against 40.9 kg in winter.

The arithmetic was wrong. The gap is 9.7 kg. Against the winter figure, which is the right denominator for milk lost, that is 23.7 percent. Against the summer figure it is 31.1 percent, which is winter expressed as a premium over summer, a different statement. The 30 percent was the second number used as though it were the first.

The population was wrong. The paper describes the cows as housed in traditional farms with free, semi-shaded yards equipped with electric fans. That is not cooled confinement. The card describes housed Gulf and desert dairy, which is the opposite kind of system. The study also had 25 cows in the summer group against 93 in winter, and does not report days in milk, so the gap is not attributable to season alone.

The magnitude is wrong for cooled barns. Published temperature-humidity slopes for hot climates run about 0.25 to 0.47 kg per THI unit above a threshold near 72, which puts a rise from THI 60 to 79 at roughly 3 to 9 percent, not 30. And Baladna, the Qatari dairy usually cited in this context, reports the opposite: 36.6 litres per cow per day in 2021 against 36.7 in 2020, a fall of 0.3 percent, through what its annual report calls the hottest summer in four years, with our essential cooling systems… functioned perfectly.

Which is a better argument anyway. A well-run desert herd is not losing thirty percent of its milk, and the people running one know it. What Honig and others measured is the gap between adequate cooling and intensive cooling in the same barn, and that gap is about a tenth of the milk. The question is not whether the cooling exists. It is where the finite water, power and labour behind it should go on any given day.

The five decisions, worked all the way down

These lists also appear at the foot of each deep-dive page, where they belong with the argument. They are repeated here so that everything the site asserts sits in one place.

Cyanotoxins at a drinking-water intake Read the decision →

The research behind the water column on the home page, now carried in full on its own page. The sampling and turnaround figures are all from regulation or from a utility’s own published wording, not from trade press.

  • Ma, L., Guerra Maldonado, J.F., Zamyadi, A., Dorner, S. & Prévost, M. Environmental Monitoring and Assessment 195(9):1042, 2023. A phycocyanin probe against the grab-sampling regime at a full-scale intake: a 93-fold change within 24 hours; 0.6 units at 12:30 rising to 55.8 at 16:30; seasonal mean 1.1; grab sampling at 08:00 to 10:00 with the authors concluding the afternoon peaks were unlikely to be ever measured. On 10 August the probe registered an alert-level 10.9 while that morning’s sample returned 0.02 mm³/L, below the vigilance threshold. Every number in the figure on that page. doi.org
  • Ohio EPA, Developing a Harmful Algal Bloom Treatment Optimization Protocol. The quotation is exact: Do not apply chlorine ahead of filtration, because any dose of chlorine is expected to lyse cyanobacteria cells. Also the powdered-carbon dose ranges, 3 to 10 mg/L routinely and 10 to 50 through an event, and drawing the offtake below a surface bloom.
  • Ohio Administrative Code 3745-90-03, effective 1 November 2022, and Oregon Administrative Rule 333-061-0540. Both set routine raw-water cyanotoxin sampling at once every two weeks through the season. Ohio allows analysis no later than five days from collection and requires a 24-hour turnaround once an action level is exceeded.
  • City of Toledo, water quality page: Water samples for microcystin are collected weekly and analyzed on Tuesday. The utility with the most institutional experience of this hazard runs a weekly cadence with a fixed analysis day.
  • Greenstein, K.E. and others, Toxins 12(5):335, 2020, for delayed release after partial oxidation: more than 96 hours after quenching, beginning 8 to 33 hours later in natural bloom samples. McKindles, K.M. and others, Applied and Environmental Microbiology 86(22):e01397-20, 2020, for viral lysis at Toledo’s own intake over three to four days.
  • Chen, H. and others, Hydrobiologia 850:123–135, 2023: at 2.1 m/s wind, about half the Microcystis biomass of the whole water column accumulated in a 1 cm surface layer, transported at roughly 1.5 to 5 percent of wind speed.
  • WHO, Cyanobacterial toxins: microcystins, WHO/HEP/ECH/WSH/2020.6, for the guideline values and the verbatim statement that gradients may be very steep and often change rapidly (within less than 1 hour); Toxic Cyanobacteria in Water, 2nd edition, chapters 5, 10 and 11, for growth rates, treatment barrier performance and monitoring design.
  • McCarty, C.L. and others, MMWR 65(35):926–929, 2016, for Toledo 2014: 3.19 µg/L on 1 August, 440,552 residents, the advisory issued 2 August and lifted 4 August, and self-reported gastrointestinal symptoms in 16.2 percent of households with no confirmed toxin-attributable illness. Bingham, M., Sinha, S.K. & Lupi, F., to the International Joint Commission, October 2015, for the roughly US$65 million loss.
  • Taumata Arowai, Water Services (Drinking Water Quality Assurance) Rules 2026, rules S3.CD.7 and S3.CD.8: microscopy weekly for high-likelihood sources, monthly for medium. Drinking Water Regulation Report 2024: most council surface water sources continue to be assessed as medium or high risk.
  • NOAA Great Lakes Environmental Research Laboratory, Lake Erie HAB Tracker and forecast, for the operational five-day product updated daily through bloom season, including the vertical distribution of buoyant colonies; and the published method for forecasting toxin concentrations five days out. Reported because it cuts against any claim to be first.
Not verified

Deliberately absent from the water page because we could not source them: the measured concentrations, population served and advisory duration for Carroll Township in 2013, which official presentations reference without stating; the measured raw and treated concentrations at Salem in 2018; a citable colony flotation velocity; the total elapsed time for an ELISA plate run; and a New Zealand powdered-activated-carbon procurement lead time in days, for which no supplier, utility or regulator publishes a figure.

One further absence is itself a finding. Extensive searching found no New Zealand drinking-water cyanotoxin advisory or maximum-acceptable-value exceedance in the public record. New Zealand’s exposure here is prospective rather than historical, and the water page says so rather than implying a local precedent that does not exist.

Two official Toledo figures are not reconcilable and both are given where used: the MMWR reports 3.19 µg/L in drinking water while an Ohio EPA chart peaks at 2.50 µg/L. They are likely different sampling points or laboratories. We do not average them.

Heat stress in confinement dairy Read the decision →

  • Baladna. 2024 Sustainability Report: 22,347 cows; 35.2 litres per cow per day; roughly 10,000 m³ of recycled water a day for livestock cooling; 60 percent water reduction from the smart soaking system. 2023 Annual Report, Operating Review Farming: the SmartSoaker pilot and its 65 percent water reduction. 2025 Annual Report: the Algeria joint venture at 51 / 49 with the National Investment Fund, USD 3.5 billion, 117,000 hectares, 240,000 to 270,000 cattle, 1.7 billion litres a year, milk production by late 2027; a 16.25 percent holding in Juhayna; accelerating discussions in additional African markets. Corporate news, 30 April 2026: USD 635 million of second-phase contracts and the airlift of 30,000 cattle on 109 flights. baladna.com
  • Al-Dobashi, H. & Wright, S. “Developing the Desert: How Qatar Achieved Dairy Self-Sufficiency Through Baladna.” Sustainability 16(24):11262, 2024. The 72 percent import share before the blockade, and the herd history. doi.org/10.3390/su162411262
  • Cheng, W.L., Saleem, A. & Sadr, R. “Recent warming trend in the coastal region of Qatar.” Theoretical and Applied Climatology 128(1):193–205, 2017. Thirty-year coastal record: maximum temperatures rising about 0.5 °C a decade, minimum temperatures about 1 °C a decade, tropical nights up by roughly thirty days. doi.org
  • Tao, S. & Dahl, G.E. “Symposium review: The influences of heat stress on bovine mammary gland function.” Journal of Dairy Science 101(5):5642–5654, 2018. Milk yield of lactating dairy cows is negatively affected by heat stress when average THI exceeds 68; the four to five kilograms a day carry-over into the following lactation; the effects on the calf. doi.org
  • Oliveira, C.P., de Sousa, F.C., da Silva, A.L., Schultz, E.B., Valderrama Londoño, R.I. & de Souza, P.A.R. “Heat Stress in Dairy Cows: Impacts, Identification, and Mitigation Strategies. A Review.” Animals 15(2):249, 2025. Verbatim: Currently, an acceptable THI limit for high-producing dairy cows (above 35 kg·day−1) is up to 68. This is the peer-reviewed restatement of the threshold. The original re-evaluation, Zimbelman and others (2009), is conference proceedings rather than a journal paper, so it is cited through this review. doi.org/10.3390/ani15020249
  • Tao, S. & Dahl, G.E. “Invited review: Heat stress effects during late gestation on dry cows and their calves.” Journal of Dairy Science 96(7):4079–4093, 2013.
  • Fabris, T.F., Laporta, J., Skibiel, A.L., Corra, F.N., Senn, B.D., Wohlgemuth, S.E. & Dahl, G.E. “Effect of heat stress during early, late, and entire dry period on dairy cattle.” Journal of Dairy Science 102(6):5647–5656, 2019. Early-lactation yield of 32.9 kg a day for cows cooled the entire dry period against 26.6 for those heat-stressed throughout; the advantage still present at thirty weeks.
  • Rodríguez-Godina, I.J., García, J.E., Morales, J.L. and others. “Effect of heat stress during the dry period on milk yield and reproductive performance of Holstein cows.” International Journal of Biometeorology 68:883–890, 2024. The smaller estimate: about 235 kg across 305 days, and no significant difference in total milk yield. Reported here because it disagrees with the figures above. doi.org
  • Wolfenson, D. & Roth, Z. “Impact of heat stress on cow reproduction and fertility.” Animal Frontiers 9(1):32–38, 2019. Conception falling from about 42.6 percent in winter to 27.7 percent in summer across an eighteen-year dataset; milk held at 98 percent of winter under efficient cooling while conception reaches only 68 percent; extreme summers costing a further five percentage points. doi.org
  • Levit, H., Pinto, S., Amon, T. and others. “Dynamic cooling strategy based on individual animal response mitigated heat stress in dairy cows.” Animal 15:100093, 2021. Individualised sensor-based cooling against a fixed schedule: heat-stress exposure 5.03 against 9.46 hours a day; improved fat, protein and energy-corrected milk; identical daily raw milk yield of 44.7 kg.
  • Israeli Dairy School, Ofier Langer. “Cow cooling, the effective Israeli way for better dairy farming.” The wet-then-dry method, applied in the waiting rooms before milking, with drying under even airflow of about two metres a second, and the cadence stated as We milk our cows three times a day. So, we keep our cows cool by doing these three cooling cycles every day. A teaching page rather than a study; it reports no production or fertility figures, and none are used here. dairyschool.co.il
  • Casarotto, L.T., Amaro, F.X., Lance, J.M. and others. “Reducing water usage to cool cows by applying smart technologies.” JDS Communications 7(2):273–277, 2025. The same soakers under a fixed run cycle against a cow-presence trigger: 184.6 litres a cow a day against 36.1, with no difference in milk yield. doi.org
  • Flamenbaum, I. & Galon, N. “Management of heat stress to improve fertility in dairy cows.” Journal of Reproduction and Development 56(Suppl.):S36–S41, 2010. Season-long intensive cooling at fifty-nine percent conception against seventeen; cooling restricted to the window around insemination at thirty-one against thirty-six. Separate experiments with separate controls, not a head-to-head trial. doi.org
  • Ferreira, F.C., Gennari, R.S., Dahl, G.E. & De Vries, A. “Economic feasibility of cooling dry cows across the United States.” Journal of Dairy Science 99:9931–9941, 2016. About eleven pounds a day more in the subsequent lactation for cooled dry cows, roughly 448 kilograms a cow a lactation, around 810 million dollars a year nationally, and a payback under six years on the equipment.
  • Montevecchio, A.B. & Chebel, R.C. “A randomized trial on the effects of heat abatement during the pre-weaning phase on growth and reproductive performance of heifers and health, reproductive and productive performances of cows.” Preventive Veterinary Medicine 230:106259, 2024. The only randomised trial found that follows a cooling allocation decision through to first-lactation fertility, and the cooled group did worse. Reported here because it cuts against the argument. doi.org
  • Espinoza-Sandoval, O., Calsamiglia, S. and others. “Modeling the profitability of investing in cooling systems in dairy farms under several intensities of heat stress in the Mediterranean.” Journal of Dairy Science, 2023. Twenty-one locations: cooling recovering 173 to 859 litres a cow a year, net margin from minus 24 to plus 239 euros, and unprofitable below a threshold annual heat load.
  • Woodward, S.J.R., Edwards, J.P., Verhoek, K.J. & Jago, J.G. “Identifying and predicting heat stress events for grazing dairy cows using rumen temperature boluses.” JDS Communications 5(5):431–435, 2024. The model outperforms the standard heat-load index; the paper states that a prediction lead time was not evaluated and that a regional forecast is future work. doi.org
  • Ortiz, X.A., Smith, J.F., Villar, F. and others. “A comparison of 2 evaporative cooling systems on a commercial dairy farm in Saudi Arabia.” Journal of Dairy Science 98(12):8710–8722, 2015. Four hundred multiparous Holsteins through a Gulf summer at 56 kg a day; no difference in milk yield, core temperature or conception between the two cooling systems tested.
  • General Authority for Statistics, Kingdom of Saudi Arabia. Livestock Statistics 2023: about 234,000 dairy cows in specialised holdings plus 39,000 in non-specialised, and 2.8 billion litres of raw milk. Almarai, Annual Report 2025: eight dairy farms, over 189,000 cows, 1.5 billion litres; Sustainability Report 2024: 166,251 cows, all in enclosed housing. The two Almarai figures are on different bases and are not reconcilable from public disclosure.
  • Al Ain Farms and Al Rawabi company pages for herd descriptions; Gulf News, 17 July 2019, for the 850 fans and the 1,800 against 400 cubic metres of water a day. Trade press, seven years old.
  • Nedap Livestock Management case study for Kuwait United Dairy Company: 2,800 head, 1,400 milking, yield from eighteen to twenty-eight litres. A vendor account, not a company disclosure.
  • USDA Foreign Agricultural Service. Egypt: Livestock and Products Annual, EG2023-0023, 3 November 2023: 4.525 million dairy cows and around seventy percent of animal husbandry in small and medium enterprises. China: Dairy and Products Annual, CH2024-0133, 23 October 2024: 70 to 80 percent of raw milk from large-scale farms.
  • China Youran Dairy Group (HKEX 9858) FY2024 filing: 621,568 dairy cows, 324,908 milkable, 12.6 tonnes a milkable cow. China Shengmu Organic Milk (HKEX 1432) FY2024 filing: 156,481 head, 12.05 tonnes. China Modern Dairy (HKEX 1117) FY2025 results: 456,945 head.
  • Wang, T., Zhong, R. & Zhou, D. “Temporal-Spatial Distribution of Risky Sites for Feeding Cattle in China Based on Temperature/Humidity Index.” Agriculture 10(11):571, 2020. Most areas above the danger threshold of 79 from June to September; up to 140 kg a cow a month lost in Xinjiang; 90 to 120 kg in Hebei, Shandong and Henan. doi.org
  • USDA National Agricultural Statistics Service. Milk Production, released 22 May 2026: milk cows by state for April 2026. These move monthly.
  • Skidmore, M., Hutchins, J. & Nolan, D. “Vulnerability of US dairy farms to extreme heat.” Food Policy 131, 2025. About 1.4 billion pounds of energy-adjusted milk lost over five years across 18,000 herds, roughly 245 million dollars; extreme-stress-day loss more than double a moderate day.
  • Vendor wording, quoted from their own pages, July 2026. smaXtec on detection and its THI 72 alert; Nedap on detecting four levels updated every fifteen minutes; Allflex and MSD on the Group Heat Stress Report; Afimilk on panting and its twenty percent group alert; Connecterra on spotting early signs.
  • Also read, and reported on Sources and notes: the figures we could not verify, including a temperature-humidity index day-count for Qatar, the corporate share of Egypt’s herd, and NADEC and Al Safi herd sizes. None of those appears above.
Not verified

Deliberately absent from that page because we could not verify them: a temperature-humidity index day-count for Doha or Qatar; the corporate or confinement share of Egypt’s dairy herd; herd sizes for NADEC and Al Safi in Saudi Arabia; a Baladna milking-cow split, a 2023 comparator for its 2024 yield, or any 2025 herd figure, none of which the company publishes; a committed Baladna project in Egypt or Nigeria, both of which exist only as trade-press reports; and the housing-type breakdown for US Southwest herds.

The St-Pierre and others (2003) heat-stress cost figures are widely quoted and are not used on that page, because every route to the primary text was paywalled and we would only have them at second hand.

Three findings are reported on that page specifically because they cut against the argument: Baladna’s own published yield shows no hot-climate penalty; the closest published trial to what we do found no gain in raw litres; and the one randomised trial that follows a cooling allocation decision through to fertility found the allocation made things worse.

The largest gap is an absence, and we characterised it rather than assumed it. We searched for a controlled study that varies a cooling policy on unchanged equipment and measures conception rate or next-lactation yield. There is not one. Roughly forty searches and forty-five documents, including the full citing list of Levit and others (2021), four recent reviews, two doctoral theses, and the conference and extension literature. Elsevier, Wiley, PubMed Central and the ADSA abstract archive all blocked automated access at some point, and the abstract archive is the one place a poster could still be hiding. Separately, we found no study at all in which a forecast, rather than a current reading, drove a cooling decision with a measured animal outcome.

Varroa and the almond window Read the decision →

  • Mshelbwala PP, Mulholland S, Doyle T, Anderson C, Hetherington S. “Spatial and environmental drivers of Varroa destructor detection in New South Wales, Australia.” Scientific Reports 15, 44438 (4 December 2025). All authors, NSW Department of Primary Industries and Regional Development. nature.com
  • Boehm Vock L, Mossman LM, Rapti Z, Dolezal AG, Clifton SM. “Spatiotemporal, environmental, and behavioral predictors of Varroa mite intensity in managed honey bee apiaries.” PLOS ONE (7 August 2025). Trained on 266 apiaries, 2018–2019; tested out of sample on 140 apiaries, 2020–2021. journals.plos.org
  • Agriculture Victoria. “Miticide resistance in Varroa mites.” Pyrethroid resistance (L925I) and formamidine resistance (Y215H) by state and date; the four unaffected products and the four no longer recommended. Page last updated 3 July 2026. agriculture.vic.gov.au
  • PIRSA (Department of Primary Industries and Regions, South Australia). “Varroa mite.” Acaricide cost of about $50 to $60 per hive per year based on four treatments; withholding period before extraction or residue testing; monitoring every three to four weeks within 25 km of a detection site. Page last updated 18 June 2026. pir.sa.gov.au
  • North West Farmer. “Gearing up for pollination” (1 July 2026). 300,000 hives into orchards by late July; the almond industry paying the beekeeping community in the vicinity of $60 million across a four-week blossom phase. northwestfarmer.com.au
  • Australian Honey Bee Industry Council. Annual Report 2025, Chair’s report (Stephen Targett). “As of December 2024, Australia’s registered hive numbers have decreased by approximately 5% to 823,291. The biggest decrease has been in NSW (-13%) where varroa is the most widespread.” 1,738 commercial and 44,172 recreational registered beekeepers; average honey prices around $4.50/kg with major packers holding excess stock; the agreed 1c/kg levy increase to repay the federal government, to remain until industry’s share of the varroa response is repaid; T2M ceasing February 2026; and, on the season, that “some beekeeping businesses resorted to under cutting prices to obtain pollination contracts for business cashflow”.
  • Professional Beekeepers (Extension Australia). “Almond pollination: hive management.” “Colonies should have a minimum of six frames of bees, with eight or more being highly desirable.” extensionaus.com.au

Selective dry-off Read the decision →

  • Veterinary Council of New Zealand. Case study: Selective Dry Cow Therapy reducing antimicrobial use and resistance, 12 May 2026. The updated prescribing advice for DCAT “comes into effect on 1 January 2027”; the Taranaki Veterinary Centre figures (80% of adult cows on DCAT in 2014 to 50% now, of which the bulk is the 19% of herds on whole-herd; teat sealants from 35% in 2014 to 65% in 2025); and the 2023 MPI audit quotation. vetcouncil.org.nz
  • Veterinary Council of New Zealand. Statement on the Authorisation of Dry Cow Therapy. Per-animal justification with individual cow SCC; the minimum record set; and Technote 14 Table 3 (2025), the two bulk-milk criteria under which whole-herd DCAT may be justified. vetcouncil.org.nz
  • DairyNZ. Technote 14: Dry Cow Strategy, January 2025. The 150,000 / 200,000 / 250,000 cells/mL thresholds, and the sensitivity and specificity of the >200,000 cut-point at the last herd test within 80 days of drying off. dairynz.co.nz
  • Rowe SM, Godden SM, Nydam DV, Gorden PJ, Lago A, Vasquez AK, Royster E, Timmerman J, Thomas MJ. “Randomized controlled non-inferiority trial investigating the effect of 2 selective dry-cow therapy protocols on antibiotic use at dry-off and dry period intramammary infection dynamics.” J. Dairy Sci. 103(7):6473–6492, 2020. doi.org/10.3168/jds.2019-17728
  • Rowe SM et al. “Randomized controlled trial investigating the effect of 2 selective dry-cow therapy protocols on udder health and performance in the subsequent lactation.” J. Dairy Sci. 103(7):6493–6503, 2020. The companion udder-health paper. doi.org/10.3168/jds.2019-17961
  • DairyNZ. New Zealand Dairy Statistics 2023–24. “The lowest-ever average somatic cell count of 161,000 cells/mL in 2023/24”; 10,485 herds; average herd size 448; 77.1% of cows herd-tested. dairynz.co.nz
  • DairySmart. Cost of mastitis in New Zealand, citing DairyNZ: an average clinical case at an estimated total economic cost of $496, and a 300-cow herd at $54,500 a year, or $180 per cow. Also, from DairySmart’s own data, that approximately 20% of high-SCC cows grow no bugs. dairysmart.nz
  • DairyNZ. SmartSAMM Guideline 14: Dry Cow Strategy, January 2025. The farmer-facing companion to Technote 14: “All cows that have a SCC over a threshold should be considered ‘infected’. Discuss with your vet as to the threshold suitable for your herd, which will be between 150,000 cell/s mL and 250,000 cells/mL.” dairynz.co.nz
  • New Zealand Veterinary Association. Position statement on bobby calves and restricted veterinary medicines. Strict application of milk withholding periods for dry cow therapy rather than a general clean-milk rule; calves treated directly with antimicrobials are not eligible for bobby calf slaughter. nzva.org.nz
  • Vet Clinic Morrinsville. “Dry cow therapy legislation changes”, Jackie Davies, 11 March 2026. How one practice is preparing its clients. vetclinicmorrinsville.co.nz
  • Inside Dairy / Farmers Weekly. “New dry cow advice supports selective use”, 14 April 2026. DairyNZ veterinarian Mitch Cooper; fewer than 5% of herds meet the bulk milk somatic cell count threshold above 250,000 that would indicate blanket treatment. farmersweekly.co.nz

Tuna allocation under a moving stock Read the decision →

  • WCPFC. CMM 2025-02: Conservation and Management Measure for Bigeye, Yellowfin and Skipjack Tuna in the Western and Central Pacific Ocean. Paragraph 27: the high-seas limits in Attachment 1 “do not confer the allocation of rights to any CCM”, and the commitment to commence in 2026 and reach agreement in 2027. The measure runs to 15 February 2028. cmm.wcpfc.int
  • WCPFC. The western and central Pacific tuna fishery: 2024 overview and status of stocks, WCPFC22-2025-1P16, 27 November 2025. High-seas catches at approximately 12% of the total in 2024, 17% in 2023, at least 25% across 1990–2007; provisional total 2024 catch of 3,059,005 t, a record. meetings.wcpfc.int
  • WCPFC. Scientific Committee, twenty-second regular session, 11–19 August 2026, Apia, Samoa. meetings.wcpfc.int
  • WCPFC Circular 2026/52, 22 June 2026. The SPC seminar on climate change impacts on WCPFC tuna stocks, Apia, 10 August 2026; the move from a CMIP5 to a CMIP6 ensemble and from two-degree to one-degree resolution, prepared “through financial support from the New Zealand Government”.
  • Forum Fisheries Agency. Milestone agreement on South Pacific albacore allocation, July 2025; South Pacific Albacore Management Procedure adopted at WCPFC22, December 2025. ffa.int
  • Bell JD et al. “Pathways to sustaining tuna-dependent Pacific Island economies during climate change.” Nature Sustainability 4:900–910, 2021. Projected reduction in annual fishing access fees to ten Pacific SIDS averaging about US$90 million a year by 2050 under RCP 8.5, in a range of about US$40–140 million. doi.org/10.1038/s41893-021-00745-z
  • Convention on the Conservation and Management of Highly Migratory Fish Stocks in the Western and Central Pacific Ocean. Article 10, on the Commission’s functions including total allowable catch and allocation criteria; Article 12, on the Scientific Committee. wcpfc.int

Which fruit survives the lane Read the decision →

  • Zespri. Kiwiflier issue 484, July 2026. The China market discussion document: the four trends, the 22 percent claims figure on fruit sold from November 2025, our premium depends on trust, the two options put to growers, and Chinese G3 availability from September to March.
  • Zespri Group Limited. Zespri Pool Policy Manual, 2026/27 season. The definition of the Sales Forecast as the allocation of NZ sourced kiwifruit to market by class, variety and growing method, based on a crop estimate; and the definitions of trays submitted, net submit, trays supplied and onshore fruit loss.
  • Zespri. Kiwiflier issue 444, February 2023, “Quality Action Plan Update”. Frank Bollen, Zespri Technical Manager, on damage almost impossible to see at the time of packing; and on impact damage taking five or more days to show damage. zespri.com
  • New Zealand Kiwifruit Growers Incorporated. Kiwifruit Book, chapter six, Harvest and Postharvest Practices. Curing at about 48 hours; packing as the key control point where the fruit is segregated into market-acceptable product; controlled-atmosphere storage; reefer charters direct to Europe, Japan, China and Korea a journey of between 2-4 weeks; the historical split of about 40 percent reefer to 60 percent container liner; the Fruit Monitoring Technician aboard each reefer from mid-March to mid-June; and KiwiStart and Time Payments. nzkgi.org.nz
  • Zespri. Annual Report 2024/25. Quality costs were around $1.40 for SunGold and $1.80 per tray for Green, where we saw rapid deterioration at the tail end of the season; 220.9 million trays sold; $5.031 billion in net global kiwifruit sales. canopy.zespri.com
  • New Zealand Kiwifruit Growers Incorporated. Kiwifruit Industry Performance Report 2025. Cost of quality per tray equivalent by variety, including Green at $0.75 in 2023 and $1.80 in 2024. nzkgi.org.nz

One wellboat, eleven sites, one summer Read the decision →

  • New Zealand King Salmon. “A step change for NZKS: introducing New Zealand’s first wellboat.” The Ronja King; its use for relocating fish between farm sites, stock grading operations, and early removal of underperforming fish; and that New Zealand had been the only major salmon-producing country not using this technology. kingsalmon.co.nz
  • New Zealand King Salmon. NZX announcement, 9 February 2026, and 1HY26 investor presentation, May 2026. The Sølvtrans lease, eight years at approximately $8.9 million a year; arrival late April and operation from late May 2026; the stated additional harvest of 2,000 metric tonnes annually; the wellboat as non-negotiable infrastructure for open ocean farming; and the current site list and fallow status. kingsalmon.co.nz
  • New Zealand King Salmon. NZX announcement, 28 April 2026: the two Blue Endeavour pilot pens installed, with fish delivery scheduled late May to early June 2026. Consent granted March 2024 for 35 years, 10,000 metric tonnes at full operation. nzx.com
  • RNZ. 25 May 2022 and 28 September 2022. 1,269 tonnes of dead fish and waste over one summer; the $55.7 million net loss for FY2022; and the first-half mortality cost. Mortality of 42% in warm-water areas against 37% in cooler towed water is from the same period, reported at the time. rnz.co.nz
  • Ministry for Primary Industries. Salmon biosecurity response, closed 2024. The September 2023 independent risk assessment finding mortalities resulted from a number of other causes, including environmental factors (for example, warmer water) and farming practices, with the bacteria having played only a minor role in the deaths. mpi.govt.nz
  • Sanford Limited. FY25 Annual Report and FY25 Sustainability Report. We are close to our maximum harvest from Big Glory Bay, due to nitrogen cap constraints; there is a maximum amount of feed we can use in a year due to a nitrogen constraint condition of our environmental consent; salmon are a cold-water breed and begin to die of stress at around 20° Celsius; a record 5,623 gross weight tonnes harvested in FY25; and oxygenation equipment deployed as a mitigation. sanford.co.nz

The team, and the research

Every claim about a scientist on this site is a claim about a public record. Each name links to that person’s Google Scholar profile from the team section on the home page, which is where citation counts, h-indices and publication lists can be checked directly and where they will be current rather than frozen at the date we wrote them.

Bergmeir, de Nijs, Genov, Sriramulu, Abolghasemi, Bean and others, Predict+Optimize Problem in Renewable Energy Scheduling, IEEE Access vol. 13, pp. 60064–60087, 2025
DOI 10.1109/ACCESS.2025.3555393. Four of ruru’s core scientists are among the authors. This is the benchmark referred to on the home page.
Smyl, Bergmeir, Dokumentov, Long, Wibowo & Schmidt, Local and global trend Bayesian exponential smoothing models, International Journal of Forecasting, 2025
Listed under selected publications on the home page.
Bandara, Hyndman & Bergmeir, MSTL: seasonal-trend decomposition for time series with multiple seasonal patterns, International Journal of Operational Research, 2025
Listed under selected publications on the home page.
Zang, Zhang, Harabor, Stuckey & Li, Online guidance graph optimization for lifelong multi-agent path finding, AAAI, 2025
Listed under selected publications on the home page.
Open-source software attributed to the team
The R forecast package, NeuralProphet, Kats and MiniZinc. Authorship of each is stated in the project’s own repository and documentation, which is the record to check.
Not verified

The individual credentials on the team cards are stated on the strength of the public record rather than a document we hold. That covers: 13,200+ citations, Clarivate top 1 percent and ARC DECRA (Bergmeir); AAAI Fellow, ACM Distinguished Scientist, Google Eureka Prize and h-index 71 (Stuckey); top 0.75 percent most-cited AI researchers and the ACM Transactions on Probabilistic Machine Learning co-editorship (Buntine); ARC DECRA, third globally at the NeurIPS 2021 ML4CO competition and PhD TU Delft (de Nijs).

The same applies to the three aggregate figures: 50,000+ citations across the team’s work, 1,000+ combined peer-reviewed publications, and top 1 percent Clarivate highly-cited in forecasting. Citation counts move, so any figure of that kind is a snapshot. Treat the Google Scholar links as the live source and these as approximations at the time of writing.

Photography

Where a photograph is somebody else’s work, it is credited here rather than under the image, so that the credit sits with the rest of the provenance.

Israeli Dairy School, Summer at the Israeli Dairy School
The hero photograph on the heat-stress page: cows resting in an open-sided barn while overhead misters and fans run a cooling cycle. Used with permission. dairyschool.co.il
Clearance pending
Seven photographs are published ahead of a final credit line, and we would rather say so than leave the impression that everything here is cleared. They are: the satellite view of a cyanobacterial bloom on the water page; a Varroa destructor mite; a dairy udder at dry-off; a desert barn; a Pacific map of exclusive economic zones and the high seas between them; kiwifruit showing fungal growth; and the wellboat Ronja King. Five of them now carry both the thumbnail and the hero on their page. Each is either a public-domain government image or is being licensed or credited; where clearance is refused the image comes down. If you hold rights in any of them, tell us and we will credit or remove it.
Everything else
Remaining photography across the site is illustrative and is being licensed.