A decade of quarterly readings

The CSO’s Table 1 records the climb without a break. The data centre share of metered electricity was 5 percent in 2015, 6 percent in 2016, 7 percent in 2017, 8 percent in 2018 and 9 percent in 2019. It reached 11 percent in 2020, 14 percent in 2021, 18 percent in 2022 and 21 percent in 2023, before the 22 percent recorded for 2024. In volume terms the series runs from 1,238 gigawatt hours to 6,969 gigawatt hours, an increase of about five and a half times, calculated from the CSO table.

The quarterly figures make the slope sharper still. Data centres used 290 gigawatt hours in the first quarter of 2015 and 1,829 gigawatt hours in the final quarter of 2024, an increase of 531 percent across the run of quarters.

In the most recent year of the series, data centre consumption rose 10 percent, from 6,335 gigawatt hours in 2023 to 6,969 gigawatt hours in 2024. All other users — residential customers and other business customers combined — rose 3 percent over the same period. Total metered electricity consumption in Ireland grew 30 percent between 2015 and 2024, from 24,600 gigawatt hours to 31,903 gigawatt hours. Arithmetic on the same table shows where that growth came from: consumption by all other customers rose about 7 percent across the decade, while data centres accounted for roughly 5,731 gigawatt hours of the 7,303 gigawatt hour increase, or nearly four-fifths of it.

A broader category in the release, large energy users — which typically includes larger data centres — reached 31 percent of total metered consumption in 2024, at 9,897 gigawatt hours, 9 percent up on the previous year. The companion Metered Electricity Consumption release, published the same day, puts total electricity consumption up 4.3 percent in 2024 against 2023, and records a median residential consumption of 3,246 kilowatt hours across 2.2 million residential meters, up from 3,174 kilowatt hours in 2023.

Finding a data centre among 2.5 million meters

None of this comes from a ready-made list, because data centres are not separately classified in the source data. The CSO’s background notes state plainly that “there is no agreed definition of a data centre”, and that data centres were not readily identifiable in the meter data. The raw material is meter data supplied by ESB Networks, covering all meters connected to the mains network. To build the series, statisticians examined around 2.5 million meters to identify the meter point reference numbers primarily used for data centre activity.

The identification runs on thresholds and cross-checks. The CSO matched names and aliases of known data centres consuming more than 0.5 gigawatt hours; looked at customers located in specific business parks above the same threshold; and reviewed all meters consuming more than 1 gigawatt hour annually. Published reports, the CSO Business Register and internet searches were used alongside those steps. The notes add two cautions worth carrying forward: a small number of data centres accounted for most of the consumption, and figures for all years are subject to revision, partly because a new data centre may use little electricity at first and sit below the search thresholds.

Commenting on the release, Dr Grzegorz Głaczyński, a statistician in the CSO’s Climate and Energy Division, said the data centre share of metered electricity consumption “grew substantially” from 5 percent in 2015 to 22 percent in 2024.

What the regulator did about it

The same growth has reshaped connection policy. On 23 November 2021, the Commission for Regulation of Utilities issued CRU/21/124, a direction to the system operators on processing data centre grid connections. The decision identified an “evolving, significant risk to electricity security of supply in Ireland”, and named the large increase in demand from the growth of the data centre industry as a significant contributory factor. The CRU directed EirGrid, as transmission system operator, and ESB Networks, as distribution system operator, to assess data centre connection applications against set criteria. It chose those connection measures over a moratorium, a step the regulator later described as one that would have been disproportionate.

That direction has since been superseded for new applications. On 12 December 2025 the CRU published a connection policy decision for large energy users requiring new data centres to provide new renewable and dispatchable electricity generation. Setting out its reasoning, the regulator noted that Ireland’s total electricity demand “has grown by 30% over the past 10 years, with data centre demand having greatly contributed to this level of growth” — in line with the 30 percent growth in total metered consumption between 2015 and 2024 in the CSO table.

The long count, then, rests on ESB Networks meter data, sifted by the CSO and tallied quarter by quarter for a decade. In 2015, data centres took 5 percent of Ireland’s metered electricity, and every other customer on the network took the remaining 95 percent. In 2024, data centres took 22 percent — 6,969 gigawatt hours — while consumption by all other users rose about 7 percent across the same ten years.

You’d expect the newer device to last longer. It hasn’t. Or rather, they haven’t. As the bulb’s longtime keeper, retired deputy fire chief Tom Bramell, put it in a report republished by FireRescue1, “The lightbulb isn’t burning out, but the webcams are.”

The bulb is the Centennial Light, and it hangs in Fire Station No. 6 in Livermore, California. Donated to the local fire department and first lit in 1901, it has burned almost without a break ever since. Its history includes interruptions for power problems, a renovation and a few moves. Guinness World Records lists it as the longest-burning bulb on the planet.

The story has a recent milestone. The Centennial Light Bulb Committee’s official website records a 125th-anniversary celebration on June 6, 2026. That followed its million-hour celebration in June 2015. The anniversary is another marker of its extraordinary longevity, although its history includes periods without power.

So why hasn’t it burned out? There is no single, conclusively established explanation for this particular bulb’s survival. Its construction and operating conditions offer clues, but examining an irreplaceable working object has obvious limits.

It is a hand-blown, carbon-filament bulb made by the Shelby Electric Company of Ohio. That matters because its filament behaves differently from the tungsten filaments used in later incandescent lamps.

Physicist and electro-optics specialist Martin Kykta explored those differences in a 2021 technical analysis hosted on the bulb’s website. He drew on measurements of another Shelby bulb and calculations to investigate the Centennial Light’s behaviour. He suggested that the strength and thickness of its carbon filament helped it survive as material was gradually lost. These are explanations based partly on a substitute bulb and modelling, rather than a destructive examination of the Centennial Light itself.

Its operating history is unusual, too. The committee’s facts page describes it being left on as a nightlight over the fire trucks. It records roughly a week without power during a 1937 renovation, a move to its present station in 1976, and other interruptions. For most of its life, however, it has simply been left glowing.

The bulb is commonly described as originally rated at 60 watts. Popular Science reports that it now draws about 4 watts, a figure also used in Kykta’s analysis. Watts measure electrical power, not the amount of visible light produced. Its faint glow should not be confused with the brightness of a modern 4-watt LED.

Kykta explains the underlying trade-off: a hotter incandescent filament produces more light, but loses material faster through evaporation. Lower-temperature operation can extend its life while reducing the light produced for each watt consumed.

He calculated that the Centennial Light’s filament operates at a comparatively low temperature and suggested that it might last another century. That is a possibility raised by his analysis, not a forecast anyone can guarantee.

On May 20, 2013, the bulb went dark. The committee’s webcam page records an interruption of nine hours and 45 minutes. The failure was in its power supply, rather than the filament. After the faulty backup system was bypassed, the bulb glowed again.

What the bulb is, and isn’t, evidence of

It is tempting to hold the Centennial Light up as proof that manufacturers once knew how to make everlasting bulbs and then stopped. Its survival alone cannot establish that. One dim antique is not a controlled comparison with generations of brighter lamps designed for everyday use.

The Phoebus cartel, however, was real. On 23 December 1924, representatives of major lighting businesses met in Geneva and founded it. They included Osram, Philips and interests associated with General Electric, which participated through overseas subsidiaries.

As professor of media studies Markus Krajewski documented for IEEE Spectrum, the cartel established a 1,000-hour standard for household bulbs. His account draws on company archives describing efforts to shorten lamp life, testing arrangements and fines for products that fell outside the agreed lifespan.

This is documented commercial history, rather than something inferred from the Centennial Light and The cartel’s history does not make the Centennial Light proof that a bright, efficient household bulb could routinely last 125 years.

A survivor under unusual conditions

What Livermore has is an exceptional object with an unusually well-documented history. Its carbon filament, low-temperature operation and long service as a dim nightlight all belong in the explanation. How much each factor contributed remains harder to establish.

The consequence, if the analysis holds, is not merely academic.

What the traditional model said

The image the beekeeping literature has favoured, since the earliest twentieth-century texts, is one of a warm, insulating layer of bees on the outside of the cluster protecting a warmer inner core. The outer bees, on this view, function like the outer wool of a blanket around a sleeping body: their bodies and the trapped air between them slow the escape of heat from the core, so the core can be maintained at or above 18 degrees Celsius even when the ambient air around the hive is 40 degrees below zero. A study led by Fabrice Requier’s group at the Université Paris-Saclay, published in Ecological Indicators in December 2024 and based on 1,083 temperature sensors distributed across 31 hives in France, Germany and Greece, independently documents the temperature architecture Mitchell is arguing about: bees form a cluster to hold the nest above the 10 degrees Celsius chill-coma threshold, with cluster centres averaging 27 to 35 degrees and surface temperatures decreasing to between 18 and 29 degrees. The very tight compaction of the mantle bees was traditionally read as evidence that this insulating behaviour was working, and the thin wooden walls of standard hives were treated as acceptable because the bees were seen as doing the thermal work themselves. Two adjacent practices grew out of this reading. In the 1930s, thin-walled hives were placed in climates that regularly reach thirty degrees below zero, on the assumption that the mantle would compensate. In the late 1960s, Canadian commercial beekeepers began keeping colonies in cold storage at around four degrees Celsius through the winter, again on the theory that clustering was benign. Since the 2020s, and increasingly in Idaho, Washington and Southern California, similar refrigeration practices have been used in summer to force bees into a brood-break state that makes chemical treatment of Varroa mites easier.

What Mitchell actually did

Mitchell’s approach was to apply the standard engineering definitions of insulation and heat sink to the cluster geometry and see which the cluster actually satisfies. On the accumulated evidence of a hundred years of experimental data from previous investigators, he built a model of a spherical cluster inside a rectangular hive inside a bare cold landscape with a radiating sky, using the effective medium theory to derive the thermal conductivity of a mixture of bee bodies, hair and air across a full range of packing densities. He then tested the mantle against four definitions of insulation: whether adding more mantle reduces heat loss, whether the effective conductivity of the mantle decreases as clustering progresses, whether the mantle has a lower conductivity than the cluster core, and whether the mantle’s thermal resistance (its R-value) increases as clustering progresses. On all four definitions, the mantle failed. In fact, the paper found that when a colony transitions from its pre-cluster distribution to a fully dense cluster, the effective thermal conductivity approximately doubles and the mantle’s R-value can decrease by more than a factor of eleven. The mantle is not adding insulation as it compresses. It is losing it.

The mechanism is the compression itself

The reason the effect operates in the opposite direction from the traditional intuition is a specific property of porous insulating materials, which behave counter-intuitively when compressed. According to Mitchell’s own summary of his findings, written for The Conversation and republished by ScienceAlert, a down jacket is the illustrative case. Down insulates because of the small pockets of still air trapped between the feathers. Compress the down and those pockets of air disappear, the feathers touch one another more directly, the thermal conductivity of the compressed feather mass rises to something closer to that of a solid leather jacket, and the insulating value collapses. A cluster of honeybees under cold stress does something structurally similar. As the outside air temperature falls, the bees on the outer surface of the mantle chill down toward 10 degrees Celsius. Below this temperature, individual mantle bees die and fall off the cluster. To avoid falling below 10 degrees Celsius, the surviving mantle bees crowd closer to the bees behind them that are still warm. That compression closes the air gaps between the mantle bees, raises the effective thermal conductivity of the mantle, and increases the rate at which heat flows from the core out through the mantle to the outside, which is precisely what a heat sink does. What Mitchell’s paper describes, in the language of engineering rather than beekeeping, is compressed down turning into leather.

What this changes for hive design

On the current best interpretation of Mitchell’s numbers, the consequence for practical beekeeping is that standard hive walls are contributing to the problem rather than being neutral. According to the University of Leeds’ announcement of the paper, earlier work by the same author established that most manufactured hives lose heat at roughly seven times the rate that natural tree hollows do. Manufactured hive walls are about 19 mm of thin wood, and the tree cavities in which Apis mellifera evolved to overwinter have walls of the order of 150 mm. Mitchell’s calculation is that hive walls need to provide substantially more insulation than 19 mm of wood, on the order of 30 mm of polystyrene or its equivalent, to bring the hive’s thermal performance close enough to the tree cavity for a colony to overwinter without being forced into a persistent stress cluster. That change would remove the compression mechanism that produces the “wrapping in leather” outcome in the first place, because the bees would not need to compress themselves against the hive wall to hold 10 degrees Celsius at the mantle surface. Mitchell has framed the argument in explicitly welfare terms. The traditional view of clustering as benign has, on his analysis, licensed a practice of housing bees in thermal conditions they would not choose themselves, and of using refrigeration to force them into a stress behaviour that requires them to burn through as much as 60 kg of honey in the most adverse conditions. Whether the practice should continue is, on Mitchell’s reading, an ethics question as much as an entomological one.

The wider point his paper makes, sitting underneath the numbers, is that the hive itself is best understood as part of the honeybee’s extended phenotype rather than as an inert container the bees happen to inhabit. Thin-walled boxes are not neutral.

When pumping stops, groundwater can fill the tunnels. Underground rock generally gets warmer with depth, and the water absorbs that natural heat. A mine that stopped producing coal, therefore, can become a useful reservoir of low-temperature heat.

That is what happened in Heerlen, in the Netherlands’ South Limburg region, although  decades after the mine closed. According to operator Mijnwater’s history of the project, Heerlen’s last coal mine closed in 1974. Pumping from the Limburg mines stopped in 1994, allowing the workings to fill gradually with groundwater.

The mine water project officially began in March 2004, exploring whether that underground resource could help heat and cool the city above.

How the Heerlen system works

The first plant went live in October 2008. A project case study hosted by Renewables Networking describes it as the world’s first mine water geothermal plant.

The early setup is easy to picture: water at two temperatures drawn from different depths. The EU-funded STORM project’s description records two deep wells in the north of the city supplying water at about 28 degrees Celsius. Two shallower wells in the south supplied water closer to 16 degrees.

Twenty-eight degrees is lukewarm. The system uses heat pumps to raise the temperature to the level buildings need for heating and hot water. That means using the mine water as a source of heat, rather than simply sending lukewarm water into conventional radiators. The cooler source can help remove heat from buildings that need cooling.

The individual components are familiar. Herman Eijdems, identified as Mijnwater’s innovation manager in a Construction21 article, put it plainly: “The funny thing is that we don’t actually use any exciting innovations at all.” The novelty lies in how heat pumps, pipes and storage work together.

From a heat supplier to a two-way grid

The original system supplied heating and cooling to a small group of customers through a largely one-way network. Expanding it meant finding a better way to use the available energy. The Renewables Networking case study also describes years of financial losses associated with design and commissioning problems.

Mijnwater says it began developing exchanges between groups of buildings from 2012.

Instead of treating the mine only as a source to draw from, the new approach also used it to store thermal energy. A building that needs cooling releases unwanted heat, which another building may be able to use. When supply and demand do not match, underground storage helps bridge the gap.

Mijnwater’s own explanation of the network gives a useful example: heat removed from a data centre can become a source for heating homes. It also makes clear that the mine workings still serve as both energy sources and thermal buffers. Heat exchange between customers does not make the mine redundant.

Eijdems described the approach as recombining established equipment: “Heat pumps, pipes and storage basements are techniques that have been around for 50 years or more,” he said. “But we have linked the various components together in a new way.”

What the results show

A shorter case study hosted by Renewables Networking reports a 65% reduction in carbon dioxide emissions associated with heating and cooling connected buildings.

Historical accounts also give snapshots of its size. Construction21 described the system as supplying 200,000 square metres of building space in 2018.

Still operating, and drilling deeper in 2026

The system remains active and is expanding. In August 2026, ThinkGeoEnergy reported that Mijnwater had six wells, three warm and three cold, and was developing a fourth warm-water source. Mijnwater’s own announcement of the drilling work identified the target as an old mine gallery about 634 metres underground, with expected water temperatures of roughly 28 degrees Celsius.

Where the idea can travel, and what it needs

The tempting conclusion is that any old mining town could do this. Eijdems has suggested that the approach “can also be used in other former mining areas to heat and cool buildings.” That is a possibility worth investigating, rather than a promise that every flooded mine will make a viable heating system.

Heerlen’s experience involved much more than finding warm water. It required wells, a pipe network, suitable buildings, funding and years of adjustments.

There is also a practical issue at the customer’s end. Mijnwater says most of its connected buildings are new construction. Existing buildings connected to the network have been upgraded to Dutch energy labels A or B to make them suitable for its heating and cooling service. Having a useful source underground does not remove the need to prepare the buildings above it.

What transfers is therefore an approach to using local energy, rather than a ready-made blueprint. The flooded mine provides a source and a place to store heat. The buildings provide demand and, when they need cooling, heat that can be recovered. The network brings those resources together.

Heerlen found a new use for the infrastructure left by an abandoned industry. Its achievement was not simply noticing that the water was warm. It was developing a system that could make that modest warmth useful, share energy between buildings and keep adapting as demand grew.

The fish itself had not changed. What had changed was the state’s operating theory about what would actually reduce their numbers.

How the invasion actually happened

The animals were brought into the United States in the 1970s to help clean aquaculture ponds in Arkansas and other Southern states, where their filter-feeding on plankton and algae was seen as a low-cost biological solution to a technical problem. Flooding and accidental releases through that decade let them escape into the Mississippi River system, where they had no natural predators, no historical parasites, and no established competitor for the ecological niche they occupied. According to the U.S. Geological Survey’s Upper Midwest Environmental Sciences Center in La Crosse, Wisconsin, which now runs population-level modelling of the two most numerous species, bighead and silver carp had spread into the Illinois River by the 1980s and now make up as much as 90 per cent of the fish biomass in some sections of the invaded waterways. In doing so they starve out the native filter feeders, most of which sit at the base of the food web supporting native game fish, mussels and diving birds. Silver carp in particular have the unusual habit of leaping several metres out of the water when startled by boat engines, and have caused a growing number of injuries to river users through the region. The fish that Illinois is now trying to sell as copi were, until 2022, one of the largest ongoing ecological problems on the American mainland.

Why physical removal has not been enough

The problem the rebrand was designed to address is not a problem of biology. It is a problem of scale. On the accumulated evidence from two decades of federal, state and interstate control programmes, mechanical removal by contract fishing crews, electric barriers on the Chicago Sanitary and Ship Canal, sonic deterrents and lock-and-dam modifications has managed to slow the range expansion of the fish but has not been able to reduce their standing biomass in the invaded rivers to anything approaching pre-invasion levels. The Illinois DNR estimates the harvestable annual biomass of copi in the Illinois River alone at 20 to 50 million pounds, with hundreds of millions of pounds more available across the wider Mississippi system. The stakes downstream of the electric barriers are large in dollar terms as well as ecological ones: the Great Lakes support a commercial and recreational fishery estimated at seven billion dollars a year and a tourism industry estimated at sixteen billion, both of which are at risk if the fish reach Lake Michigan. Removing tens of millions of pounds of copi year after year at a rate faster than the population can replace itself is only economically viable if there is a market willing to buy them once they are on land.

In the absence of that market, government-funded contract fishing is essentially the only way to remove them, and that is expensive, slow, and politically fragile.

What Span actually did, and the historical parallels

The reason Illinois needed a branding agency and two years of consumer research rather than a straightforward marketing campaign is that “carp” is a name American consumers have long associated with muddy-tasting bottom-feeders, despite copi being a top-feeder that eats plankton and has clean, mild flesh. The rebrand also had to sidestep the label “Asian carp,” which the American Society of Ichthyologists and Herpetologists and the American Fisheries Society had never formally adopted, and which Asian-American legislators in the Midwest had raised objections to on the grounds that pairing the word Asian with an invasive species being violently killed on camera was doing damage to their communities. Span’s process, involving surveys of Illinois residents, produced “copi” as a name that tested well on brevity, pronounceability, and lack of prior negative association. According to the consumer-facing brand site the IDNR launched alongside the announcement, which the Monterey Bay Aquarium’s Seafood Watch programme has listed as a recommended sustainable choice under its silver, bighead and grass carp entries, copi is now being pitched to shoppers as a wild-caught, mild-flavoured, mercury-safe white fish with heart-healthy omega-3 content. The historical precedents Illinois cited when explaining the strategy included several fish whose commercial fortunes had turned on their names. Slimehead became orange roughy in the 1970s and grew into a substantial international fishery. Patagonian toothfish became Chilean sea bass around 1994. Mud crab became peekytoe crab. Illinois has applied to the U.S. Food and Drug Administration to have copi formally recognised as a market name and has moved to register the word as a trademark, so that industry groups can develop their own quality control standards around it.

What the record actually shows so far

Progress since the launch has been real but is still small against the size of the underlying problem. Distribution deals were signed at launch with Gordon Food Service, one of the largest foodservice distributors in North America, and copi has since appeared on Chicago menus and in fresh-fish sections of Midwestern retailers. The critique that environmental groups including the Alliance for the Great Lakes have raised in response is essentially that while a commercial fishery is welcome, it cannot on its own be relied upon to prevent copi from reaching Lake Michigan, and that the more consequential piece of infrastructure remains the Brandon Road Lock and Dam project on the Des Plaines River south-west of Chicago, which is being built to a design combining electric barriers, acoustic deterrents, air-bubble curtains and a flushing lock intended to keep the fish out of the Great Lakes drainage. On the strongest current reading, the copi rebrand is neither a solution on its own nor a distraction from the physical infrastructure. It is one of the two things the state is doing at the same time, and its function within the wider strategy is to make the ongoing removal of the fish a business proposition rather than a permanent public expenditure.

In late 1991, a camera clamped to a retort stand (the metal pole and clamp that holds flasks and test tubes in school chemistry labs) was aimed at a coffee machine in a corridor of the University of Cambridge Computer Laboratory. Its goal was modest. Nobody wanted to climb the stairs for nothing.

Quentin Stafford-Fraser, one of the researchers behind it, set out the problem in a short biography of the Trojan Room coffee pot first published in 1995. About 15 people shared a single machine outside a part of the lab known as the Trojan Room. Colleagues elsewhere in the building had to tackle several flights of stairs to reach it, and often found the room’s all-night regulars had got there first.

Turning a spare frame-grabber into a coffee monitor

What does a research group do with a frame-grabber gathering dust? A frame-grabber is a circuit board that pulls still pictures out of a video signal, and one was sitting unused on a test computer in the room. The team ran the camera cable under the floor to it. Paul Jardetzky wrote a server program to capture the images, and Stafford-Fraser wrote a client program that anyone in the group could run, which put a thumbnail of the pot in the corner of their screen. In plain terms, the server did the watching and the client did the showing.

Pictures refreshed about three times a minute, which suited a pot that filled slowly. They were also greyscale, a shortcoming Stafford-Fraser waved away because, in his words, “so was the coffee.”

Assembling the rig took a day or so. He later rated it more useful than anything else he produced during that stretch of networking research, and a much easier sell at dinner parties than the day job.

Putting the pot on the web in 1993

Early web browsers could only display text. That changed in 1993, when a program called Mosaic arrived with the ability to show images, as Stafford-Fraser recalled in a 2001 essay for Communications of the ACM, a magazine published by the Association for Computing Machinery.

The team then twigged something that sounds obvious today: a web server doesn’t have to send back the same picture every time it is asked. Daniel Gordon adapted the coffee server to do exactly that, so each visit to the page fetched a fresh snapshot. In November 1993, with help from Martyn Johnson, he put it online, and the coffee was open to inspection from anywhere with an internet connection.

The lab’s coffee pot page, still hosted on the university’s servers, carries a couple of housekeeping notes. A small lamp was usually left pointing at the pot so night-time viewers could see it, and anyone hoping to drink the stuff needed to join the coffee club first, travel costs not included.

Traffic snowballed. By around 1996 the image had passed a million hits, and one journalist pointed out that the pot had drawn more “visitors” than King’s College Chapel, making it, on paper, the top tourist attraction in East Anglia. Two million followed in 1998, per a timeline kept by Stafford-Fraser.

Selling a broken coffee machine on eBay

“It leaks water and we’ve cut off the mains plug.”

So ran the eBay listing in August 2001, as reported by The Register. The lab was moving to new premises, and the machine on offer, a Krups ProAroma owned by the coffee club rather than the university, had stopped working. Several pots had taken a turn on camera over the years, and this one had served longest, starring from October 1997 until March 2001. Club members hoped it might fetch a few hundred pounds, enough for a decent espresso machine.

Spiegel Online, the website of German news magazine Der Spiegel, won with a bid of £3,350. Tim Hill, who organised the sale, joked that the priciest coffee maker he could now find on the web was the one they had just sold. Stafford-Fraser’s timeline records that the money was meant for coffee facilities in the new building.

The last frame and a second life in Germany

At 10.54 on 22 August 2001, three people reached for the power switch on the Acorn Archimedes computer that had been capturing the pictures. The camera’s final image shows their fingers pressing it, closing a web run of seven years and nine months.

In the ACM essay, Stafford-Fraser explained why it had to go. The ageing hardware was the last surviving piece of the lab’s home-built network and increasingly hard to keep alive, and the Trojan Room itself would disappear once the lab moved to West Cambridge. News of the shutdown brought the pot its biggest burst of publicity yet, including front-page coverage in the London Times and the Washington Post. His conclusion was that nobody makes headlines by putting something online any more. Switching it off, on the other hand, works a treat.

The machine went to Hamburg along with its camera, a coffee grinder and other memorabilia, and Krups got it brewing again, as recounted by the Heinz Nixdorf MuseumsForum, a computer museum in Paderborn. The pot has been on permanent loan there since 2016, sitting in an eleven-metre wall display on the history of the internet with a live camera trained on it once more.

For an actual cup, the museum’s blog directs visitors to the café on the ground floor.

Taste buds are the small clusters of cells, arranged like the segments of an orange, tucked into the bumps and grooves of the tongue. An adult carries between 2,000 and 4,000 of them, per the InformedHealth.org guide to taste from Germany’s Institute for Quality and Efficiency in Health Care. Each bud holds 10 to 50 sensory cells, and those are renewed about once a week. Extra taste cells turn up at the back of the throat and even in the upper part of the food pipe.

None of this comes with a memo.

How the ten-day figure was first measured

In 1965, two biologists at Florida State University set out to time the shifts. They injected rats with thymidine, one of the raw ingredients a cell uses to copy its DNA before dividing, tagged with a faint radioactive label. Cells that split shortly afterwards soaked up the tag and passed it to their daughters, and it showed up as dark specks once thin slices of tongue were coated in photographic emulsion and left to develop.

Writing in the Journal of Cell Biology, the pair reported that new cells were born in the lining around each bud, then drifted inward toward its centre. An average cell lasted about 250 hours, give or take 50, which rounds to roughly ten days. Some burned out much sooner and others stuck around far longer. All of it was measured in rats.

Some taste cells outlast others

Does every cell in a bud clock off on the same schedule? A University of Miami team put that to the test in mice, in a paper published in PLOS One in 2013, sorting the cells by type before starting the clock.

Type II cells, which carry the receptors for sweet, bitter and savoury (umami) flavours, had a half-life of eight days. Half-life means the time it takes for half of a batch to disappear, so within just over a week, half the receptor cells from any given intake were gone. Type III cells, the most nerve-like members of the bud, were in no hurry, with a half-life of 22 days.

That is one study, looking at one patch at the rear of the tongue, so the exact figures belong to that setup. It does fit the spread the Florida State pair spotted decades earlier, and it recasts ten days as an average smoothed across very different careers.

Where the replacement cells come from

“Taste tissue regenerates continuously throughout the life span in mammals.” That line opens the plain-language summary of a 2014 paper in PNAS from the Monell Chemical Senses Center, which went looking for the source of all those new recruits.

Its authors isolated single stem cells from mouse taste tissue, identifying them by a surface protein called Lgr5 (or its close relative Lgr6), which marks out adult stem cells in several tissues. Grown in a dish, a single one of these cells could multiply into a small three-dimensional clump known as an organoid. Those clumps contained all the mature taste cell types, the cells reacted to taste chemicals, and they managed all of it with no nerve attached.

That knack for renewal is rare among the body’s sensing cells. As a taste biology lab at the University of Colorado Anschutz Medical Campus points out, the light-sensing rods and cones of the eye and the hair cells of the inner ear lack it. Wear those out and no replacement order is coming.

What happens when the renewal stops

Basal cell carcinoma, a common skin cancer, can be treated with drugs that block a growth signal called Hedgehog, a chemical message that helps developing cells decide what to become. Many patients on these drugs report severe taste disturbances, which prompted University of Michigan researchers to give one of them, sonidegib, to mice and track the fallout, as documented in PNAS in 2017.

The precursor cells that feed each bud were thrown off their usual pattern of dividing and maturing, and the buds disappeared. Nerve responses to taste were wiped out, while responses to touch and cold carried on. Once the drug was withdrawn, the buds grew back and the taste signals returned.

It is a mouse result, though it lines up neatly with reports of patients’ taste recovering after they finish treatment.

Taste can feel like one of the most settled things about a person, as fixed as an accent. A decades-long grudge against coriander has been maintained by taste cells replaced hundreds of times over since the feud began.

Whenever a dog stopped to relieve itself off the lead, in open ground away from roads, power lines and steel structures, its owner took a bearing along the animal’s spine. The tally, published in a 2013 paper in Frontiers in Zoology, covered 70 dogs from 37 breeds, with 1,893 defecations logged from 55 of them and 5,582 urinations from 59. Leading the project were Vlastimil Hart and Hynek Burda of the Czech University of Life Sciences Prague and the University of Duisburg-Essen, whose group had earlier claimed, in a 2008 study in PNAS, that cattle and deer tend to line up north-south.

The dog work went on to win the 2014 Ig Nobel Prize for biology, per APS News, an award for research that makes people laugh before it makes them think.

Sorting the walks by magnetic conditions

The first pass through the data was underwhelming. Pooled together, the headings showed a weak, heavily scattered trend that fell apart once each dog was averaged separately. Only after sampling ended did the team sort every walk by what the planet’s field was doing at that moment, using readings from a geomagnetic observatory near Munich.

When magnetic declination held completely steady, the dogs had a highly significant preference for a north-south axis. As the rate of change crept upward their headings scattered, and with declination drifting faster than a fiftieth of a degree an hour, they were random. The researchers also found the dogs avoided facing east-west, and because the preference ran along an axis, pointing south counted the same as pointing north.

Shifts in the field’s direction predicted the switch better than shifts in its strength.

How magnetic north shifts

Why would a compass direction move at all? Magnetic north and true north sit in different places, and the angle between them, known as declination, varies with both location and time, as the US National Centers for Environmental Information explains. Over decades it drifts. Within a day it wobbles too, and Hart’s team describes westward nudges in the morning, eastward ones in the afternoon and erratic jolts during magnetic storms.

These movements are tiny. In one worked example, declination moved 10 arcminutes, a sixth of a degree, over four hours, about twice the rate at which the dogs’ headings turned random. Nobody holding a lead would notice.

Sunlight, which switches sides of the sky between breakfast and dinner, is an obvious rival explanation. The authors countered that calm-field alignment held at every hour, and that most walks probably happened under cloud.

Later dog studies disagree

In city parks in Eilat, at Israel’s southern tip, high school students checked the headings of squatting dogs with phone compasses, sometimes after tucking a small bar magnet into the grass, as reported by Science in Poland. Their study, published in the Journal of Veterinary Behavior in 2020, found north-south alignment that shifted when the magnet was nearby.

A year later, a University of St Andrews team repeated those methods in five Lyon dog parks and detected no magnetic alignment. Those dogs did line up with each park’s layout, and the researchers warned that volunteer observers can unconsciously bias what they record.

Back in 2014, Columbia University statistician Andrew Gelman flagged a multiple-comparisons problem in the original paper: slice one dataset enough ways and a fluke starts to look meaningful. He wanted a preregistered replication, with the test fixed before any data came in. Its authors maintained that the late sorting made their work genuinely blind, since nobody had expected declination to matter. For now the canine evidence is contested, with two supporting studies, one failed replication and no consensus.

Hunting dogs and the compass run

“Compass run” is the name the Czech group later gave to a habit of hunting dogs. Writing in eLife in 2020, Kateřina Benediktová and colleagues fitted 27 hunting dogs with GPS collars and action cameras and followed more than 600 forest excursions. Dogs returning to their owner by a fresh route usually began with a dash of about 20 metres along the north-south axis, whichever way home lay, and those runs made the trip back more efficient.

Hart and Burda appear on both author lists, so this extends the group’s own thinking and can’t count as independent confirmation. The proposed explanation is that the run brings a mental map into line with a magnetic sense of direction, an idea the 2013 paper had already floated, likening a squatting dog to a hiker turning a map towards north.

That paper also held that such calibration only makes sense against a stable reference, which would explain why alignment vanished once declination started sliding. By its estimate, the field is calm for only about a fifth of daylight hours. On most walks, then, a dog squatting east-west may simply have judged the map not worth reading.

I’ve seen this incident get flattened into two competing shorthand versions online. One treats it as a minor equipment glitch, the kind of thing that happens once and gets quietly fixed. The other treats it as proof the whole technology is unready and dangerous. Neither version survives much contact with what actually happened on the ground, and the actual record turns out to be more serious than “minor glitch” while stopping well short of “the technology is a disaster.”

The claim: a broken blade is a rounding error

The dismissive version of this story usually goes something like this: offshore wind machinery breaks sometimes, just like any large piece of industrial equipment, and a single incident at a single project a few years into build-out doesn’t tell you much of anything. On its own, that reasoning isn’t unreasonable. Industrial hardware fails. A car recall or a bridge inspection finding doesn’t indict an entire category of infrastructure by itself, and it would be unfair to treat one manufacturing defect as proof that offshore wind can’t work.

That logic only holds up if the failure actually stayed small, though, and here it didn’t. The federal government shut down an entire commercial power project for six months, not just the one damaged turbine. Six public beaches closed during the busiest tourist weeks of the year. A settlement in the eight figures followed a year later. Those aren’t the marks of a rounding error, and treating the incident as one requires skipping past most of what actually happened.

What fifteen miles offshore turned into on the beach

The blade broke roughly 15 miles from shore, but the debris didn’t stay offshore. Fragments began washing onto Nantucket’s south shore within days of the failure, according to local reporting from Nantucket Current. Harbormaster Sheila Lucey told residents at the time, “The water is closed to swimming on all south shore beaches, due to large floating debris and sharp fiberglass shards.” A total of six public beaches on the island’s south shore were closed to swimming that summer, according to CBS News Boston, right as Nantucket was heading into the weeks it depends on most for tourism revenue.

A Vineyard Wind spokesman said the blade had snapped about 20 meters from its base. GE Vernova, the company that manufactured it, later attributed the failure to an error at its own factory, according to GBH News. That’s a specific, named cause, on the record, and it’s part of why regulators treated what came next as seriously as they did.

Why regulators shut the whole project down

The federal Bureau of Safety and Environmental Enforcement suspended construction and power generation across the entire Vineyard Wind 1 project, not just the single damaged turbine, within days of the failure. That suspension held for six months and only lifted in mid-January 2025, and the agency’s own account of that order lists real conditions attached: the operator had to commit to removing blades installed before the failure from every turbine already in the water, complete an environmental review of the cleanup, and prove that replacement blades met the original design specifications. Regulators were blunt about one piece of it. As the agency put it, “BSEE will not allow installation of new blades manufactured at this facility.”

A shutdown lasting half a year, applied to an entire operating power project rather than a single component, carries real financial and operational weight. Regulators reserve a response like that for failures they consider genuinely unresolved.

What the $10.5 million settlement actually covers

A year after the blade broke, in July 2025, Nantucket and GE Vernova announced a $10.5 million settlement. The money is earmarked to compensate local businesses for lost revenue during the disrupted summer, evaluated through a claims process run by an independent administrator, according to town officials. Select Board Chair Brooke Mohr called it a genuine win for the town: “We are confident that this is a really positive outcome for Nantucket.”

The agreement is narrower than it might sound, though. According to town officials, it does not limit Nantucket’s options if a similar failure happens again at the same project, meaning the settlement resolves what already happened without signing away the town’s ability to respond to whatever comes next. That distinction matters for reading this fairly. A settlement resolves a specific, already-documented set of losses between one town and one manufacturer. Whether the broader technology is safe, or worth the tradeoff long term, is a separate question this payout doesn’t answer on its own.

The trade Nantucket actually made

I’m not an energy engineer or a regulator, and I have no professional stake in whether offshore wind succeeds or stalls in New England. What I do have is a strong personal opinion about how to weigh a hard year against a long-term arrangement, and it has nothing to do with turbines specifically. I think happiness, or at least not being made miserable by your own circumstances, is close to non-negotiable. If something in your life is making you unhappy, the honest move is usually to fix it. The one exception I actually believe in is when the discomfort is temporary and in service of a goal you genuinely want, in which case it’s worth sitting with rather than escaping.

A town obviously isn’t a single person weighing a single decision, and Nantucket didn’t get to opt out of hosting this project the way you might turn down a job or end a relationship. But the same basic question still applies to how residents get to judge the arrangement going forward. Six months of federal limbo, closed beaches during the busiest weeks of the year, and a full year waiting on compensation is real, specific discomfort, not an abstraction on a spreadsheet somewhere.

Whether that discomfort was worth absorbing depends entirely on what the town actually gets back from hosting offshore wind over the long run, not on how convincingly either side tells the story. If the long-term arrangement genuinely pays off in local jobs, lease revenue, and years of functioning clean power, a rough year and a settlement check might turn out to be exactly the kind of temporary cost worth accepting. If it doesn’t pay off, the math changes, and no settlement number fixes that after the fact. Nantucket will find out which version it got long after this particular blade is forgotten.

A cheese plant making mozzarella and provolone was producing salty wastewater and paying to have it hauled away. Meanwhile, a county road crew was buying rock salt and losing some of it off the pavement before it could do much good.

That was the beginning of Polk County’s cheese-brine project, which entered full use in 2009. The arrangement later ended when its local dairy supplier closed. But the idea continued elsewhere: Green County, in southern Wisconsin, was still using cheese-production brine according to its 2024–2025 report.

Two waste problems that suggested one solution

The cheese brine in this story was a used saltwater solution from the cheesemaking process. It was already full of salt, which is exactly what road crews use to help melt ice.

Emil Norby, a Polk County highway employee, saw a possible connection between the dairy’s disposal headache and the county’s wasted road salt. As he told Modern Farmer, “I knew we had a lot of dairies looking to get rid of this stuff.” The brine came from F&A Dairy in nearby Dresser.

The method is called pre-wetting. Instead of spreading dry crystals, crews coat the rock salt in liquid to reduce the amount that bounces or is scattered off the road. According to a University of Wisconsin–Madison account of the project, two Wisconsin Department of Natural Resources employees helped the dairy and Polk County secure approval and develop application rates. The first season of full use was in 2009.

What Polk County reported saving

Outside the Beltway, summarizing contemporary coverage, reported that Polk County officials said they used 30 percent less road salt than usual in the first year. That fits the intended benefit of pre-wetting: if less salt is scattered away, less may be needed.

The money moved on both sides. Polk County saved $40,000 in its first season of full use in 2009. Modern Farmer also reported that F&A saved nearly $30,000 in brine-disposal costs. Chuck Engdahl, the dairy’s wastewater manager, summed it up as “Everybody wins.”

From a cost standpoint, it is easy to see the appeal. Whether the practice also reduces environmental harm depends on how much salt and other material ultimately reaches the surrounding water.

To give a sense of scale, NPR reported in January 2014 that Polk County had used at least 150,000 gallons of cheese brine over the preceding years. Norby said 38,000 gallons had been applied the previous winter alone.

Why Polk County stopped

The catch was the same thing that helped make the arrangement work: proximity.

A cheap byproduct can become expensive once hauling and storage are included. In February 2024, Wisconsin Public Radio reported that Polk and Burnett counties had stopped using cheese brine when their local cheese factory closed.

In the same interview, Green County highway commissioner Chris Narveson explained that the economics depend on how far crews must travel to collect the brine. Trucks, storage tanks, and suitable facilities also have to be available. The material’s low cost at the dairy is only part of the calculation.

Green County continued using it

Green County is a separate case. The Green County Water Quality Coalition’s 2024–2025 report, published through UW–Madison Extension, confirms that its highway department continued using brine from cheese production. The coalition credits the practice with reducing rock-salt applications and saving money.

Less rock salt still does not mean salt-free

Salt can pollute freshwater, so reducing its use matters. In a study announced by the U.S. Geological Survey in December 2014, researchers led by Steven Corsi examined chloride trends in urban streams and emphasized the need to “consider deicer management options that minimize the use of road salt while still maintaining safe conditions.” Across the study’s 30 monitoring sites on 19 streams, 29 percent of sites exceeded the EPA’s chronic aquatic-life criterion for chloride.

The scale of salt use helps explain the concern. Penn State Extension put annual U.S. road-salt use at roughly 20 million tons. Cheese brine does not eliminate the pollution problem: it contains dissolved salt too. Evaluating the environmental benefit therefore means counting the salt in both the brine and the rock salt, rather than looking only at how much dry salt a county buys.

Perhaps, the broader lesson is to look at what one operation pays to throw away and ask whether another nearby operation is paying to buy something that serves the same purpose.

What actually opened the door

The reason the golden jackal has walked halfway across Europe over the last four decades is not primarily climate, though climate helps, and it is not intentional reintroduction, because there hasn’t been any. It is the collapse and fragmentation of European grey wolf populations across the same window. Golden jackals are considerably smaller than wolves, they avoid areas where wolves are actively defending territories, and where wolves have historically been present in numbers the jackals have been kept out or kept scarce. When wolves were shot out of most of central and western Europe over the course of the nineteenth and twentieth centuries, the mid-sized carnivore niche they had held opened up, and a smaller, more dietarily flexible canid could move into that space without being killed by a resident pack on arrival. Ecologists call this pattern mesopredator release, and it has been documented in a large number of settings. The North American parallel is the coyote, which has expanded across most of the continent over the same period that grey wolves were being extirpated from it.

Europe got its predator back without anyone deciding to give it one. Yellowstone is the opposite case, and the video covers what happened there after 14 wolves were let out of crates in January 1995.

The German case in specific

The Böcker paper describes the German breeding site in some detail. It sits at the edge of a low mountain range in Baden-Württemberg, in a mosaic of bogs, reeds and open oak and pine forest, adjacent to two nature reserves and near the river Danube. The area contains red fox, badger, raccoon, roe deer, fallow deer and wild boar, all of which are the sort of prey and scavenging opportunities golden jackals make good use of. Critically, the area contains no wolves and no other large carnivores. Four camera traps set across roughly 20 hectares registered 90 confirmed jackal events between October 2021 and August 2022. Scat samples analysed with a trained detection dog and subsequent genetic identification at the Senckenberg Center for Wildlife Genetics established five distinct individuals, likely siblings, and a probable father assigned through DNA from saliva left on a scavenged sheep carcass. What makes the case biologically significant is the distance from the nearest previously documented breeding sites, which the paper lists as approximately 400 to 500 kilometres for the closest Italian reproduction, 525 kilometres for a Czech site, 605 kilometres for an Austrian site, and 965 kilometres for the Polish site at Kwidzyn that had held the northernmost reproduction record in Europe up to that point. Whatever animal or animals founded the German population had covered that distance on their own.

How far the dispersers have got

The German case is the westernmost confirmed reproduction. Individual dispersers, without confirmed breeding, have reached considerably further. A 2024 paper in Mammalian Biology by Ilpo Kojola, Heikki Henttonen and colleagues at the Natural Resources Institute Finland describes six verified records of golden jackals in Finland between 2018 and 2022. The first was a camera trap image in east-central Finland in August 2018. The most northerly was a male, weighing about ten kilograms and in good physical condition, that was inadvertently caught in a fox snare in Lapland in March 2022, about 120 kilometres north of the Arctic Circle. His origin is unknown, but the most likely source population for the Finnish records is Estonia, where a resident golden jackal population has been breeding since 2012 on the country’s western coast, and the most plausible route into Finland is the isthmus of dry land between lakes Ladoga and Onega, since the sprawl of St Petersburg blocks the more southerly crossing. In the Netherlands, an individual golden jackal was camera-trapped in the Veluwe Nature Reserve in 2016, the country’s first record, and a second was DNA-confirmed from a sheep predation event in the Ooijpolder in 2021. In Denmark, the species has been recorded sporadically since 2015, again with DNA-confirmed livestock predation. None of these populations are yet reproducing. All of them show the same pattern: single animals arriving from further east and south, hundreds of kilometres from any breeding group.

What the pattern actually shows

On the accumulated evidence from the last decade of European carnivore monitoring, what the golden jackal is doing is a natural experiment in what happens when the top predator in a landscape stops being there. Wolves are now recovering in parts of Europe, including Germany, and where they arrive in numbers the jackal expansion appears to slow or reverse locally, which is consistent with the Balkan pattern where jackal densities are low in areas that hold established wolf packs. In the meantime, the smaller canid has done in roughly forty years what nobody thought a Balkan-restricted species could do, which is walk from Bulgaria to Lapland without a single reintroduction programme, a single translocation, or a single management plan authorising any of it. The species is doing this by using farmland margins, wetlands, river valleys, roadkill, feeding stations set out for other animals, and the sort of human-modified landscape that most rewilding conversations treat as biologically dead. That the jackals are managing it without help is the point of the story. The niche was open, the animal was capable of the movement, and the animal moved.

Marshall reported no symptoms other than stomach gurgling during the first week. On day seven he felt unusually full after dinner; on day eight he vomited mucus, and colleagues noticed bad breath during the second week. Biopsies taken on day ten showed both spiral bacteria and acute inflammation. By day fourteen, the paper reported that the gastritis had largely resolved.

On 3 October 2005, the Nobel Assembly at Karolinska Institutet announced the Nobel Prize in Physiology or Medicine jointly for Marshall and pathologist J. Robin Warren. It recognised their discovery of the bacterium now called Helicobacter pylori and its role in gastritis and peptic ulcer disease. The drink was one experiment within that larger body of work, and what it proved is narrower than the Nobel citation.

Perth, 1981: a pathologist’s slides and an unwelcome bacterium

At Royal Perth Hospital, Warren had seen curved bacteria in stomach biopsies and noticed inflammation in tissue near them. Marshall joined the investigation in 1981, and the pair studied biopsies from 100 patients. Marshall cultivated an organism that would later be named H. pylori; they found it in almost all the patients in their study who had gastric inflammation, a duodenal ulcer or a gastric ulcer.

The finding challenged a view of peptic ulcers centred on acid, stress and lifestyle. Finding bacteria beside inflamed tissue, however, did not show whether they caused the inflammation or merely occupied tissue already affected by disease. Marshall’s self-experiment was designed to test that distinction.

What the broth actually proved

In the 1985 paper, Marshall and coauthors John A. Armstrong, David B. McGechie and pathologist Ross J. Glancy described their attempt to fulfil Koch’s postulates. Marshall’s stomach lining was histologically normal before he swallowed the cultured organism. The team then documented infection and gastritis in biopsies taken ten days later.

The authors described acute pyloric Campylobacter gastritis, using the organism’s name at the time. They proposed that an infection persisting beyond the acute stage might predispose a person to peptic ulcers. That was a proposal drawn from the experiment, not an ulcer demonstrated in Marshall: he did not develop one.

The result answered an important objection. Bacteria from a patient’s stomach had infected a previously healthy volunteer and caused observable inflammation. It supplied evidence that the organism could be a cause of gastritis, while leaving the proposed connection to ulcers to be tested through further work.

From gastritis to eradication, and to Stockholm

The evidence for ulcers grew through patient studies and treatment trials. As the Nobel Assembly summarised, ulcers associated with the infection often returned after acid suppression alone, while studies found lasting treatment depended on eradicating the bacteria. That evidence addressed a different question from Marshall’s two-week self-experiment: whether removing the organism changed the course of peptic ulcer disease.

The Nobel citation brought the discoveries together: Warren’s observations, the patient biopsies, cultivation of the organism, the human infection experiment and evidence from treatment. Its press release said H. pylori colonised the stomachs of about half of all humans and estimated that 10 to 15 per cent of infected people would experience peptic ulcer disease at some point.

Marshall’s drink in 1984 demonstrated that a cultured stomach bacterium could cause acute gastritis in a healthy human volunteer. The prize awarded 21 years later recognised the broader finding that the same organism plays a major role in peptic ulcers and that eradicating it can change the outcome for patients whose ulcers are linked to the infection.

The only regular management was an annual hay cut from 2011 onward, with the cuttings baled and taken away, plus hand-pulling of ragwort before each cut. Orchids were among the first newcomers, in 2013, and after 2016 one species was so abundant that the surveyors stopped mapping individual plants.

The study, by Carl Sayer of University College London and three co-authors, was published in Restoration Ecology on 29 June 2026 and is open access. Its title makes the argument directly: “No need to seed”.

Six fallow years before the first hay cut

The field, known as Sayer’s Old Meadow, lies in the village of Bodham. According to UCL it is owned by the lead author’s father, Derek Sayer, who is one of the paper’s co-authors. It was pasture for dairy cows until the 1950s, then alternated between arable crops and sown ryegrass grazed by cattle until the early 1970s, after which it was permanently arable: wheat, barley, oats, sugar beet, once potatoes, and from the late 1990s oilseed rape. The paper says fertilisers and other agrochemicals were applied every year during that arable phase.

After the final rape harvest in 2005 the field lay fallow for six years. Conservation management began in 2011, with most of the field cut and baled each August or September. A small wet area in the north and an adjoining north-eastern block left to develop into willow woodland were not cut.

That approach runs against the usual practice in meadow restoration. Agri-environment schemes and many conservation projects sow commercial or locally collected seed, and some strip topsoil or plough deeply to reduce fertility first. The paper says UK studies of grassland recovery on former arable land appear to be limited to a four-year, multi-site experiment in southern England published in 2002, which compared natural regeneration with seeding and concluded that, at least over that short period, seeding after deep cultivation worked best.

The order in which the newcomers arrived

Carl Sayer and a friend, Pete Robinson, surveyed the field every two years from 2011 to 2019 and again in 2022, after Covid-19 stopped the planned 2021 survey. Each time they recorded every species along two fixed lines of twelve 1-metre-square plots, and walked the whole field in parallel lanes until 10 to 15 minutes passed without a new find.

In 2011 and 2013 the plots were dominated by grasses and by weedy species such as field horsetail and creeping thistle. Common field weeds including scarlet pimpernel and the common poppy were gone by 2013 and were not seen again. Grass cover in the plots fell from 93 percent in 2011 to 64.7 percent in 2017, while cover of clovers, vetches and other legumes rose from almost none to 52.7 percent. Each group’s cover was estimated separately and the plants overlap, so the figures can add up to more than 100 percent.

The average number of species in a single square-metre plot roughly doubled. The paper does not give the figures in its text, but its chart shows about 9 species per plot in 2011 and about 18 by 2019 and 2022, and UCL’s news release describes the change as “around 10 in 2011 to almost 20 in 2022”.

The first southern marsh orchid, the earliest of the locally scarce arrivals, appeared in 2013 as a single flowering spike in the south-west corner, while 12 spikes of common spotted orchid were found in that corner and the centre of the field. By 2016 common spotted orchids occupied large parts of the field, and after that, the paper’s results say, they were too abundant to map. By 2022 they grew across the whole meadow, often densely.

Yellow rattle, a partly parasitic plant that taps the roots of its neighbours, was first found in 2016 in a single 2-square-metre patch. By 2022 it made up 8.8 percent of the plant cover in the survey plots, and the authors suggest its parasitism is a likely explanation for the decline of Yorkshire fog, a dominant grass, and of legumes between 2017 and 2022.

The paper’s “locally scarce” label is defined against a detailed local benchmark: species that did not appear in extensive plant surveys of all habitats, chiefly hedges, field margins and arable fields, in the same 1-kilometre grid square. The 11 present in 2022 included the southern marsh orchid, yellow rattle, common centaury and wild carrot, along with five sedges such as greater tussock-sedge. The common spotted orchid, though widespread, is not on the list.

Seeds with no obvious route to the field

How the rarer plants arrived is the part the authors cannot explain. The orchids are the easy case, because their dust-like seeds travel on the wind. Most of the others have no such help, and the field sits in a mainly arable landscape.

The nearest old species-rich grassland is the churchyard of Bodham’s 13th-century church, 0.9 kilometres away, but it holds only one of the locally scarce species that turned up in the meadow: field wood-rush, which the meadow’s surveys recorded in 2019 but not in 2022. Four valley fen grasslands 3.4 to 4.7 kilometres away together hold all the locally scarce species present in the meadow. The contractor who cuts the hay otherwise works on species-poor commercial grassland, which the authors say makes machinery an unlikely route.

Yellow rattle is the hardest to account for. Its nearest known population is 3.4 kilometres away, its seeds are not thought to lie dormant in the soil, and their large size is thought to limit dispersal to within a few metres of the parent plant. The authors suggest deer as the most likely carriers, noting that muntjac, roe deer and red deer are all common locally. They did not test that, and the paper itself says the farmland between the meadow and the fens probably greatly reduces the chance of animals carrying seed across.

They also point to the soil. The field is free-draining and sandy, and the paper says fertiliser use was “probably not as intensive as for many United Kingdom farmed fields due to the small-scale nature of the Sayer family farming operation”. The paper says how soil nutrients changed during the study is unknown, and the seed bank in the soil was not examined before the field was abandoned.

What a single family field can settle

The paper is candid that this is a single site. It says the survey was “restricted to two transects on the western side of the meadow”. Its own methods section, however, places the same two transects “on the eastern side of the field”, so which half the plot data describe is unclear from the text.

The site is also unusually well placed for a long study. UCL says the family’s ownership ensured the hay cut could be kept up through the eleven years of the study, and the paper says it guarantees the meadow’s management from here on. It also means that two of the paper’s four authors are the landowner and his son.

The study had no comparison field, seeded or otherwise, so it cannot say whether sowing would have produced a richer or poorer meadow in the same time, only what this field became when left unsown. The authors describe the meadow as moving toward three nationally valued grassland types, but they say whether it will reach them “is unknown” and that more management or deliberate introductions of species might yet help.

The coverage has rounded the story up in places. The UCL release describes a two-hectare field and says the researchers made no interventions other than hay cutting; the paper gives 1.7 hectares and describes ragwort being pulled by hand every year. ScienceDaily’s headline, “Scientists left this farm field alone. Thousands of orchids appeared”, goes further: the field was cut for hay every year, and the “thousands” is Sayer’s own description in the release rather than a count, since the paper stopped mapping orchids once they became too numerous.

Sayer’s case for trying the approach more widely rests on local character as well as cost; the paper says natural regeneration has the “potential to save money”. “Natural plant recovery better safeguards genetic diversity than seeding and ensures that local species thrive, making meadows less generic,” he said in the release. The study did not measure genetic diversity. The paper notes that the vast majority of seeded agri-environment grasslands and field margins in the region are dominated by oxeye daisy, a knapweed and bird’s-foot trefoil, all of which are present in Old Meadow but rare.

What the paper offers against the single-site objection is precedent from abroad. It says similar recoveries have been recorded on many former arable fields in continental Europe, especially dry grasslands in Romania, the Czech Republic and Hungary, usually over 10 to 20 years and less often 20 to 40. Old Meadow, 17 years after its last crop, sits inside that range, and the paper says research of this kind is widely lacking in the United Kingdom.

The work, whose lead author is Lesia Hryhorenko of Adelaide University’s School of Animal and Veterinary Sciences and whose co-authors include Monarto Safari Park staff, was published in Applied Animal Behaviour Science. It describes how the lions’ behaviour reorganised itself around a feast-and-fast schedule. The authors are careful to say it documents behaviour and does not demonstrate that the lions were better off.

The study ran from 10 February to 10 May 2025. For the first four weeks the three males ate the park’s standard diet of horse, deer, rabbit, turkey and pig, mostly as cut portions or partial carcass sections. For the next eight weeks keepers instead placed whole kangaroo or goat carcasses, about 200 kilograms in total per feed, in an off-exhibit yard, typically between 9 and 10 in the morning.

The carcasses stayed available for two to three days and were then removed, leaving six to seven days without food before the next one. The authors cite earlier work reporting that lions in the wild typically feed every 1.5 to 2.5 days but have been seen fasting for up to eight.

Eight infrared cameras recorded a 10-second clip every five minutes, day and night. Two trained observers scored the footage, and on the 23.5 percent of sessions that both scored, their agreement averaged above 97 percent. Because individual lions could not be reliably told apart on camera, behaviour was scored for the group as a whole rather than for each animal.

Almost three times as long at the carcass

On the carcass days analysed, seven in the paper’s count, at least one lion was feeding for an average of 76.4 minutes a day, against 25.7 minutes a day at baseline, a ratio of just under three. The university’s news release says the lions spent “almost three times longer feeding than they normally would”. The lions began eating on average 16.4 minutes after the carcasses went in, and fed in more frequent bouts than before.

The span between the first and last feeding scan after a carcass went in stretched to as long as three hours and 35 minutes. The lions also ate at the same time far more often: scans with two or more lions feeding rose from 1.2 percent at baseline to 19.0 percent on carcass days. The authors add, as an anecdotal observation, that the three males usually split up to eat, each taking a carcass of his own rather than working on one together.

Feeding did not rise as a monthly average in a way the statistics could confirm. The measure tended to be higher in the first month of carcass feeding, but the overall effect across months was not significant. The increase shows up when carcass days are compared with baseline days, and within the carcass months feeding was concentrated in the first three days of each cycle.

A nine-day rhythm of lying down and moving

On days one to three of each cycle, which the paper treats as the feeding days, lions were recorded lying down in 52.5 percent of scans. That fell to 30.4 percent on days four to six and 17.0 percent on days seven to nine. Averaged by month, lying down roughly doubled, from 15.8 percent at baseline to 34.9 and 32.8 percent in the two months of carcass feeding.

Walking about moved the other way. It was lowest on days one to three, at 1.53 percent of scans, rose to 3.08 percent in the middle of the fast and stayed slightly raised at 2.39 percent near its end. The authors compare this with wild lions, which alternate heavy feeding with long spells of reduced activity broken by searching or waiting.

Pacing rose over the study instead of falling as predicted

Pacing, the back-and-forth walking along a fixed route that is often interpreted as a stereotypic behaviour in captive carnivores, did not follow the script. Based on earlier work, the team expected it to fall once carcass feeding began, while noting that any change might depend on the length of the study and the gap between feeds. Instead it rose, from 1.79 percent of scans at baseline to 3.83 percent in the second month of carcass feeding, a difference that was statistically significant.

The university’s release describes this as pacing that “increased slightly over the course of the study”. In the paper’s own figures it more than doubled, although from a low starting point, and both numbers describe the group rather than any one lion.

What pacing did not do was change significantly across the nine-day cycle. Averages went from 2.26 percent on days one to three to 3.78 and 3.81 percent on the later days, differences the statistical models did not find significant.

Plotted hour by hour, pacing peaked around 9 to 10 in the morning and 3 to 4 in the afternoon, when keepers arrived and moved the lions between enclosures. The authors read this as anticipation of predictable routines rather than a build-up driven by the fast.

The paper handles this point unevenly. Its abstract says pacing “rose toward the end of each nine-day interval, suggesting increasing anticipation before feeding events”, while its results describe that rise as small and not statistically significant, and its discussion refers to “the absence of escalation across fasting days”.

This article follows the results. The authors also note that their scoring could not separate pacing from other repetitive or potentially goal-directed movement, such as patrolling.

A control group that differed from the start

The study had four other lions, two males aged three and two females aged five, kept on the usual diet as a comparison. During the baseline month the two groups already behaved differently: the test males spent much less time inactive and more time active than the control lions.

The authors therefore did not compare the groups once carcass feeding began. The comparisons with baseline are before-and-after changes in the same three males, and the cycle comparisons are made within the carcass months.

That leaves other explanations open. When carcass feeding started, the three males were also kept out of public feeding encounters and moved between enclosures on a different pattern, which the authors say “may have contributed to some changes in pacing or visibility”. At night the lions were often in a large exhibit where parts of the ground were out of camera view. A check in the paper’s appendix found no significant link between time out of sight and recorded pacing, but it tested pacing only, and the paper notes that resting and lying down could also happen out of view in the larger enclosure.

Because the conclusions rest on changes within three males of the same age, the authors write that this “limits generalization to other social structures, such as mixed-sex groups or prides”. The results also come from one facility and a single three-month stretch from late summer into autumn, scored for the group rather than each lion.

The authors add that “behavioural change alone cannot be used to infer welfare outcomes”. They also report, without giving figures, that no increase in aggression or social instability was seen around the carcasses.

Why the authors question treating carcasses as mere enrichment

The release frames the study as showing that carcass feeding “encourages behaviours more closely aligned with those seen in the wild”. The paper goes further in its introduction, questioning whether whole carcasses should be treated only as enrichment. For obligate carnivores that evolved to eat whole prey, the authors write, providing carcasses “could be argued to represent a core component of meeting biological and behavioural needs rather than an optional ‘extra'”.

Their concern is with the label. Framing carcass feeding solely as enrichment, they argue, “risks lowering the perceived welfare baseline by implying that less natural feeding methods constitute an acceptable standard of care, rather than distinguishing between fundamental feeding requirements and additional forms of environmental stimulation”.

The three males at Monarto went from cut portions and partial carcass sections to whole kangaroo and goat carcasses, and the study’s own data cannot say whether that made their lives better. What it records is that they fed, rested and moved to a nine-day clock, set by the carcasses and by the changes in where they were kept that came with them.

What turns a geophone into an actual scientific instrument is everything wrapped around it: the housing, the circuit board that reads its signal, the software that decides whether a wobble is a passing truck or an earthquake. That part usually belongs to a lab with a budget, not a bedroom.

Vivien He built hers in a bedroom. She was sixteen when she started, a junior at Palos Verdes Peninsula High School in Southern California, and she finished the project the way a lot of ambitious 2020 projects got finished, alone, at home, with the rest of the world shut down around her. The device that came out of it costs less than $100 to make, fits in a hand about the size of a Rubik’s cube, and has spent every year since doing exactly what she designed it to do.

I’m not an engineer or a scientist, and most of what actually happens inside a working seismometer is past my ability to explain with any authority. What I can follow is the shape of the story, and this one has a shape worth sitting with: an ordinary teenager, a pile of parts you could buy with a part-time job’s paycheck, and a device that now sits in the same conversation as instruments built by real seismology labs.

The idea came from a research paper, not a disaster. During the early lockdowns, seismologists noticed the planet had gone quieter, literally, with less human vibration moving through the ground as fewer cars hit the road and fewer people commuted to work. He read about that seismic hush and got curious about whether she could measure anything like it herself. “I was wondering whether I would be able to measure that from my own home,” she recalled, “and then that quickly evolved into, I wonder if I can measure in my own home and apply it to earthquake early warning?”

That is a big jump for a high schooler to make on her own, from noticing an interesting data point to deciding she would build a functioning early-warning device. She had no formal training in the electronics or programming the project needed, and no class was going to hand it to her either. Her method, by her own account, amounted to looking things up as she went, the same way any curious teenager teaches herself an unfamiliar skill off the internet now, one search at a time until the picture holds together.

Her lab was a corner of her bedroom, drawers full of spare wires and geophones, plus a bathroom down the hall she borrowed because the lighting was better for soldering. Laser-cutting the acrylic case meant hauling a table into the backyard, with her dad standing by as a safety precaution, which turned out to be a reasonable call. One soldering session started a small fire she asked her family not to mention to her mother.

She plugged in the finished device for the first time one night after midnight in September 2020, then went to bed. The next morning, Los Angeles had an earthquake, and the seismometer had recorded it. “And then I went to sleep, and then the next day I woke up and there had been an earthquake in Los Angeles and I was like, oh, it’s fate!” She checked her instrument’s waveform against a nearby U.S. Geological Survey station near her house and found the two traces matched.

That first catch was not a fluke. By the following spring, her device had detected every earthquake above magnitude 3.0 around Los Angeles since the day she switched it on, giving anyone using it a few to tens of seconds of warning before the shaking arrived, enough time to get under a table, step away from a shelf, or, in a commercial setting, shut equipment down automatically. Small local earthquakes are notoriously hard to catch consistently. They are too frequent and too regional for the big national systems to be built around them, and too faint for most consumer devices to register over background noise. Going a full stretch without missing one, on hardware she designed herself, is the part working seismologists found worth paying attention to.

They did pay attention. She presented her research at the Seismological Society of America’s 2021 annual meeting and was the only high school student that year to win the organization’s Student Travel Grant. One of the researchers whose work she had read while teaching herself the field was Richard M. Allen, director of the Berkeley Seismology Lab. She went on to compete at the International Science and Engineering Fair, founded a nonprofit called Melior Earth aimed at getting cheap early-warning devices into lower-income neighborhoods with older, less earthquake-resistant buildings, and by 2022 had been named a Davidson Fellow and a Regeneron Science Talent Search Scholar on her way to Stanford. None of that changes what the device itself was built to do. Asked why she kept going, she did not reach for the recognition. “The whole point of it is that it would be a consumer product, but I’m not focused on the monetary gain of it,” she said. “I’m more focused on the science of it and just the impact in general on people and on earthquake disaster prevention.”

What holds my attention in this story is the consistency. I’m a competitive person, always have been. I like being good at things, genuinely good, not just good enough to get by, and I have spent real effort learning how to want that without letting it take over the calmer home life I am trying to build around it. Reading that a seventeen-year-old set out to catch something as small and slippery as a magnitude 3.0 earthquake, then checking the record and finding she actually did it, every time, for as long as anyone tracked it, gives me something close to recognition. I don’t have the background to judge her instrument, but I recognize the discipline behind never missing one, the kind of standard a person keeps for herself long after the novelty wears off and nobody is watching anymore.

A cheap geophone in a plastic case does not know it is supposed to be impressive. It just sits there, waiting for the ground to move, doing the one job it was built for. Somewhere in Southern California, one still is.

What the argument had been

The reason the removals were seen as radical at the time is that the assumption running through most of twentieth-century European water policy was that dams were what a modern country built, and the biological argument for taking one out was thought to be unserious next to the argument for keeping it. Both of the Loire dams that came down in 1998-99 had been operated by Electricité de France, and both had come up for licence renewal in 1994 under a piece of French policy called the Plan Loire Grandeur Nature. That plan, adopted by the French government in January 1994, was itself the result of a long, hard-fought campaign by conservation groups and local associations against a proposed sequence of new dams on the upper Loire and the Allier. What the plan did, in the end, was reverse the direction of policy: rather than approve more dams, it set out a strategy of restoring ecological continuity to the basin. The non-renewal of the Saint-Étienne-du-Vigan and Maisons-Rouges licences was the first serious test of what that reversal meant on the ground.

Two rivers both get called Europe’s last wild one, and the difference is the whole point. The Loire is free-flowing because France undid its dams; the Vjosa is free-flowing because nobody ever got to build them. Watch below:

What happened when the water started running again

The response upstream was faster than anyone had expected. According to the European Rivers Network’s case file on the Saint-Étienne-du-Vigan removal, five salmon spawning sites were observed in the reach immediately above the former dam site in the winter of 1998-99, which was the first winter after the demolition. That is, salmon returning from the Atlantic that autumn had found the obstacle gone, swum through the section that had been sealed off since 1895, and started cutting redds in the gravel within months. Downstream on the Vienne, once the Maisons-Rouges dam was removed, the effect was even more dramatic in absolute numbers. Dam Removal Europe’s monitoring record for the site shows that a fish-counting station about 20 kilometres upstream of the demolished dam went from recording essentially no allis shad, sea lamprey or Atlantic salmon before 1999 to registering nine salmon, several thousand shad and several thousand sea lamprey by the end of that year, and to registering 57 salmon and more than 41,000 sea lamprey by 2007. Roughly 35 kilometres of previously blocked river were reopened. Today the Vienne basin holds about 80 per cent of the Loire basin’s sea lamprey population and 15 to 20 per cent of its Atlantic salmon.

Why the Loire is now what it is

The Loire is a little over a thousand kilometres long, from its source at Mont Gerbier-de-Jonc in the Ardèche to its estuary at Saint-Nazaire, and it drains around a fifth of France. There are still dams on parts of its upper course and on some of its tributaries: Villerest and Grangent in the upper basin, Naussac on the Allier, a partially reconfigured hydropower structure called Poutès further down the Allier. What the Loire does not have, and what makes the description “the longest free-flowing stretch of water in western Europe” defensible, is any dam of consequence on the middle and lower main stem, which is the section that runs from below Villerest all the way to the sea. That stretch is close to a thousand kilometres of connected river, and there is nothing else in western Europe that comes close. The Rhine has been dammed, canalised and re-engineered for two centuries. The Rhône runs through a chain of hydropower dams. The Thames and the Elbe were long ago cut off from their salmon runs entirely.

What the two removals actually settled

What Saint-Étienne-du-Vigan and Maisons-Rouges established, on the accumulated evidence of the two decades since, was that the biological response to reopening a river was not a hopeful projection but a measured fact. Salmon do come back. Shad and lamprey come back faster. The riverbed reforms into gravel bars and riffles within a couple of high-flow seasons. Sediment starts moving again downstream in the way it used to before impoundment stopped it. The specific French policy shift that made the two removals possible has since been followed by roughly a dozen other removals in the basin, including a partial demolition of the eighteen-metre Poutès dam on the upper Allier that was completed in 2022 to allow salmon a genuine chance at the highest spawning grounds. What was considered eccentric in 1998 is now an established branch of European river management. The 1,000 kilometres of connected water that runs down the middle and lower Loire is what a river doing that management looks like after a generation.

I am not a materials scientist, a civil engineer, or a chemist, and nothing below should be read as expertise I do not have. What pulled me into this particular paper was less the concrete and more the shape of the experiment: a team took a material nobody especially wants back, ground it down, and ran a longer, more patient test than the standard actually required of them.

Why anyone bothers grinding up glass for concrete

Glass recycling has a reputation it does not earn. In the United States, only 31.3 percent of glass containers actually got recycled in 2018, the most recent year the EPA has published figures for, with most of the rest going straight to landfill. Glass does not break down there. It just sits, taking up space it will occupy for longer than anyone reading this article will be alive.

That gap between how recyclable glass technically is and how rarely it actually gets reused is what sent a team at the Federal University of Mato Grosso, in the Brazilian city of Sinop, looking for a use for it inside concrete blocks. As the researchers put it in their own opening line: “Glass waste is generated in large quantities worldwide, yet only a small portion is recycled or reused.” Concrete blocks, it turns out, were willing to take some of it off their hands.

What the researchers actually swapped

Half the sand, by weight

The team built a plain control batch of dry concrete blocks with no glass at all, then built three more batches where crushed glass stood in for a quarter, half, and three quarters of the sand. The batch at exactly half, the one named in this article’s title, ended up mattering most by the end of the study, for reasons that only showed up once the testing ran long enough to see them.

Ground fine enough to disappear into the mix

The glass was not tossed in as chunks. It was ground down to a particle size under 600 microns, roughly the width of a coarse grain of table salt, before it went anywhere near the mixer. At that size, crushed glass packs into a mix instead of sitting inside it as an oversized, sharp-edged shard, which is a meaningfully different job for the same waste material to do.

The 28-day number everyone would have accepted

Twenty-eight days is roughly the age at which concrete gets judged by default. It is when the plain control blocks in the Mato Grosso study reached 3.11 megapascals of compressive strength, the baseline every other number in the paper gets measured against. Had the researchers closed their notebooks there, the results would still have looked fine. Every glass mixture, from a quarter replacement up to three quarters, beat the control by a wide and statistically significant margin, an average of 32 to 35 percent higher across the different substitution levels.

The number that took an extra month to show up

The single best result of the study did not arrive at day 28. It showed up at day 56, four weeks past the point where most testing would have wrapped up, when the batch built with half sand and half ground glass reached 4.35 megapascals. That mix also posted a characteristic strength of 3.62 megapascals, clearing Brazil’s required minimum of 3 megapascals for load-bearing blocks. A quarter or three quarters glass beat the control too, but only the fifty-fifty batch, tested well past the usual finish line, actually met code.

The absorption test the blocks still had to pass

Strength was not the only thing riding on this. Concrete blocks also get tested for how much water they soak up, since a block that drinks in water is one that eventually cracks, spalls, or grows mold behind a wall. Every mixture in the study, glass or no glass, absorbed between 6.0 and 7.5 percent water, comfortably under the 10 percent ceiling the standard allows. Replacing half the sand with glass did not make the blocks any thirstier.

What one study can tell you, and what it can’t yet

One laboratory result is not an industry standard. The blocks in this study were dry-pressed concrete units tested against a specific Brazilian building code, not a stand-in for every kind of concrete made everywhere. As far as I could find, nobody has independently repeated this particular experiment yet, which matters if you want a settled fact rather than a promising early one. What held up was the design itself: a controlled comparison against a real baseline, a statistically significant result, and a number that had to clear an actual government-grade minimum, not just an internal sense of what counts as strong enough.

Why the day 56 number is the one that stuck with me

I hold a fairly stubborn opinion that discipline is what keeps a person from getting swept along by a fast, distracting world that has no particular vision of where it wants you to end up. That belief has nothing to do with construction materials, but it is the reason the 56-day number sat with me longer than the 28-day one did.

Stopping at day 28 would have been enough to publish something defensible. The control number was already sitting there, the glass mixtures were already ahead of it, and nobody would have faulted the team for stopping. Instead they kept the blocks curing and kept measuring for another month before writing anything down as final. The gain between day 28 and day 56 barely shows up as a bump on a chart. It only exists at all because somebody was still checking.

A result that needs eight weeks instead of four is a harder thing to market and an easier thing to skip past. It is also, based on what stuck with me in this particular paper, usually the more honest one. Most progress worth having, in a lab or anywhere else, tends to belong to whoever was still paying attention after the point where it would have been reasonable to stop.

What the fMRI showed

To understand what that something was, Longcamp’s team put a subset of the participants into a functional MRI scanner and showed them the characters again. The instruction this time was passive. Look at the characters, do not respond, do not move, do not attempt to produce them. What the scanner picked up was that the participants who had learned to write those characters by hand were activating, during purely visual recognition, brain regions that have nothing to do with vision. Specifically, the handwriting-trained participants showed elevated activation in the left Broca’s area, which is one of the classical language-production regions, and in the bilateral inferior parietal lobules, which are associated with the execution, imagery and observation of hand actions. The typing-trained group showed no such activation for the same stimuli. Their brains were treating the characters as pictures. The hand-trained brains were treating them as actions.

If you want a friendly, practical companion to all this, below is a short video on the six things a pen-and-paper journalling habit quietly does for you, with a few concrete prompts to try.

The action-perception loop the study exposed

What Longcamp’s group had documented, on the strongest current reading of that result, was a version of what cognitive neuroscience calls an action-perception coupling. When a person learns to produce a specific movement associated with a specific visual stimulus, the visual stimulus subsequently activates the motor programme for producing it, even when there is no requirement to act. The same phenomenon has been documented for tool use, musical instruments, and sign language. What Longcamp’s 2008 paper added was that the effect operates at the level of individual written characters, and that it appears within a few weeks of training. Reading, in this view, is not a pure visual process. It is a partially motor one, in which recognising a shape includes silently rehearsing the movement that would produce it.

This was already implied by an earlier fMRI study by the same group in 2003, which had shown that adult right-handers passively viewing alphabetical letters activated a region of left premotor cortex that overlapped with the region they activated when actually writing those letters. The 2008 study extended the finding by demonstrating that the coupling can be built experimentally in adults, in a matter of weeks, and that it does not develop in the same way when the training uses a keyboard.

What later work has added

The Longcamp finding has been replicated and extended by groups working with different techniques and different tasks. In 2021, a research team at the National Institute of Information and Communications Technology in Japan, led by Aya Ihara, ran a conceptually similar experiment on adult native Japanese speakers who were learning novel Indonesian words. According to the resulting paper in Frontiers in Human Neuroscience, participants who learned the words by writing them out by hand, whether with an ink pen on paper or with a digital pen on a tablet, subsequently showed a larger N400 event-related potential response to those words than participants who had learned the same words by typing. The N400 is one of the most consistently measured electrophysiological indices of semantic processing during word recognition, and a larger repetition-priming effect on N400 is generally taken to indicate that the learner has formed a more accessible semantic representation of the stimulus. The Ihara team’s finding, in other words, is what Longcamp’s finding predicts: written-by-hand stimuli come with a richer downstream response when later read.

The oscillation pattern picked up separately

A different NTNU study in 2020, published by Ose Askvik, Van der Weel and Van der Meer in Frontiers in Psychology, used high-density 256-channel EEG in twelve adults and twelve 12-year-olds to look at what their brains were doing during the act of writing itself. When adults wrote cursively with a digital pen on a touchscreen, their parietal and central brain regions showed event-related synchronised activity in the theta band, between 4 and 8 Hz, which is the frequency range that a large existing literature associates with the encoding of new information and the formation of working memory. When the same adults typed the same words on a keyboard, those regions showed the opposite pattern, event-related desynchronisation in the theta and alpha bands, which does not have that association with encoding. The 12-year-olds showed the same pattern less distinctly, consistent with a system that is still developing.

What the accumulated evidence points to

Taken together, on the accumulated evidence from these three lines of work, the underlying claim is fairly specific. When a person learns a written character by producing it with their hand, the visual representation of that character in their brain becomes coupled to the motor programme that produced it, and subsequent recognition of the character will partially reactivate that motor programme. When the character is learned by pressing a key, no such coupling is built, because the motor programme for pressing a key does not encode the character’s shape. This does not mean typing is useless, or that reading is broken in people who never learned to write by hand. It does mean that reading, in the fully wired version that literate cultures have historically produced, is a coupled visual-motor skill rather than a purely visual one. The recognition of a letter is, at the neural level, partially a silent enactment of drawing it.

The engine, meanwhile, keeps running hot.

How caffeine blocks the tiredness signal

What exactly gets disconnected? That wire carries a molecule called adenosine, a by-product of the brain’s ordinary daily graft. In a 1997 paper in Science, researchers at Harvard Medical School measured adenosine in the brains of cats and found it climbed during waking hours, kept climbing through prolonged wakefulness and drained away only slowly during recovery sleep. That was one lab working with animals, but it cast adenosine as a running tally of time spent awake.

Nerve cells read that count through adenosine receptors, and caffeine fits the same slots. The StatPearls clinical reference on caffeine, hosted by the US National Library of Medicine, explains that it blocks all four receptor types, with one subtype, known as A2a, doing most of the work of keeping people awake. Think of a key snapped off inside a lock. Adenosine keeps turning up, finds the keyhole jammed, and the message about tiredness never gets through.

Blocking the receptors leaves the count itself untouched. Adenosine goes on piling up behind the jammed locks for as long as the caffeine holds, which helps explain why the tail end of a long, heavily caffeinated day can land like a dropped piano.

Why an afternoon coffee lingers until midnight

Caffeine clears out of the body slowly. StatPearls puts its half-life in the average adult at roughly five hours, so half of any dose is still circulating that long after the last sip.

Run that against the 2pm long black. By 7pm, half remains. By midnight, two half-lives on, about a quarter is still parked on those receptors, and a quarter of a strong coffee is hardly nothing.

Five hours is only an average, too. That same entry notes that smoking can shorten the half-life by up to 50 per cent, while late pregnancy can stretch it to 15 hours, and liver disease, or medicines that gum up the enzymes doing the clearing, drag it out further. Two colleagues downing identical cups at neighbouring desks may be carrying very different loads by bedtime.

What daily coffee does to baseline alertness

‘Merely returns it to baseline.’ Those five words close the abstract of a 2010 study in Neuropsychopharmacology, led by Peter Rogers at the University of Bristol, and they hit like a bucket of cold water.

Rogers and colleagues kept 379 people off caffeine for 16 hours, then gave them either caffeine or a placebo. Regular drinkers handed the placebo grew less alert and more headachy, and caffeine did little more than haul them back to normal. Among people who rarely touch the stuff, it failed to lift alertness at all. Their conclusion was that frequent drinkers gain no net benefit, because the caffeine mainly cancels out withdrawal.

A plausible reason sits in the StatPearls entry on caffeine withdrawal, which describes how heavy, long-term use can prompt the brain to add extra adenosine receptors. More receptors make the tiredness signal louder whenever caffeine goes missing, producing the fatigue, drowsiness and headache that many daily drinkers know from a skipped morning cup. The Bristol verdict comes from a single study, so it stops well short of settled science, though it fits the receptor story uncomfortably well.

Caffeine six hours before bed and lost sleep

Twelve healthy sleepers once spent four days swallowing three pills a day, timed at six, three and zero hours before bed. On three of those days, one pill held 400 milligrams of caffeine and the other two were dummies. On the remaining day, every pill was a dummy. The trial, run at the Henry Ford Hospital sleep research centre in Detroit and published in the Journal of Clinical Sleep Medicine, found significant sleep disruption at all three timings.

The six-hour dose is the one that stings.

In a press release, the American Academy of Sleep Medicine reported that caffeine taken six hours ahead of bedtime still cut objectively measured sleep by more than an hour. Participants’ own sleep diaries suggested they hadn’t noticed. Twelve people on a dose the academy equates to two or three cups makes for a small, blunt experiment, but it points exactly where the half-life maths does.

Put that beside the withdrawal findings and a grim little loop suggests itself, though joining those dots is speculation. A late coffee shaves an hour off the night, nobody clocks the loss, the morning feels rough, and the remedy of choice is another cup that, on the Bristol evidence, only restores normal service.

That jolt has a name.

Sleep scientists call it a hypnic jerk, from the Greek hypnos, meaning sleep, and it also answers to sleep start or hypnagogic jerk. It belongs to a family of brief, involuntary muscle twitches known as myoclonus, a group that includes the humble hiccup. For something so jarring, it is remarkably ordinary.

How common hypnic jerks are

Most people have had one.

A guide to sleep starts from the American Academy of Sleep Medicine (AASM) puts the figure at 60 to 70 per cent of people, adding that many go unrecalled, which is why the jolt can seem rarer than it is. A typical episode is one strong contraction, mostly in the arms and legs, occasionally with a sharp cry thrown in. Some people feel themselves falling at the same instant, or catch a sensory flash or a scrap of dream. Bed partners, who tend to clock the kick well before its owner does, can fill in the rest.

None of this is new to medicine. In 1959, Ian Oswald’s paper in the journal Brain tackled the subject under the gloriously literal title “Sudden bodily jerks on falling asleep”. These days the International Classification of Sleep Disorders, the field’s reference manual, files hypnic jerks under isolated symptoms and normal variants, as a 2018 review by neurologists at Toronto’s Hospital for Sick Children explains. That is clinical shorthand for a generally harmless quirk.

Why relaxing muscles might feel like falling

So what is the brain actually reacting to? One leading explanation starts with the mechanics of dozing off. As sleep takes hold, muscles lose tone and the body goes slack. Per the Sleep Foundation’s overview, researchers suspect a misfire between nerves in the reticular brainstem, part of the same machinery that drives the startle response. The brain, the theory goes, registers that sudden loss of tension, mistakes it for a genuine fall, and fires off a twitch to catch a body that isn’t going anywhere.

Think of it as a smoke alarm set a touch too sensitive, shrieking because someone opened the oven. It is a neat explanation, though an unconfirmed one. A second possibility, floated in the same overview, is that the jerk is a reaction to the dreamlike imagery that sometimes arrives with it, and the Toronto review concludes that the origin and physiology of hypnic jerks are still poorly understood.

An evolutionary idea involving tree branches

“An archaic reflex.” That is how psychologist Frederick Coolidge and anthropologist Thomas Wynn, both of the University of Colorado, Colorado Springs, described one common hypothesis in a 2006 paper published in Before Farming. Picture an early primate asleep on a branch or in a nest high in the canopy. Its muscles relax, its brain reads the slackness as the start of a plunge, and the jolt wakes it just long enough to shift back to safety. Coolidge and Wynn also suggest the reflex may have nudged sleepers to review their position before anything went wrong.

The stakes would have been real. The pair cite research putting great ape tree nests at an average of 10 to 20 metres off the ground, a long way to drop mid-snooze. Still, this is one paper, and the jerk is a passing point in a larger argument about how sleeping on the ground may have shaped human thinking. The authors present the falling-primate idea as a hypothesis, and it has stayed one.

What makes hypnic jerks more frequent

A late espresso, a deadline that won’t let go and a hard gym session after work stack the odds towards a jumpy night. The AASM lists heavy caffeine or stimulant use, earlier intense exercise or physical work, and emotional stress as things that make sleep starts more frequent and more forceful. Nicotine and plain old sleep deprivation make the foundation’s list too.

There is a nasty loop tucked inside all this. Frequent or intense jerks can leave some people dreading sleep itself, and the AASM notes that this anxiety can make nodding off harder and, over time, tip into chronic insomnia. Its threshold for seeing a sleep specialist is practical: when the jerks, or the fear of them, are costing you sleep. Daytime jerking is a different matter: persistent contractions that spread while awake could point to another kind of myoclonus, and the foundation suggests seeing a doctor.

For everyone else, the advice from the Sleep Foundation is about as unglamorous as advice gets: ease off caffeine and nicotine, shift hard workouts earlier, keep a regular bedtime and find some way to shed the day’s stress before lights out. It concedes that the jolts are unlikely to disappear entirely, even for the most disciplined sleeper.