Three hundred thousand hives go into the almonds by late July. What decides whether a colony arrives at eight frames or four is already happening, out of sight, under capped brood.
Every yard visit is the same two calls. Treat this colony, or leave it and spend the treatment on another one. Send this colony to the almonds, or hold it back and send one you trust more. Both calls are made weeks before the thing they are actually about, which is how strong the colony is on the day it is unloaded in an orchard. Both are made while standing in front of a hive that looks, today, fine.
And the margin on that call is thin. In the season AHBIC reported on, some beekeeping businesses resorted to under cutting prices to obtain pollination contracts for business cashflow
, and almond pollination was complicated due to cross border rules due to varroa
. Honey sat around $4.50 a kilogram with packers holding excess stock. The pollination cheque is carrying more of the year than it used to.
You are not paid for mites. You are paid for frames of bees. The working standard for almond pollination is a minimum of six frames of bees, with eight or more highly desirable. Varroa is one of the things that decides which of those two numbers you deliver, and the only one nobody can see on the day.
The decision is made in June. The evidence arrives in August.
the structural problem, in one lineA yard in June. You shake three hundred bees off a brood frame into a jar, wash them, count what comes off. The number is low. On that evidence the colony does not need treating, so it is not treated. There are twelve more yards to get through today, and the product runs to around fifty to sixty dollars a hive a year across four treatments.
What the wash counted was the mites riding on adult bees. The ones that matter are not there. They are sealed inside capped brood cells, where a founder mite walked in before the cell was capped and is now reproducing on the developing bee: unreachable, and uncountable. How much of the colony’s mite population is hidden in that compartment depends on how much brood is capped on the day you sample. So the wash reads low when the colony is building hardest.
Six weeks on, that brood has emerged. Every cell that held a founder has released her and her daughters onto the adult population, and those mites have gone straight back into the next round of cells. The bees being reared on those frames now are the bees that will fly the almonds in August.
And your yard’s load is not only your yard’s making. Mites arrive from outside, carried in on bees robbing out a collapsing colony nearby, or drifting between hives. A single count cannot show you that. A count is a level, and this is a rate.
Late July, the truck is loaded. The colony that read clean in June is unloaded in an orchard at four frames. Nothing went wrong on the day. It went wrong six weeks earlier, in the one part of the hive nobody looked in, because nobody can.
Hives go into the orchards by late July, and the almond industry will pay the beekeeping community in the vicinity of sixty million dollars across a four-week blossom phase. Whatever a colony is on that date is what it is worth. There is no version of the season in which you get another fortnight to fix it.
Pyrethroid resistance was first confirmed in northern New South Wales in January 2026 and in Queensland the following month, and the same mutation has since been confirmed in Victoria. Formamidine resistance, which covers amitraz, has been confirmed in northern NSW and southeast Queensland apiaries, and found in Victoria and South Australia. Bayvarol, Apistan, Apivar and Apitraz are no longer recommended. Four unaffected products remain, and all four are organic acids.
Transition to Management, the national plan that followed the abandonment of eradication, ceased in February 2026, on AHBIC’s own account. Industry has agreed to a 1c/kg honey levy increase to repay the federal government for its share of the cost of the varroa response and T2M, to stay in place until that share is repaid. The national support has stopped and the bill has started. Registered hive numbers fell approximately 5% to 823,291 by December 2024, and by 13% in New South Wales, where varroa is most widespread.
Those two facts change what a treatment decision is. It used to be a cost, and a modest one: around fifty to sixty dollars a hive a year, four treatments, a line in the budget. It is now also a withdrawal. Efficacy is a shared resource. It is spent by being used, and it never refills. Every colony treated that did not need it is spent out of the four products left for the colonies that will.
You cannot simply treat more often either. Label directions carry a potential withholding period before extraction, or residue testing. The window in which you are allowed to treat runs into the window in which the hive earns. And inside the detection zones the monitoring load is set for you: beekeepers with hives within twenty-five kilometres of a detection site shorten their monitoring intervals to every three to four weeks.
A budget that refills rewards spending it early. A budget that does not rewards spending it correctly.
The rule is a three-hundred-bee alcohol wash against a seasonal action level: sample, count, treat above the line. It costs nothing but time, it measures something real in the yard in front of you, and it is completely honest about what it is. In most weeks of most seasons it returns the same answer a model would. Any forecast that cannot beat it is not worth what it costs to run.
There is also a published model, and a serious one. Mshelbwala, Mulholland, Doyle, Anderson and Hetherington, all five at the New South Wales Department of Primary Industries and Regional Development, fitted a Bayesian hierarchical logistic regression to 10,118 apiaries surveyed between September 2022 and July 2025, of which 984, or 9.7%, were positive. This is the department that ran the response, modelling its own surveillance data, in Scientific Reports, 4 December 2025.
It answers where varroa will be found. It does not answer what a colony’s mite load will be, and the authors say so themselves. The binary nature of the detection outcome, they write, limits the ability to capture the full dynamics of mite infestations.
The model did not account for key risk factors, such as informal hive movements, landscape connectivity
. That is re-invasion, named. And surveillance conducted in colder months may underestimate infestation levels.
That gap is not ours to assert. It is stated in print by the people best placed to state it.
So the bar is not find varroa. The bar is this. Does a forecast, made on the day you are standing in the yard with the product in your hand, change which colony you spend it on? And does that change show up as frames on the truck?
One published number says it might, and it did not come from us. Boehm Vock, Mossman, Rapti, Dolezal and Clifton, in PLOS ONE, trained on 266 apiaries and tested out of sample on a further 140: colonies that were monitored and then treated grew mites at 0.5% per day; colonies treated without monitoring grew at 1.7% per day. Their own conclusion is that treating without monitoring is no more effective than not treating at all.
The chemical was never the thing. Knowing which colony to put it in was the thing.
our reading of Boehm Vock et al., 2025It is worth more now than it was a year ago, because four of the eight products Agriculture Victoria lists are no longer recommended.
The wash is not lying. It reports the one compartment it can reach, which is mites riding on adult bees. As the colony builds brood toward the almonds, more and more of the population moves into sealed cells, where nothing can count it. So the reading drifts sideways, and can even fall, while the population underneath it climbs. The colony crosses the action level in early July and stays over it right through the bloom, and the sample never says so. Underneath sits the resource you would spend to act. Half of it has already been withdrawn by resistance.
Schematic, on a log scale. The population is drawn growing at up to 1.7% per day, the published rate for colonies treated without monitoring (Boehm Vock et al., 2025), scaled by the brood available to reproduce in. The share hidden in sealed brood is illustrative, not fitted. The bloom window, the action-level rule and the product list are sourced below.
Frame count is what a pollination hive is assessed on and what it is paid for; six is the minimum and eight or more is the working benchmark. Mite load is one driver of it among several. Forecasting the priced quantity is more useful than forecasting the parasite. It is also easier to grade honestly, because the answer arrives on the day the truck is unloaded, with everyone standing there looking at it.
A colony’s mite population under closed-population reproduction has a known growth shape. Where the observed trajectory departs from that shape, mites are arriving from outside. That is measurement of something already happening, not prediction of something unmodelled. It also identifies which of your yards sit near a source, without ever observing anyone else’s hives.
Four products, each with a withholding period, against a run of yards and a bloom date that will not move. The question is not which colonies are worst. It is which treatments, in which order, across which yards, buy the most frames on the truck, under the constraint that efficacy spent is efficacy gone.
Frames at delivery, predicted against what was actually unloaded, across seasons already lived, using only the data that existed on the day each call was made. Scored against the wash and the action level as they are practised now, not against doing nothing.
We have not found an Australian pollination agreement that ties a numeric mite count to a payment band or to a rejection. Varroa appears in them as a compliance and documentation obligation. Frame count is what is priced. That is why the forecast should be of frames.
A pilot runs your own decisions again across seasons you have already lived, scored against what actually happened. You already know how those seasons went. That is what makes it a test rather than a demonstration.