The number, the notice and the signature.

There is no international harmful algal bloom threshold. There are national and state ones, a small number of guidelines from the World Health Organisation, and a large amount of local practice that was set by a committee in the nineteen-nineties and never revisited.

Anyone writing or reviewing a programme should start by accepting that. What follows is not a list of numbers to adopt. It is an account of what kinds of thresholds exist, what each one can carry, and the published figures a directorate can cite without being embarrassed by a reviewer.

1. Three different objects, all called a threshold

The word does a lot of work and hides three unlike things.

A guideline value is a concentration in a matrix, derived from toxicology, above which harm is expected over a defined exposure. It is a health number and it knows nothing about your treatment works.

WHO publishes three of them for microcystins, and this is where most programme documents go wrong. The provisional lifetime drinking-water value is one microgram per litre. There is a separate provisional short-term drinking-water value of twelve micrograms per litre, which WHO describes as indicating the extent to which the lifetime value may be exceeded for up to two weeks while treatment is augmented. And there is a provisional recreational water value of twenty-four. All three are derived from studies using microcystin-LR and are applied to total microcystins.

A document that carries only the one is not being cautious. It is missing the number a duty engineer needs at four in the morning, which is the short-term one, and it will produce a do-not-drink notice in a situation WHO's own guidance says does not require one.

An action level is what your organisation does at a given reading, and it belongs to you. It is set below the guideline value by a margin that reflects your sampling frequency, your laboratory turnaround and how fast your source water can change. A utility sampling weekly needs a wider margin than one sampling daily. That is a management decision, not a scientific one.

A trigger is the observation that starts a process — a satellite anomaly, a duty operator's report of discoloured water, a cell count crossing a screening level. Triggers start testing. They do not, on their own, close anything.

Programmes get into difficulty when these three are written into the same column of the same table.

2. Drinking water

Cyanotoxins in a raw water source are the case where the administrative chain is shortest and the consequences of getting it wrong are worst. Toledo in August 2014 is the reference event: microcystin in treated water, and roughly half a million residents told not to drink from the tap.

What that event exposed was not an absence of science. It was the gap between a source-water monitoring programme run by an environment agency and a treatment decision that had to be made by a utility at four in the morning, using the last result available rather than the current condition of the intake.

Four things belong in a drinking-water annexe and are frequently missing. A named action level for raw water that is lower than the finished-water guideline. A written treatment response for each level, agreed with the regulator in advance. A standing arrangement for who may issue a do-not-drink notice outside office hours. And an explicit statement of which WHO value the notice is being judged against, lifetime or short-term, because the two are twelve-fold apart and the difference is a city's water supply.

3. Shellfish, and why cell counts are not enough

Filter feeders concentrate toxin from water in which no human would be harmed by swimming. The exposure route is the meal, not the water, and the number that matters is in the tissue.

Domoic acid is the clearest case for a directorate to understand, because its consequences are severe and its handling is counter-intuitive. Produced by Pseudo-nitzschia blooms, it causes amnesic shellfish poisoning: vomiting, diarrhoea, headache, cramps and disorientation at low doses; trouble breathing, seizures, coma, permanent loss of short-term memory and death at higher ones. Onset runs from fifteen minutes to nearly forty hours after the meal, which is why case-finding is hard. There is no specific treatment. Cooking does not destroy it and neither does freezing, and contaminated shellfish look, smell and taste normal.

The at-risk list is wider than most notices say. California's health department names older people and people with kidney disease or diabetes as being at greater risk of severe illness, and adds pregnant women, infants and children as groups who should take advice before eating shellfish at all. A notice that says only "the elderly" has under-warned three of those five.

So the closure rests on tissue testing, the tissue testing rests on a laboratory queue, and the queue is the real constraint on how quickly a fishery reopens. Cell counts of the producing organism are how you decide to test. They are rarely, on their own, how you decide to close.

This table is deliberately two rows long. It contains the toxin classes for which we can point a reviewer at a published figure or a published clinical description. Saxitoxin, ciguatera, brevetoxin and the rest are real and important; we have not tabulated numbers for them because we do not have a source we would defend, and a table of half-remembered limits is worse than no table.

4. Recreational water

Bathing advisories are the most visible part of a programme and the least standardised. Jurisdictions differ on whether the trigger is a cell count, a toxin concentration, a scum observation or an officer's judgement, and they differ on whether the notice is advisory or prohibitive.

For cyanotoxins there is at least a starting point: WHO's provisional recreational water value of twenty-four micrograms per litre, set out above. It is a health number for incidental ingestion while swimming. It is not a scum-avoidance rule, and it says nothing about the dermal and inhalation exposures that dominate complaints on a busy beach.

Beyond that we are not going to publish numbers. Marine toxins in bathing water are handled differently in almost every jurisdiction we have read, and lifting a figure from a neighbouring country without its sampling regime, its species assemblage and its legal framing is how a programme acquires a threshold nobody can defend at an inquiry.

What does travel between jurisdictions is the process: a defined observation trigger, a defined confirmation step, a named signatory, a published notice with a plain-language explanation of the risk, and pre-agreed rescission criteria.

5. The step nobody rehearses

Every programme document we have read describes how an advisory is issued. Perhaps a third describe how it comes off.

That asymmetry is expensive. A closure that outlives the hazard costs the same as one that was never justified, and it is harder to defend, because the operator can see clear water. Rescission criteria written during a season are written under pressure from people with a commercial interest in the answer.

Write them in the winter. Consecutive clear results, a minimum number of stations, a minimum elapsed time, and a named person who signs the withdrawal.

6. The decision chain, in order

1. Trigger Something is observed — a satellite anomaly, a discoloured surface, an operator's report, a routine count crossing a screening level. This step is cheap and should be generous; false triggers cost a sample, not a season.
2. Confirm A sample is taken and analysed by a laboratory whose method and detection limits are stated in the programme document. If the method is not written down, the result is not defensible.
3. Classify Which organism, at what abundance, and — for the exposure routes that need it — how much toxin in which matrix. This is where the queue lives, and where most programmes lose their days.
4. Decide Against the action level your organisation set in advance, not against the guideline value and not against the mood of the meeting.
5. Publish A notice with a named signatory, a plain statement of what is and is not affected, and the criteria under which it will be lifted. The last part is what stops the telephone ringing.
6. Rescind Against the criteria agreed before the season. Then record what the event cost, because the ledger is the only thing that ever moves a budget.

7. Where the chain runs out

Every step above is about knowing and telling. None of it changes the water.

That is the honest boundary of an advisory regime, and it is the point at which a directorate has to decide whether it is content to describe events or wants the option of intervening in them. In-water intervention is treated on the nutrient policy page as a scheduling question, because that is what it is: something that acts in weeks, sitting alongside catchment measures that act in decades.

What it does not do is remove a single line from the chain above. If anything, it adds one, because a treated water body needs verification as well as surveillance.