I counted, which in retrospect was a mistake because the number is dispiriting. Eleven attempts at a milk-reduced sweet before one came out the way it was supposed to.
The failures were not random. They were the same failure repeated, and it took me an absurdly long time to see the pattern.
The failure was always heat
Every unsuccessful attempt had the same cause. I ran the heat too high because reducing milk takes a long time and I was impatient.
What happens at too high a heat is that the milk solids at the base of the pan cook faster than the rest. They catch, they brown, and then that browned flavour distributes through everything.
The result is not obviously burnt. It is subtly wrong — a slightly bitter background note that you notice on the second mouthful and cannot get rid of.
The tenth attempt taught me this because I did it on a genuinely low flame while doing something else in the kitchen, and it worked, and I assumed it was luck.
How long it actually takes
Reducing a litre of milk by two thirds, on a heat low enough not to catch, in a wide heavy pan, takes about an hour.
Not the twenty five minutes I have seen written. An hour, sometimes more, and there is no shortcut that does not affect the result.
The pan matters enormously. Something heavy distributes heat and gives you margin. A thin pan concentrates heat at the hot spots and will catch no matter how careful you are.
Width matters as much as weight, because evaporation happens at the surface. A wide shallow pan reduces significantly faster than a tall narrow one at the same heat.
The stirring question
Received wisdom is constant stirring. I now think that is close to wrong.
What matters is scraping the base, not agitating the whole thing. The risk is solids settling and catching, which happens at the bottom.
So: a flat-edged implement, dragged across the base every couple of minutes, going into the corners. Between those passes, it can sit.
Constant stirring incorporates air, cools the surface and makes the whole hour more miserable than it needs to be. Periodic scraping does the actual job.
The skin, and what to do with it
A skin forms and people stir it back in, which is what I did.
Stirring it in produces lumps that never fully disperse and that end up as slightly rubbery bits in the finished sweet.
What worked better was either lifting it off and discarding it, or lifting it off and folding it back at the very end when the mixture is thick enough to absorb it evenly.
Some traditions deliberately collect the skins as their own preparation, which suggests people worked this out a long time ago.
Sugar late
Adding sugar early is the second most common mistake and it caused three of my eleven failures.
Sugar in the pan for an hour will caramelise, and beyond a certain point it will catch. It also changes the viscosity, which means the milk behaves differently and reduces less predictably.
Sugar goes in near the end, once the reduction is where you want it. The mixture will loosen when you add it and then thicken again as the sugar dissolves and some water cooks off. That loosening frightened me the first time and is entirely normal.
Knowing when to stop
The hardest judgement and the one that only comes with repetition.
The mixture continues to thicken as it cools, sometimes substantially. Something that looks right in the pan will be too stiff on the plate.
The test I use is that it should just about hold a line when you draw the spoon across the base, and that line should close again slowly. If the line stays open, it is already too far.
Undershooting is recoverable — you can cook it a little longer. Overshooting is not, and produces something dry and grainy.
Why it is worth the hour
An honest question given that these things are available in every sweet shop.
The shop version is generally sweeter and frequently made faster with milk powder or with additions that shorten the process. It is fine and it is a different thing.
What an hour of slow reduction produces is a caramelised, nutty depth that comes from milk sugars slowly transforming, and there is genuinely no way to get there quickly.
I make it perhaps four times a year, which feels about right for something that occupies an entire afternoon.
What I would tell someone attempting it
Use full fat milk, and do not substitute. The whole thing depends on the solids and the fat, and a reduced fat version produces something thin and chalky no matter how long you cook it.
Start with more milk than the recipe suggests, because everybody underestimates how much reduction is needed.
Do it on an afternoon when you are in the kitchen anyway. Trying to fit it around going out is how the heat gets turned up.
And accept that the first attempt will probably not be right. Eleven is an unusual number of failures and two or three is normal, and the reason is that the judgement of when to stop can only be learned by having stopped in the wrong place.