Your wine rack as a data file: what should we open tonight?

Part 1 put your taste into a project profile. That answers the question of what to buy. There is a second question that costs people more money: what is already sitting there, and what are you drinking too late?

Bottles forgotten in a rack are the quietest way to throw money away. Not because wine spoils, exactly, but because the window in which a bottle is at its best does eventually close.

The job

I wanted one place that says what is in the rack, in a form a model can read. No app with a subscription, no scanner. A text file, because a text file outlives every tool.

Twelve lines is enough to start. The craft is in which columns you pick.

My actual prompt

First the file. A CSV with six columns, nothing more:

name,producer,year,type,paid,window_note
Classic Brut,Domein Holset,2022,sparkling,24,bought at the estate
Assyrtiko,Artemis Karamolegos,2023,white,18,maker says better after two years
Brunello,La Fortuna,2021,red,45,tasted closed

That last column is the important one, and the one everybody leaves empty. Put in what you noticed yourself or heard from the grower, in your own words. That is the information no database has.

Then the prompt, with the file attached in the project:

This is my wine rack. Use my taste profile from the instructions.

Tonight I'm eating [dish] with [number] people.
Which bottle from this list should I open, and why that one?
Give the serving temperature.
Also tell me which bottle to leave alone, and until when.
If you don't know a drinking window, say so instead of guessing.

That last line is the same hard rule from part 1, and this is where it pays off fastest. Drinking windows are exactly the kind of fact a model invents with great confidence.

Taken apart

The transferable pattern has nothing to do with wine: give the model a list of what you already have, not just a question about what you want.

That changes the kind of answer you get. Without a list you get advice drawn from the general stock of the internet. With a list you get a choice out of your own reality, and that is a choice you can actually act on.

The same works for your bookshelf, your tools, your pantry, your record collection. Three things make such a list usable.

Few columns. Six is plenty. Every extra column you invent is a column you will stop maintaining.

One column for your own observation. The model roughly knows the facts. What it cannot know is that you found that bottle closed last year.

Plain text. CSV or markdown, not some app’s export format. You need to be able to open it in five years.

Where it gets useful fast is the reverse question. Not “what do I drink tonight” but “what has been sitting too long”. Ask which three bottles from the list are up first and why, and you get an order instead of a suggestion. That is exactly the list you never make yourself, because you always reach for whatever is at the front.

What broke

Two things you will almost certainly hit.

The model does arithmetic with drinking windows it does not have. Ask about a wine it doesn’t know and out comes a tidy five-year window resting on nothing. That is what the last line of the prompt is for, and even then you have to watch it.

And it steers towards the most expensive bottle. Ask for something for an ordinary evening and it reaches for the Brunello, because that is the wine with the most words attached to it. So state the occasion, and state what you are willing to open.

Ethics note

A rack list says more about you than you would think. What you buy, what you spend, how fast it goes. That is precisely the profile insurers and advertisers would love.

So keep the list local as the master file and decide each time what you send along. To answer what to open tonight, a model does not need the purchase price. You can drop that column from what you paste.

The second consideration is more mundane but costs real money: a model that suggests is a model that gets you buying. This prompt is deliberately built the other way around. It chooses from what you already own.

Take it

This is my [stock / rack / shelf] as a list.

[paste the CSV]

Columns: name, origin, year, type, [your own column], my own observation.

Question: I have [situation] and I want [goal].
Pick one thing from this list and explain why that one.
Also tell me what to save, and until when.
If you are unsure about something, say so instead of guessing.

Starting exercise: write down twelve lines of what is in your rack right now. Do not try to make it complete. That is the trap. Twelve is enough to open a better bottle tonight than you otherwise would have.

What happens in the glass after that is a different craft. I write about that at VinoVonk.