Cellar records rarely die when you set them up, because that is the fun part. They die in the upkeep. You buy two bottles, you are not at your desk at that moment, and three weeks later the list no longer matches the shelf.
The text file has the same problem. It is only easier to repair.
What you want
Take the data entry out. One action at the moment it happens: a photo of the receipt or the label, into a folder. Everything else happens afterwards, without you.
This is work that runs over weeks with files trickling in. A project with a standing instruction suits that better than a chat, because a chat starts over every time, and starting over is exactly the action that has to go.
The instruction
This lives in the project’s instructions, not in a one-off conversation.
Photos of receipts and labels arrive in inbox/.
For every new photo:
1. Read what you can read with certainty: name, year, count, price, shop.
2. Leave anything you are unsure about empty. Do not fill in and do not
guess from the name.
3. Add a row to kelder.csv with the columns name, year, style, count,
bought, price, source.
4. If you cannot determine the style from what is legible, leave the column
empty and put the row at the bottom of openstaand.md with the reason.
Move processed photos to verwerkt/. Never delete anything.
If a bottle is already in kelder.csv, add to the count instead of adding a
second row. Ask me when you are unsure whether it is the same one.Once a week you ask one question yourself: what is in openstaand.md and what do I need to decide about it? That is the only moment you actually take part.
Taken apart
Put the doubt in a separate file. This is the core of the whole setup. An incomplete row does not go half into your records, it goes onto a list of open items. That keeps the main file trustworthy, and it makes uncertainty visible instead of letting it dissolve into an empty cell.
Move, don’t delete. As long as the photos are still there, any mistake traces back to its source. That is the difference between records you trust and records you re-audit every quarter.
Add to the count instead of adding a row. Without that rule the list fills up with duplicates of the same wine, which is the mess that kills these systems.
One fixed moment where you decide. Automation that never asks you anything is automation making choices quietly. One question a week is enough, and it is honest work: if nobody reads the open items, the file is a bin within a month.
Where it goes wrong
Two things are structural, and you can design around them from the start.
A supermarket receipt is less legible than a label. It says “R WIJN ITAL 75CL” and nothing else: no year, no producer. Any attempt to match that against what is already in the list is exactly the kind of helpfulness that wrecks a set of records. Rows like that belong in the open list, unchanged.
And a label photo gives you the name but not how many bottles you bought. That is on the receipt, and nothing ties the two photos together except the day you took them. The date in the filename works, right up until you visited two shops in one day.
The third is not a fault but an expectation worth adjusting. This saves data entry, not thinking. The expensive part remains deciding what to do with the open list, and that stays manual. What you gain is that the records no longer collapse on the day you cannot face them.
Ethics note
A receipt holds more than wine. It shows where you were, at what time, which card you paid with and what else you bought. Put that photo in a project and all of it goes along. Crop above the first line, or photograph only the wine rows.
Then the subject itself. A system that tracks exactly what you buy also tracks exactly how much you drink. That can go two ways. Seeing the numbers is clarifying for some people, and for someone with a strained relationship with alcohol a counter in the house is not automatically helpful. Know which of the two you are before switching this on.
Take it
This works for any records that demand data entry: expenses, books, tools, parts.
Photos or files arrive in [folder].
For each new file: read what you can read with certainty.
Leave anything uncertain empty. Do not fill in, do not guess.
Add a row to [file] with the columns [columns].
If you cannot determine [the awkward field], put the row in openstaand.md
with the reason.
Move processed files to verwerkt/. Never delete anything.
If the row already exists, add to it instead of appending a new one.A small exercise this week: don’t look at your records, look at your open list. If it is empty and you have never worked through it, something is filling gaps.