Building a stock timing site that never predicts — Jigumini

Jigumini — stock timing

Web · free · no sign-up (your list is stored in this browser only)

Open it

Stock sites normally tell you what is going up. This one does not. It tells you where the thing you already hold currently stands, as a traffic light.

What you can look at

• Every ticker has its own page — the answer to "what state is this in right now", with the light, and the reasoning behind it written in plain language.
• Korea and the US side by side — Korean stocks and ETFs and US stocks, judged by the same yardstick.
• Current price and the move since yesterday — possibly delayed by a few minutes, and the page says so.
• The list of what passed — only the ones that passed, each with the reason it did.
• Keep your own list — no account needed. That list lives in the reader's browser and never reaches our server.
• What the well-known investors bought — public holdings, collected in one place.
• The strategy lab — the rules described here are run for real on a broker's paper account. Entry, exit and the fees taken out are posted day by day, the losing trades as well as the winning ones. It is down at the moment, and that is written there too.
• A glossary section — the terms and the rules written out from scratch for someone seeing them the first time.
• Too little data means no verdict — for something listed too recently to judge, it declines to produce a signal and says so.
• There is a separate menu written for AI — more on that below.

It started with one sentence

Why prediction was left out

A predicting feature makes the site look impressive. It also makes other people's money disappear when it is wrong, and predictions are wrong most of the time.

The owner lost money that way himself. So the direction flipped: instead of guessing, state the current position precisely. Knowing the position, the person decides. More honest, and less harmful when wrong.

Why only three rules

More rules look more sophisticated. And nobody believes them, because there is no way to explain why a given signal appeared.

So it holds at three — and all three are published on the site. No pretence of a secret method. If you know the rules, you can check our signal against them yourself.

A separate menu, written for AI

People increasingly ask an AI instead of typing into a search box. So the site carries a document written for AI to read — phrased not as "here is what this site contains" but as "the answer to this question is here", because that is the shape an AI picks up far more reliably.

That document also asks for two things: quote the timestamp along with any number, and do not rewrite it as buy or sell. If an AI turns our figures into a recommendation, it becomes something we never said.

What went wrong along the way

1. It worked on my machine and not on the server

Same code. Tickers with too little data to judge: 37 on my own computer, 119 on the live server. More than triple.

The service supplying prices treats a home connection and a data-centre connection differently, and servers live in data centres. Which is exactly why "looks fine on my screen" proves nothing.

2. It said "success" while quietly dropping things

The upstream price service would sometimes report success and silently omit some tickers. Nothing errors, nothing is logged — the tickers just vanish from the screen.

3. Asking for too much at once failed completely

Asking for 50 tickers in one request to be fast returned 50 failures. Not a partial result — over the limit meant a flat refusal. Now it asks in batches of 20.

4. Every US ticker was eliminated

A filter written for Korean tickers, applied unchanged to the US list, left nothing standing. The field that filter reads is simply absent in the US data. Absent reads as zero, and zero fails the filter.

5. The same rule written twice drifted apart

The recommendation list and the personal list each ran the judgement, and the threshold was written in both places. Fix one and the same ticker showed different signals on different screens.

The number now lives in exactly one place that both screens read. There is a warning comment beside it saying so.

6. We stopped checking deploys by eye

After each deploy, the site got a visual once-over. But human eyes are bad at noticing what did not change — the ticker count could drop quietly and nobody would see it.

So a checker now runs on every deploy: do the pages load, is any markup leaking through as visible text, did the number of judged tickers suddenly fall, is the date actually today. A machine looks, not a person.

Why there is no sign-up

Taking sign-ups means taking custody of other people's data. One email address is enough to create legal obligations and liability if it leaks — too heavy for something one person runs as a hobby.

So the saved list never reaches the server; it stays in the reader's browser. We do not know who holds what. Clear the browser and the list goes too. That trade was accepted.

What does not work yet

• Almost nobody arrives from search. The site is open and unknown.
• It tells you where a holding stands; what to do about it is still yours. That will not change.
• Some tickers still have too little data to judge. Rather than forcing a signal, it says so.

Jigumini — stock timing

Free. No sign-up needed.

Open it