METHOD

Why are predictions locked before the match?

A prediction that can be changed afterwards is worthless. How the lock works in LigLab, what a "never-seen season" means, and why predictions that miss are not deleted.

LigLab Editorial Team · · 5 min read

After a match is over, everyone is right. Saying "it was obvious" is easy; what is hard is writing the same thing before the match and signing it.

LigLab's whole claim rests on that difference. Every prediction is recorded before the match, locked, and stays where it is whatever the result. This article explains how the lock works and why it is indispensable.

The problem: looking back, everything seems obvious

The human mind tends, once it knows a result, to find that result "predictable". If the derby ended 3–0, the winner's superiority looks plain in hindsight. Before the match, the same person was undecided.

For a site or app that makes predictions, that tendency is a convenience. If predictions aren't recorded somewhere:

  • the ones that land are pushed to the front and the ones that miss are quietly forgotten;
  • explanations added afterwards, like "we did mention a draw too", can't be tested;
  • the accuracy rate can be pulled wherever you like, depending on which predictions are counted.

None of this needs bad intent. Without a record, it happens by itself.

The solution: the lock

In LigLab the rule is simple: once a match has started, no prediction changes.

Each week's predictions are published and recorded before that week's matches start. After the match is played, the result is written next to the prediction. If the prediction landed it says "Correct", if it missed it says "Missed"; both stay on the same page, at the same size.

You can browse this record yourself. Every league's every matchday has its own page; for example, from the matchday archive at the bottom of the Süper Lig page you can go to past weeks. Predictions that miss are not deleted, because if they were, the count of what is left would mean nothing.

Let us state the limits of the record too. The pre-match record started on 9 October 2026 for the Süper Lig and on 30–31 October 2026 for the European leagues. The predictions for earlier weeks were calculated afterwards with the same model, using only data known before that match; under those weeks in the archive there is a note saying so. We also closed a gap: until 16 November 2026 the lock protected a prediction only until the match was played; the archived prediction for a played match could be recalculated when the model was updated. Since that date, the predictions for played matches are stored permanently too.

The same rule applies to you in the prediction game in the app: you can change your score until the match starts, then it locks. The bots follow the same rule.

The lock's second layer: never-seen seasons

A match-by-match lock keeps the daily predictions honest. But when it comes to a model's "accuracy rate", there is a subtler trap.

A model is built from past matches. If you then look at how well it predicts those same past matches, the result always comes out better than it really is, because the model has already "seen" the results of those matches; it's like sitting an exam knowing the questions.

That's why we measure our rates on seasons the model has never seen. The model is built from matches up to a certain date; the three seasons after it are set aside and opened only once, for testing. Each match is predicted using only the information from matches played before it, just as it would be in real life.

The "locked test" rates on the site are the result of this measurement. They were announced once and don't change with live data. This season's rate is shown separately, updated every day, on the Record page. We keep the two apart: one is the model's exam grade, the other is this season's course.

The third layer: writing the rule in advance

There is one more trap. When you try a new idea (say, "does the referee affect the match result?"), it is possible to choose the measure of success after seeing the result. If you look at ten different measures and pick the one that works, every idea "works".

To prevent this, we write the test rule for every new idea in advance: with what data, with what measure, and over what threshold it will be accepted. Then we run the test once and publish the result as it is.

A natural consequence is that some ideas fail the test. We write the ones that fail on the Method page too, as "failed". A method page that only tells the successes is no different from a prediction list without a lock.

What does the lock cost us?

To be honest, in bad weeks it's uncomfortable. When there is a week in which we got three of nine matches right, that week stays in the archive and in the record on our social media account, exactly as it was.

But without the lock, what we'd be left with would be a claim, not a measurement. We explained in a separate article why numbers like "87% accuracy" can't be tested; what they have in common is that the predictions weren't recorded in advance.

Three questions to ask any source of predictions

The idea of the lock isn't only for LigLab; it works for every prediction source you meet:

  • Where was the prediction written before the match? Can a date and a record be shown?
  • Where are the ones that missed? Can all past predictions be seen, or only selected examples?
  • On which matches was the rate measured? Matches the model had seen before, or matches it hadn't?

If you can get a clear answer to all three, you have a reason to trust the number.

In short

  • Looking back, every result seems obvious; that's why a prediction's value lies in being recorded before the match.
  • In LigLab no prediction changes after a match has started; the ones that miss stay in the archive.
  • Our accuracy rates are measured on seasons the model has never seen.
  • The test rule for new ideas is written in advance; ideas that fail the test are published too.

Statistical model prediction; for entertainment only, not betting advice. 18+ How this article is produced and where the numbers come from: How we work. You can add LigLab as a preferred source on Google.

Related articles

Guide
How to set up a prediction league with friends

The steps to start a friends' league in LigLab, the scoring rules, and a few tips for competing on the same table as the bots. No money, no prizes, no betting; just predictions.

16 November 2026 · 4 min read

Guide
Connect LigLab to Claude and ChatGPT: ask for the locked predictions directly

How to connect LigLab's MCP server to Claude, ChatGPT or Claude Code, what you can ask, and where the answers come from. Free, read-only, no betting data.

16 November 2026 · 3 min read

Method
How to predict a football match: the seven bots' methods

Goals, shots, form, ratings? Each of LigLab's seven prediction bots looks at a match from a different place. What does each rely on, and why are there seven bots?

16 November 2026 · 6 min read

← All articles