Two matches on 10 October 2026 started with a BetScore home-win prediction and ended differently: Fukuoka won; Gimcheon lost. Yet correctly picking Fukuoka did not make the predicted score or the supporting explanation correct in every respect.
This review places a prediction stored before kick-off, the official final result, and the limits of its original reasoning side by side. We deliberately selected contrasting cases, not a random or complete sample. They cannot establish the service's overall accuracy or return.
First check: did the prediction exist before the result?
On 13 October we checked the server creation and last-modification times of the original prediction records. All times below are Korea Standard Time (UTC+9); kick-off is the scheduled time.
| Match | First stored | Last modified | Scheduled kick-off |
|---|---|---|---|
| Fukuoka–Okayama | 3 Oct, 15:50:40 | 9 Oct, 09:20:08 | 10 Oct, 15:00 |
| Gimcheon–Anyang | 3 Oct, 16:30:41 | 9 Oct, 09:20:08 | 10 Oct, 16:30 |
Both records' last modifications preceded kick-off. This is an internal check that the currently retrieved prediction values were stored before the match. It is not an independent performance audit or a guarantee of a complete, immutable revision history.
There was a gap between the last save and kick-off. These values should not be described as forecasts incorporating confirmed starting line-ups and all last-minute information.
Grade the same questions for both matches
Scores are home team first. An outcome hit means the selected home win, draw or away win occurred. An exact-score hit requires both final goal counts to match.
| Measure | Fukuoka–Okayama | Gimcheon–Anyang |
|---|---|---|
| Stored selection | Fukuoka win | Gimcheon win |
| Predicted score | 1–0 | 2–1 |
| Official final score | 2–1 | 0–1 |
| Outcome hit | Yes | No |
| Exact-score hit | No | No |
| Home / draw / away probability | 51.5% / 32.5% / 16.0% | 52.5% / 29.0% / 18.5% |
| Separately reported confidence | 63 | 66 |
Results were checked against the official J.League match record and the official K League club report. Links to the BetScore match records appear below.
Confidence values of 63 and 66 are separate model-reported fields. They are neither the home-win probabilities of 51.5% and 52.5% nor observed historical accuracy.
Case 1: Fukuoka won, but the 1–0 score forecast missed
Fukuoka finished 2–1 ahead. The official record shows a Fukuoka goal in the eighth minute, an Okayama equaliser in the 61st and Fukuoka's next goal in the 76th. These are events checked after the match, not a sequence the model had predicted. Official match record
The stored forecast was a Fukuoka win, 1–0. The selected winner was correct, but both teams' goal counts and the total differed. Calling that a fully accurate reading of the match would overstate what was achieved.
The probabilities matter too. A home win was the largest individual option at 51.5%, but draw and away win together accounted for 48.5%. Displaying only the leading choice can make the favourite look stronger than the full distribution suggests.
The supported conclusion is that the first-choice outcome matched the result. Exact-score precision, long-run probability quality and profitability remain separate questions.
Case 2: Gimcheon's 52.5% home-win forecast met a 0–1 defeat
The record assigned 52.5% to a Gimcheon win, 29.0% to a draw and 18.5% to an Anyang win. Anyang won. Neither the leading selection nor the forecast score of 2–1 matched the result. Official K League result
A probability of 52.5% does not make a defeat impossible. Equally, saying that probabilistic forecasts sometimes fail cannot end the evaluation. We need enough comparable forecasts to ask whether events assigned similar probabilities occur at roughly those frequencies.
One defeat cannot establish that the whole model is wrong. But one case can still reveal whether its supporting information was appropriate.
A shared problem: the age of the evidence
Both original explanations disclosed an important limit. Fukuoka's referred to data through 2024; Gimcheon's said it was based on 2024 data. Fukuoka's explanation also acknowledged a lack of current 2026 team-strength and form information. Neither stored prediction included entries in its reference list.
We do not repeat those explanations as verified descriptions of the teams at the time. The quality issue is that the current information used to explain a 2026 match cannot readily be traced. An empty reference list does not prove every claim false, but it leaves the reader without a reliable route to check it.
For Gimcheon, a club report published on 20 September, before the prediction was generated, had already discussed the tenth intake's final match and military discharge. Such roster changes are relevant context to check before applying older home-performance claims. Our investigation does not establish that this change caused the 0–1 defeat. Information available before the prediction
The same standard applies to Fukuoka's correct outcome pick. A successful result does not repair stale supporting evidence. Scrutinising misses while excusing hits risks mistaking a fortunate outcome for sound analysis.
Four notes to keep before the next match
- What was predicted? Separate the match outcome, exact score and total goals.
- When was it saved? Check whether the record predates the game and whether it was later changed.
- How current is the evidence? A familiar team name can hide changes in personnel, management and competition.
- What is still unverified? Leave unsupported claims, missing line-ups and unchecked variables explicitly unresolved.
After the match, add the result to the original notes. Do not insert later injury information or tactical explanations as though they had been known at the time. Our guides to recent form and small samples provide further context.
Records and scope
Checked on 13 October 2026. Prediction values and save times reflect the original records retrieved that day; official result sources are linked in the text. Some additional analysis on match pages may require login, but the comparison in this article is fully readable without signing in. The cover is an AI-generated conceptual illustration, not a photograph of these matches or a statistical chart.
▶
▶
▶