Live data · updated weekly
Does the research still hold?
The original analysis is frozen. New seasons are added separately to see whether the patterns continue.
- Last updated
- 5 Oct 2026 · 21:10 IST
- Latest match
- 20 Sep 2026
Frozen
2015/16 — 2024/25
The published research. Never changes.
Live
2025/26 complete · 2026/27 early season
New seasons, estimated the same way and shown beside it.
Data may be stale
Last successful update: 5 Oct 2026. The latest season data is not currently available.
Partial update
1 2026/27 match not processed in this snapshot (La Liga: raw match file not cached this week).
The weekly run reported 250 2026/27 matches; this snapshot has 249.
Historical baselineFrozen
2015/16 — 2024/25, ten seasons. Fixed at publication · manifest checksum 4dae6a51f310.
2025/26Complete
Season progress · 100% of fixtures played
- Matches processed
- 1,752
- Matches through
- 24 May 2026
- Last updated
- 5 Oct 2026
By league
- Premier League380/380
- La Liga380/380
- Serie A380/380
- Bundesliga306/306
- Ligue 1306/306
Processed / fixtures · played share from the weekly Understat fetch
2026/27Early seasonData may be stale
Season progress · 14% of fixtures played
- Matches processed
- 249
- Matches through
- 20 Sep 2026
- Last updated
- 5 Oct 2026
By league
- Premier League50/380
- La Liga68/380
- Serie A50/380
- Bundesliga36/306
- Ligue 145/306
Processed / fixtures · played share from the weekly Understat fetch
Research vs reality
The research made a set of claims. New seasons give us a chance to see whether those claims survive.
Claim 01 · Home advantage
The research claimedPlaying at home increases attacking output.
+26.6%Frozen historical baseline · xG · home vs away, same player
+27.1%Complete 2025/26 · xG · home vs away, same player
Historical baseline +26.6% · this season's 95% interval +23 to +32
+22.0%Early seasonData may be stale 2026/27 · xG · home vs away, same player
Historical baseline +26.6% · this season's 95% interval +11 to +34
The 2025/26 home xG effect (+27.1%) is in line with the historical baseline (+26.6%, inside this season's 95% interval). 2026/27 so far (early season): +22.0%, in line with the historical baseline. Provisional — its interval is wide and will narrow as matches are added.
xG home edge · baseline vs live seasons
Historical baseline 2015/16 — 2024/25 (◇) · live seasons with 95% intervals. Live seasons never change the baseline.
All five metrics
| Metric | Baseline | 2025/26Complete | 2026/27Early season |
|---|---|---|---|
| xG | +26.6% | +27.1%+23 to +32 | +22.0%+11 to +34 |
| xA | +24.7% | +27.0%+22 to +32 | +17.4%+6 to +30 |
| Shots | +21.7% | +22.9%+20 to +26 | +18.1%+11 to +26 |
| Key passes | +20.8% | +22.3%+19 to +26 | +14.6%+8 to +22 |
| Goals | +23.8% | +22.6%+16 to +30 | +28.7%+9 to +52 |
Show all historical seasons
The frozen outputs have no season-by-season home estimate — Study 2 split the decade by crowd status instead, so that is what is shown here rather than invented single seasons.
- With fans, before 2020+27.7%+26 to +30
- Behind closed doors+11.8%+8 to +16
- Fans back+24.9%+22 to +28
- Frozen+26.6%with fans, pooled
- 2025/26+27.1%complete · +23 to +32
- 2026/27+22.0%early season · +11 to +34
Study 02 · same player, home ÷ away xG per minute
Claim 02 · Opposition
The research claimedStronger opponents suppress attacking output.
−13.8%Frozen historical baseline · xG · per +1 SD of opponent strength
−11.4%Complete 2025/26 · xG · per +1 SD of opponent strength
Historical baseline −13.8% · this season's 95% interval −13 to −9
−13.3%Early seasonData may be stale 2026/27 · xG · per +1 SD of opponent strength
Historical baseline −13.8% · this season's 95% interval −19 to −7
The 2025/26 opposition effect on xG (−11.4%) is weaker than baseline (−13.8% sits outside this season's 95% interval, −13 to −9). 2026/27 so far (early season): −13.3%, in line with the historical baseline. Provisional — its interval is wide and will narrow as matches are added. Consistency, not proof.
xG opposition effect · baseline vs live seasons
Historical baseline 2015/16 — 2024/25 (◇) · live seasons with 95% intervals. Live seasons never change the baseline.
All five metrics
| Metric | Baseline | 2025/26Complete | 2026/27Early season |
|---|---|---|---|
| xG | −13.8% | −11.4%−13 to −9 | −13.3%−19 to −7 |
| xA | −15.1% | −12.6%−15 to −10 | −11.3%−17 to −5 |
| Shots | −10.4% | −10.6%−12 to −9 | −12.3%−16 to −8 |
| Key passes | −11.4% | −11.5%−13 to −10 | −10.3%−14 to −6 |
| Goals | −15.8% | −13.4%−16 to −10 | −22.3%−31 to −13 |
Show all historical seasons
Opposition effect on xG, season by season
Each frozen season (Study 3), the pooled frozen estimate, then the live seasons. 95% intervals.
Single frozen seasons range from −16.1% to −11.8% per SD — the normal year-to-year variation. Live: 2025/26: −11.4%; 2026/27: −13.3%.
Study 03 · per-season models (frozen) · live tracker
Claims 03 – 04 · Player level
Two predictions about individual players
A live season appears here only once the pipeline has re-run the test. The live tests compare a player's frozen-period record with the new season, so the method differs slightly from the frozen test; each row says which.
Claim 03 · Home specialists
Past home edge does not predict future home edge.
- HistoricalFrozenr = 0.018595 players · raw home ÷ away xG/90 per player, 2015/16–2018/19 vs 2022/23–2024/25 · near zero
- 2025/26Completer = 0.0111,155 players · frozen-period home edge vs 2025/26 · near zero
- 2026/27Early seasonData may be staleNot yet recalculated for the live dataset.Not enough players with a full home and away sample yet.
Study 02 · out-of-sample check (frozen) · live tracker
Claim 04 · Elite opponents
Overall adjusted level predicts elite-opponent output better than past elite-opponent record.
- HistoricalFrozen0.84 vs 0.752,047 players · odd vs even seasons; elite = top 20% of opponents
- 2025/26Complete0.76 vs 0.68419 players · frozen-period record vs 2025/26 · overall level higher in 4 of 4 metrics
- 2026/27Early seasonData may be staleHistorical result · live validation pendingNot enough elite-opponent minutes yet.
xG correlation with elite-opponent output: overall adjusted level vs past elite record · Study 03 · live tracker
Summary
What's changed?
This is the baseline itself. Choose a live season to compare it with.
Historical baseline → 2025/26 (complete)
Home xG
+26.6% to +27.1%
Broadly stable
Frozen estimate inside this season's 95% interval
Opposition
−13.8% to −11.4%
Weaker than baseline
Frozen estimate outside this season's 95% interval; outside the range of single frozen seasons (−16.1% to −11.8%)
Historical baseline → 2026/27 (early season)
Home xG
+26.6% to +22.0%
Early · Broadly stable
Frozen estimate inside this season's 95% interval
Opposition
−13.8% to −13.3%
Early · Broadly stable
Frozen estimate inside this season's 95% interval; inside the range of single frozen seasons (−16.1% to −11.8%)
Labels come from fixed rules, not judgement: broadly stable = the baseline sits inside the season's 95% interval; higher / lower (home) or stronger / weaker (opposition) = it sits outside; insufficient evidence = the interval includes zero; early = the season is in progress.
Frozen research protection
The historical baseline is immutable.
New seasons are evaluated separately and do not change previously published estimates. Every frozen number on this page is read from a fixed manifest; each weekly rebuild recomputes it from the frozen outputs and stops if anything has moved.
Manifest checksum 4dae6a51f310b401 · fixed 6 Oct 2026 · corrections are published in the findings, never applied silently.
Research status
Where each study stands
- 01Environment
- Not yet frozen · its own freeze checklist is still open
- Crowd results final
- Five-league weather models
- 1,000-permutation temperature placebo, pooled and per league
- Final weather conclusion
- Freeze: git tag, checksum manifest, pinned packages
- 02Home advantage
- Frozen · complete on the frozen 2015/16 — 2024/25 data
- 03Opposition
- Frozen · complete on the frozen 2015/16 — 2024/25 data
- 04Similarity
- Frozen · complete on the frozen 2015/16 — 2024/25 data
- Known-pairs check pending
- 05Transferability
- Frozen · complete on the frozen 2015/16 — 2024/25 data
- League-effect improvement planned
From each study's findings file · open items stay visible until the research closes them
Open questions
What are we investigating next?
Temperature
The small temperature effect survived the placebo check, mostly in Serie A. What does the reviewed final conclusion make of it?
Similarity validation
Do known similar-player pairs behave as expected?
Transfer model
Can a percentage-based league effect improve predictions for low-output central midfielders moving into the Premier League?
From the research's open items · nothing here is a result yet
Update log
This is maintained
- Study 1: five-league weather models finished
- Study 1: 1,000-permutation temperature placebo, pooled and per league finished
- 2026/27 live data updated · matches through 20 Sep 2026
- Weekly pipeline run completed
- Correction published: Understat's match “forecast” is post-match (never used in any model)
- 2025/26 season complete · last match processed
Events recorded by the pipeline (weekly run logs, snapshot timestamp) and dated entries in the findings files
Data provenance
- Live data
- Updated weekly (Monday pipeline → validation → snapshot → this page)
- Frozen baseline
- 2015/16 — 2024/25
- Current seasons
- 2025/26 · 2026/27
- Odds
- market-average closing (AvgCH/AvgCD/AvgCA) for live seasons; correlation with the frozen ratings 0.9997 on 10,705 matches
- Sources
- Understat · football-data.co.uk (pre-match odds) · fixturedownload.com · openfootball