luxury-lakehouse/obso-pausa-values

OBSO/PAUSA Values — Pass Timing and Target Quality

Per-pass Off-Ball Scoring Opportunity (OBSO) and PAUSA scores measuring pass timing quality and spatial target selection. Each row captures the actual, peak, and optimal OBSO at the moment of a pass, plus two derived ratios: temporal judgment (did the passer release at the right moment?) and spatial selection (did the passer choose the best available target?).

Part of the (Right! Luxury!) Lakehouse soccer analytics platform.

Quick Start

from datasets import load_dataset

ds = load_dataset("luxury-lakehouse/obso-pausa-values")
df = ds["train"].to_pandas()

# Top 10 passes by temporal judgment (best-timed releases)
best_timed = df.nlargest(10, "temporal_judgment")[
    ["match_id", "player_id", "temporal_judgment", "spatial_selection", "actual_obso"]
]
print(best_timed)

Explore interactively: Soccer Analytics App

What Is This Dataset?

OBSO (Off-Ball Scoring Opportunity) quantifies the goal-scoring threat created by off-ball positioning at the moment a pass is released. The model computes pitch control over a grid, multiplies by expected possession value (EPV), and sums the resulting surface to produce a single scalar threat value.

PAUSA ("La Pausa") extends OBSO by evaluating pass timing and target selection:

• Temporal judgment = actual_obso / peak_obso — how close to the optimal release moment the passer acted. A value of 1.0 means the pass was released at the exact peak of opportunity.
• Spatial selection = actual_obso / optimal_obso — how well the chosen target compares to the best available teammate. A value of 1.0 means the passer selected the highest-value target.

These metrics are described in Lee, Jo, Hong, Bauer, & Ko (2026), building on the pitch control and EPV frameworks of Spearman (2018) and Fernandez & Bornn (2018).

Data Fields

Column	Type	Description
match_id	string	Match identifier
pass_id	string	Composite key uniquely identifying the pass
event_id	string	Source event identifier
player_id	string	DFL PersonId of the passer
team	string	Team affiliation (home or away)
period	int	Match period (1 or 2)
timestamp_seconds	double	Elapsed time in seconds from period start
frame_id	int	Tracking frame at the moment of the pass
ball_x	double	Ball x-coordinate at release (StatsBomb 120-yard scale)
ball_y	double	Ball y-coordinate at release (StatsBomb 80-yard scale)
receiver_x	double	Target x-coordinate (StatsBomb 120-yard scale)
receiver_y	double	Target y-coordinate (StatsBomb 80-yard scale)
actual_obso	double	OBSO at the moment of pass release
peak_obso	double	Maximum OBSO observed in the ghost trajectory window
optimal_obso	double	Maximum OBSO across all teammate target locations
temporal_judgment	double	Timing quality (0–1, higher = better; actual/peak)
spatial_selection	double	Target quality (0–1, higher = better; actual/optimal)
alignment_confidence	double	ELASTIC sync confidence (0–1) for the event-tracking alignment

Coordinate System

All coordinates use the StatsBomb 120×80 yards scale. The origin (0, 0) is at the bottom-left corner of the pitch. The pitch control grid resolution is 104×68 cells. Ghost trajectory window: 3.0 seconds before to 1.0 second after the event, sampled at 25fps.

Data Sources

Source	Matches	License
IDSSE Open Data	7	CC-BY 4.0

Computed from 7 IDSSE tracking matches via ELASTIC event-tracking synchronization. Event coordinates are normalized to the StatsBomb scale during the OBSO pipeline.

PR 7 (ADR-013 second application — 2026-04-27): the gold mart fct_pausa_values is now built by dbt with contract: enforced: true from bronze.pausa_values (the writer src/ingestion/pausa.py retargets bronze; previously wrote gold directly). The mart inherits Kimball surrogate FKs (match_key, team_key, player_key) via INNER JOIN fct_passes ON pass_id. The HF dataset payload itself remains as-published by scripts/compute_obso_hf.py (a pre-mart compute output) and does not carry surrogate keys — per ADR-013, ML inference outputs are joined back to identity facts at the mart layer. Consumers querying the lakehouse should join fct_pausa_values directly; consumers reading this HF dataset should join via pass_id to the published line-breaking-passes dataset (which carries match_key/team_key/passer_player_key post-PR-7).

Companion Resources

Resource	Description
OBSO/PAUSA Inputs	Source events and ELASTIC sync used as inputs
OBSO Trained Grids	Reachability, EPV, and completion grids powering the OBSO model
Pitch Control Tracking	Per-player per-frame pitch control from tracking data

Limitations

• IDSSE matches only: 7 Bundesliga matches from the IDSSE open data release. The sample is too small to draw league-wide conclusions.
• ELASTIC alignment accuracy: OBSO values depend on correct event-tracking synchronization. Events with low alignment_confidence may have pitch control computed at the wrong frame.
• Relative scores: PAUSA temporal judgment and spatial selection are ratios, not absolute quality measures. A temporal judgment of 0.8 means 80% of the peak opportunity was captured — whether 0.8 is "good" depends on the difficulty of the pass context.
• Ghost trajectory assumption: The ghost window assumes players continue their current velocity for 3 seconds before the event. Sudden direction changes or stops are not modeled.
• No opponent intent modeling: Pitch control treats defenders as physics objects. Tactical intent (e.g., deliberate pressing traps) is not captured.

Citation

If you use this dataset, please cite the following:

@inproceedings{lee2026pausa,
  title={Valuing La Pausa: Quantifying the Timing and Quality of Soccer Passes Using Off-Ball Scoring Opportunities},
  author={Lee, Minho and Jo, Hyunsung and Hong, Seungwon and Bauer, Pascal and Ko, Sangkuk},
  booktitle={MIT Sloan Sports Analytics Conference},
  year={2026}
}

@inproceedings{spearman2018beyond,
  title={Beyond Expected Goals},
  author={Spearman, William},
  booktitle={MIT Sloan Sports Analytics Conference},
  year={2018}
}

@inproceedings{fernandez2018wide,
  title={Wide Open Spaces: A statistical technique for measuring space creation in professional soccer},
  author={Fernandez, Javier and Bornn, Luke},
  booktitle={MIT Sloan Sports Analytics Conference},
  year={2018}
}

@inproceedings{kim2025elastic,
  title={ELASTIC: Event-Tracking Data Synchronization in Soccer Without Annotated Event Locations},
  author={Kim, Hyunsung and Theis, Fabian and Kim, Hanjun},
  booktitle={ECML-PKDD MLSA Workshop},
  year={2025},
  eprint={2508.09238},
  archiveprefix={arXiv}
}

@article{bassek2025idsse,
  title={An integrated dataset of spatiotemporal and event data in elite soccer},
  author={Bassek, Manuel and Weber, Henrik and Rein, Robert and Memmert, Daniel},
  journal={Scientific Data},
  volume={12},
  pages={283},
  year={2025},
  publisher={Nature Publishing Group}
}

More Information

Explore interactively: Soccer Analytics App

• License: MIT
• Publish script: scripts/compute_obso_hf.py