Why this notebook exists

Start from the analyst’s actual problem

Competitive preparation is full of partial evidence. A replay shows actions, but not every intention. A statistical summary can reveal repeated behavior, but it may hide context. A model can organize information at scale, yet still inherit the limits of the records it receives. This site exists to help readers move through those layers without mistaking one for another.

Our editorial approach is fieldnote-first. We begin with the observable event, ask what pattern the evidence supports, and then consider what a team could reasonably do with that knowledge. We favor conditional claims, explicit uncertainty, practical examples, and methods another analyst can challenge. We do not present AI as an oracle or a replacement for coaching judgment.

Who the site is for

The intended readers are coaches, analysts, players, and curious readers who want to understand how opponent behavior can be studied responsibly. Some may work with large collections of match footage; others may have only a few useful replays and limited review time. The principles remain similar: define the question, retain context, compare fairly, and make the resulting note understandable to the people who need it.

Our articles are educational and explanatory. They discuss analytical workflows and decision-making, rather than promising that any particular technique will improve a team’s results. Competitive outcomes depend on many factors, including execution, communication, opponent adaptation, and chance. A well-made analysis can inform a decision; it cannot control the match.

How we frame claims

We distinguish observation from inference. We try to describe what a reader could locate in replay before suggesting why a team may have acted. We also call out the denominator, sample limits, counterexamples, and conditions that may change a tendency. Where an example is hypothetical, we label it as such rather than implying it documents a real match.

We value accessible explanations over technical theater. Statistical and machine-learning methods can be useful, but a complicated model is not automatically more reliable. We encourage readers to compare methods with clear baselines, check performance on data not used to discover a pattern, and test whether the output improves an actual analytical workflow.

What “useful” means here

A useful opponent read helps a team pay attention to a relevant signal and preserve good options. It says what has been observed, under which conditions, and what would make the interpretation stronger or weaker. It supports a plan with contingencies. If a note encourages certainty without evidence, it needs more work.

The site is arranged as a sequence: replay evidence, tendency models, tactical preparation, and model validation. Those topics are connected, but they are not interchangeable. Clear evidence does not guarantee a sound model; a sound model does not guarantee a useful plan; and a useful plan still needs review after the match.

Our commitment to careful reading

We aim to write for thoughtful peers. That means avoiding inflated claims, respecting the limits of available information, and making the analytical path legible. Readers should be able to ask where a statement came from, what assumptions it relies on, and how they might test it in their own context.

If you notice an unclear explanation or want to suggest a topic, use the contact form. You can also browse the analyst desk questions or start with the replay evidence guide.