Brief 01 / Player decisions
Reading the decision tree
Judge the choice at the moment it was made. Reconstruct the available information before the scoreline makes one route look inevitable.
Method brief / Conceptual examples
Support available?
Time still available?
A decision is not the same thing as the action visible on screen. Moving through a doorway may express a deliberate timing, a response to a teammate, or an incomplete understanding of the map. AI-assisted review becomes useful when it helps organise those possibilities rather than assigning a confident motive. The starting point is a bounded question: given the information reasonably available, which options deserved consideration?
Freeze before the result
Choose a timestamp before the decisive interaction. Record the player's viewpoint, the objective state, the remaining time and the resources that could actually be used. In Counter-Strike 2, that might include utility, equipment and a recent sound cue. In VALORANT, ability availability and revealed positions may matter. In League of Legends, vision, wave state and objective timing create different constraints. These are separate decision environments, not interchangeable examples of a universal score.
Keep observer knowledge in another column. A spectator may see an opponent behind cover; the player may only have a last known position. Reviewing with unrestricted vision can quietly convert uncertainty into an apparent mistake. If the replay cannot reconstruct what was visible or audible, mark the gap rather than filling it with hindsight.
Build branches a player could take
A useful alternative must be feasible at that moment. “Wait for support” is not meaningful if the objective timer makes waiting impossible. “Take another route” is weak advice if that route was inaccessible or unreasonably exposed. Name the resource cost, the timing cost and the assumptions behind each branch. Two or three defensible options usually support a clearer discussion than a sprawling tree of imagined perfect actions.
| Layer | Record | Avoid assuming |
|---|---|---|
| Observation | A cue visible or audible in the replay | The player noticed it |
| Option | A route available within the time window | Every theoretical route was practical |
| Interpretation | A reason the option may fit | The player's actual intention |
| Outcome | What happened after the choice | The choice caused everything that followed |
Ask the model for comparisons
A sequence model or a clustering tool can retrieve situations with similar visible features. That can reduce the effort of finding review material, but similarity depends on the representation. A system that emphasises location may overlook resource differences. A system trained on match summaries may miss the timing of the choice entirely. Inspect which features define the comparison before interpreting the returned examples.
Include cases where similar decisions led to different outcomes. This guards against a highlight-only sample and helps reveal whether the proposed lesson depends on opponent behaviour. A win probability estimate, where legitimately available and validated, describes a model's expectation under its assumptions. It does not establish that a particular alternative would have won the round.
Separate execution from selection
A reasonable decision may fail because of execution, an unexpected opponent response or ordinary variation. An unsound decision may succeed because the opponent misses an opportunity. Discuss the selection and the execution separately. Ask whether the player entered a defensible situation, then ask how effectively the action was performed. Combining both into a single grade hides the mechanism a practice session should address.
Teammates also change the option set. A route that appears individually dangerous may be supported by a trade position or a coordinated distraction. Conversely, apparent support may arrive too late to matter. Connect the decision tree to the team sequence before describing an individual action as isolated error.
Finish with a reviewable question
Write down the cue, the candidate options and the evidence that could overturn your interpretation. A useful practice question might be: when this cue appears and support is one movement away, what information should be confirmed before committing? That is narrower and more testable than telling a player to improve awareness. In the next review, look for the same condition without selecting only successful rounds.
Keep the original timestamp and the model's retrieval criteria alongside the note. If later footage or player explanation changes the reading, revise it. The goal is not to defend the first diagnosis. It is to produce an explanation that survives contact with better evidence.
