A scouting analyst
that shows its work.

How does a question become a shortlist?

1. You ask in plain language

   The answer follows the language of your question. Follow-up questions build on the answers before them.

2. The agent queries the data

   Claude plans the search and calls the data tools. Its answer may use only what they return.

   find_playerget_player_profilesearch_playersfind_similar_playerscompare_players

3. The server builds the cards

   The model names the players and gives one reason for each. The server looks every player up and fills in the numbers on the cards and in the comparison table.

4. You share the result

   Print the answer as a report, or send a read-only link that opens without an account. A shared answer is a snapshot: later data updates do not change it.

Why should I trust the numbers?

show_players
{
  "mode": "shortlist",
  "reference": {"player_id": 4127},
  "players": [    {"player_id": 5310,
     "reason": "Closest match on
       carrying and chance creation;
       presses less than the
       reference."},    {"player_id": 6092, …},
    {"player_id": 7745, …}
  ]
}

• Only what the tools return

  The agent is instructed to use nothing but the results of its data tools: no outside knowledge about players, clubs, fees, ages or careers. An evaluation harness checks the names and numbers in answers against those results.

• Missing data is named

  Ask about market value, contracts, salaries or injuries and the answer says that the data does not hold them, then gives what it can.

• Every answer is dated

  The server checks the data source for changes every 10 minutes and reloads only the leagues that changed. Each answer ends with the date of the data behind it.

• Each answer keeps its trace

  Under an answer you can open how it was produced: the steps the agent took and what each data call returned.

What data does it cover?

metrics per player

47

between checks for new data

10 min

Data current to .

Source

Licensed professional football data: season statistics for each player, per 90 minutes and as ratios.

Metrics

Six families: shooting and chance creation, passing and progression, carrying, On-Ball Value, defending, goalkeeping.

Positions

Eight position groups, from goalkeeper to striker. Each has its own set of style metrics for profiles and similarity.

Seasons

An answer uses each league’s latest complete season unless you ask for the season in progress. The season is shown with the player cards.

Percentiles

A player is ranked against players of the same position group with at least 900 minutes, in leagues of the same tier. 100 is always best.

League tiers

Three: the top five European leagues; other strong leagues and the second divisions of Germany, Spain, Italy and France; all other leagues.

Similarity

How close a player’s style is to a reference player in the same position group, from 0 to 100. Every metric weighs the same, and league level is not part of it.

Level

On-Ball Value per 90 minutes for outfield players; goals saved above average per 90 minutes for goalkeepers.

Who built it?

Client

Clara Vista

The product answers the recruitment questions of its decision-makers and analysts.

Builder

ScrumLaunch

A software development agency. It designed and built the product and hosts it.

Model

Claude, by Anthropic

The agent is a Claude model working through tool use: it plans the queries, and the tools return the data.

Data

Licensed football data

Professional player statistics, used under the client’s licence.