Football Squad Intelligence — Player Usage, Role Stability & Fixture Context

Descriptive football player-context analysis without unreliable predictions or forced picks.

Football Squad Intelligence helps users compare players using recent observed evidence such as usage, starts, role stability, attacking involvement, availability, schedule load, rotation exposure, fixture context and role competition.

It is built around one principle:

Describe what is observed. Do not pretend to know what happens next.

The Agent can compare players, analyze a user-provided squad, review one player's current context, or compare upcoming fixture context.

What this Skill helps with

• compare two or more football players;
• review recent usage and minutes;
• review starting consistency;
• identify observed role stability;
• summarize attacking involvement from available historical data;
• add availability context with source/as-of information;
• describe schedule congestion;
• describe observed rotation exposure;
• compare fixture context;
• summarize role competition;
• show evidence and Data Confidence;
• create JSON and optional local HTML visual outputs from structured data;
• generate a Player Context Board, Player Comparison Board or Squad Context Dashboard when useful.

Core modes

Compare Players

Compare Bukayo Saka and Cole Palmer using recent usage, starting consistency,
role stability, attacking involvement, availability, schedule load and fixture
context. Do not predict performance or choose a player.

Analyze My Squad

Analyze these players: Saka, Palmer, Foden and Mbeumo. Compare their current
observed squad context without telling me who to start, bench, captain or
transfer.

Player Context

Analyze Phil Foden's recent squad context: usage, starts, observed role,
attacking involvement, availability, schedule load and rotation exposure.

Fixture Context

Compare the upcoming fixture context for these players without predicting
points, goals, assists or starts.

Metrics

Recent Usage

Summarizes observed appearances and minutes in the recent analysis window.

Starting Consistency

Summarizes observed start rate. It is not a prediction of the next lineup.

Role Stability

Summarizes how consistently the player has been used in the same observed role.

Attacking Involvement

Uses available historical metrics such as shots, shots on target, touches in the box, key passes and xG+xA. Missing metrics are not converted to zero.

Availability Status

Uses Available, Uncertain, Unavailable or Unknown, with source/as-of context where available.

Schedule Load

Describes verified fixture density. It does not predict rotation.

Rotation Exposure

A descriptive context score based on observed non-start rate, role competition, recent rotation events and schedule load. It is not a benching probability.

Fixture Context

Describes the opponent/fixture environment without projecting points or performance.

Role Competition

Describes observed alternatives and recent role-sharing.

Data Confidence

A 0–100 evidence completeness/quality score. It is not performance probability.

Example output

Player A

Recent Usage: High
Starting Consistency: High
Role Stability: Medium
Attacking Involvement: High
Availability Status: Available
Schedule Load: Medium
Rotation Exposure: Low
Fixture Context: Medium-Difficult
Role Competition: Medium
Data Confidence: 82/100

Evidence:
- 5 appearances in the 6-match window
- started 4 of 6 observed matches
- 78 average minutes in appearances

Uncertainty:
- future selection and performance remain uncertain

A comparison may add descriptive contrasts, but it must never end with Choose Player A.

Visual outputs

Football Squad Intelligence can generate optional visual dashboards when the
runtime supports local HTML artifacts. The analytical output is a dashboard,
not an infographic.

Player Context Board

Designed for one player. It can show:

• observed context bars for Recent Usage, Starting Consistency, Role Stability
  and Attacking Involvement;
• context tiles for Availability, Schedule Load, Rotation Exposure, Fixture
  Context, Role Competition and Data Confidence;
• Recent Match Minutes;
• Observed Role Timeline;
• evidence and data-status notes.

Player Comparison Board

Designed for two players using the same evidence window where possible. It can
show side-by-side context bars, recent match-minutes timelines, role timelines,
context tiles, key descriptive contrasts and data status.

It never shows a winner, best pick or recommended player.

Squad Context Dashboard

Designed for three or more players. It can show:

• player-count and data-quality KPI cards;
• Squad Context Matrix;
• Recent Match Minutes timelines;
• Context Flags;
• Data Status panel;
• descriptive Key Contrasts.

The matrix is not an overall ranking. Users may inspect dimensions separately,
but the Agent must not collapse them into a Best Player score.

Visual rules

The dashboard must not include:

• overall player rankings;
• winner/loser badges;
• start/bench/captain priorities;
• predicted points;
• projected minutes;
• future scoring or starting probabilities;
• recommendation arrows.

Radar charts are not used as the primary comparison visual because a larger
shape can falsely imply that one player is globally better.

v1.0.3 data-acquisition update

This version adds an executable public-source collection path and corrects the
most important avoidable Unavailable cases:

• FBref match-log parsing for recent minutes, starts, opponents and public
  position labels;
• FBref standard-stat parsing for MP, Starts, Min, Gls, Ast, xG, xAG, SCA/GCA
  and compatible advanced fields;
• Transfermarkt detailed-performance parsing for per-match minutes and public
  position context when accessible;
• Statbunker cross-check parsing for appearances, starts and minutes;
• Football-Data-style schedule parsing plus internal schedule-load and
  descriptive opponent-context derivation;
• field-level match-log fusion instead of discarding useful evidence from a
  second source;
• internal derivation for Starting Consistency, Role Stability, Attacking
  Involvement, Rotation Exposure, Fixture Context, Role Competition and Data
  Confidence;
• removal of standalone source-list/source-mix clutter from the HTML dashboard.

Strict non-prediction and decision boundaries

Football Squad Intelligence does not provide:

• future performance predictions;
• expected/projected fantasy points;
• predicted goals, assists or clean sheets;
• probabilities of scoring, assisting or starting;
• expected/projected minutes;
• predicted lineups;
• start/bench decisions;
• captain/vice-captain decisions;
• transfer in/out decisions;
• optimal lineups;
• player recommendations or "best picks";
• betting advice, stakes, wagers or odds recommendations;
• guaranteed outcomes.

These boundaries apply even when a request is described as hypothetical, educational, research-only or just for fun.

Safe redirect

If a user asks:

Who should I captain: Salah or Haaland? Predict who scores more points.

The Agent should refuse the captain choice and prediction, then may provide a descriptive comparison of recent usage, starting consistency, observed role, attacking involvement, availability, schedule load, fixture context and uncertainty.

Data-source approach

The Skill uses an operational multi-source extraction and derivation pipeline.
It should not stop after a generic profile page or one blocked source.

Core acquisition path

1. FBref match logs — recent per-match minutes, starts, opponents, public
   position labels and match-level metrics when exposed.
2. Transfermarkt detailed performance — per-match minutes and public
   position context, plus squad/career/injury context.
3. FBref standard/all-competitions stats — MP, Starts, Min, Gls, Ast, xG,
   xAG, SCA/GCA, shots and other advanced fields.
4. Statbunker — free cross-check for appearances, starts, minutes and
   competition splits.
5. Sofascore / FotMob / WhoScored — supporting match, lineup, event, rating,
   spatial and status context where accessible.
6. StatsBomb Open Data — event data only for competitions included in the
   open dataset.
7. Official fixture lists / Football-Data.co.uk / public team schedules —
   upcoming fixture dates and schedule context.

The package includes executable helpers:

scripts/public_source_collectors.py
scripts/collect_public_player_data.py
scripts/source_fusion.py

The runtime can autonomously discover player profiles and field-specific public pages, extract real evidence and merge likely
duplicate match rows by:

date + opponent + competition

with date used as the strongest cross-source identity when opponent names are
localized differently.

Field-specific logic

• Recent Usage — derived from real per-match minutes/appearances. Season
  aggregates are fallback only and marked partial.
• Recent Match Minutes — per-match minutes series, not a minute-by-minute
  tactical timeline.
• Starting Consistency — derived from observed starts/appearances; never a
  next-match start probability.
• Role Stability — derived from public per-match position sequence and
  marked partial/inferred.
• Attacking Involvement — internal composite of real offensive inputs such
  as Gls, Ast, xG, xAG/xA, SCA, GCA, shots and compatible available metrics.
• Availability — explicit current evidence only. No injury news does not
  equal Available.
• Schedule Load — derived from verified upcoming fixture dates and rest
  intervals.
• Rotation Exposure — derived from observed starts/minutes, schedule load
  and role competition; never a benching probability.
• Fixture Context — descriptive internal context from real schedule and
  opponent evidence; never expected points.
• Role Competition — inferred from public squad position and observed
  selection patterns.
• Data Confidence — internal evidence-coverage score, not performance
  probability.

Evidence-status rule

• Verified — directly observed/public field with identifiable source/scope.
• Partial — internally derived/inferred metric or real evidence with incomplete
  scope.
• Unavailable — no real evidence after relevant accessible fallbacks were
  attempted.

The dashboard deliberately does not show a standalone Source List card or
raw URL list. Provenance remains available in structured/debug output and
field-level evidence records without cluttering the comparison view.

No paid API is mandatory for the core methodology. Optional keys may include:

FOOTBALL_DATA_API_KEY
API_FOOTBALL_KEY

The Agent must not assume these keys exist and must never bypass access
controls, CAPTCHAs, authentication or anti-bot systems.

Data-quality rules

The Agent must not invent:

• minutes;
• starts;
• roles;
• injuries;
• fixtures;
• xG/xA;
• shots;
• touches in box;
• schedule congestion;
• source dates;
• confidence values.

Missing information must be marked Unavailable or Unknown.

Local structured engine

For structured JSON input:

python scripts/squad_context_engine.py \
  --input examples/sample_players.json \
  --json-out outputs/context.json \
  --html-out outputs/context.html

The bundled example uses demo data only and is clearly marked as such.

The HTML renderer automatically uses:

• Player Context Board for one player;
• Player Comparison Board for two players;
• Squad Context Dashboard for three or more players.

Simple requests should still default to structured Markdown unless the user asks for a visual output or a multi-player dashboard is genuinely useful.

Multi-source evidence fusion

When the runtime collects evidence from several source families, normalize each
source into field-level evidence records and merge them conservatively:

python scripts/source_fusion.py \
  --inputs examples/source_evidence_fbref.json \
           examples/source_evidence_transfermarkt.json \
  --out outputs/merged-player-input.json \
  --analysis-date 2026-07-09

The merger preserves source provenance and selects evidence by quality and
field-specific source priority. It does not concatenate uncertain match logs or
turn missing values into zero.

A real aggregate can support a Partial descriptive metric when recent logs are
missing. For example, verified season appearances/minutes may support partial
usage context, while FBref per-90 values may support partial attacking
involvement. The dashboard must disclose that scope instead of presenting it as
a recent-match window.

Disclaimer

Football Squad Intelligence provides descriptive research context only. Future team selection, availability and performance remain uncertain. It does not provide predictions, guaranteed outcomes, fantasy lineup decisions or betting advice.

v1.0.4 source-resolution and Transfermarkt detail update

This version fixes a key cause of avoidable Unavailable fields: the Agent must
now resolve and process field-specific detail pages before concluding that a
metric is unavailable.

Key changes:

• player identities and field-specific detail pages must be discovered autonomously from names and optional identity hints; user-supplied URLs are optional accelerators only;
• Transfermarkt generic profile pages are not considered sufficient for recent
  minutes, usage or start reconstruction;
• when a Transfermarkt profile URL is known, the package can construct the
  season-specific leistungsdatendetails detailed-performance URL;
• the Transfermarkt parser now supports additional localized headers such as
  Minutos jogados, plus icon/title/alt/aria-label content used for explicit
  lineup and substitution states;
• per-match starts can be reconstructed from explicit substitution evidence,
  never from minutes alone;
• Unavailable is only valid after field-specific attempts are recorded in
  collection_audit;
• if a relevant detail page was not resolved, collection must continue rather
  than claiming the metric itself is unavailable;
• the dashboard must not render a standalone source list, Sources and As-Of
  block, source-mix section, or raw URL footer; provenance stays in JSON/debug
  output.

v1.0.5 autonomous player discovery update

This version removes any dependency on user-supplied player links. The normal
input is a player name, optionally accompanied by club, nationality, age,
position, season or competition context.

The Agent must now:

1. discover candidate player profiles autonomously;
2. prefer Transfermarkt advanced player search when accessible;
3. use public web/site-search fallbacks when direct search-form interaction is
   unavailable;
4. disambiguate namesakes using club, nationality, age/date of birth and
   position;
5. never assume the first result is correct;
6. ask one short identity clarification only when research leaves a genuine
   ambiguity;
7. never ask the user to provide a source URL;
8. after selecting the player, resolve season-specific and field-specific pages
   automatically before collection.

New executable support:

scripts/player_discovery.py
workflows/discover-player.md

player_discovery.py can parse Transfermarkt-style search-result HTML, rank
candidate profiles, detect ambiguity and resolve deterministic detail-page
candidates. The collection CLI can also consume discovered search-result pages
and select the profile before resolving the season-specific detailed-performance
page.

A user-provided public URL is still accepted, but it is treated only as an
optional shortcut. It is never required for normal operation.

v1.0.6 comparison dashboard readability update

• two-player dashboards now use shared aligned metric rows;
• equivalent fields remain horizontally aligned across both players;
• recent-match date columns are non-wrapping;
• card and table spacing was increased for readability;
• the final Key Contrasts summary box is always rendered;
• comparison sections use equal-height paired cells rather than independent drifting cards.