Evidence & Methods

Caveat — Hypothetical Research Plan

This research programme is a conceptual prototype. It outlines how Football State Language could be tested, validated, and operationalised, but it has not yet been reviewed, audited, or endorsed by experienced analysts, coaches, sport scientists, or research professionals within the football industry. All methods, definitions, and statistical plans should be treated as provisional and subject to revision following expert consultation, pilot studies, and formal methodological review.

FSL is better understood as a behavioural research framework for generating and testing football hypotheses than as a settled account of how football works. The pages on this site that read as confident — Vocabulary, Foundations — are offering a language to think with; this page is where that language is meant to be put under pressure.

1. Methods & Research Plan

FSL is a behavioural state model of football. A behavioural model must be observable, reliable, and falsifiable. This section describes how the 13-state architecture is tested in practice.

1.1 Coding System

FSL uses a structured coding system to classify each team's behaviour into one of thirteen states. Coders assign one primary state per team per minute, based solely on observable behaviour.

  • Coding unit: 1 minute

  • One primary state per team per minute

  • Transitions logged separately

  • Coders work independently and blind to outcome

  • Commentary muted during first pass

  • No tactical labels, reputation, or scoreline used in coding decisions

1.2 Data Sources

  • Behavioural observation — state labels coded from broadcast footage; transitions and player-level behaviours coded separately.

  • Quantitative match data — xG, event data, possession, tracking data (Phase 3), and player/squad data.

1.3 Research Phases

  • Phase 1 — Reliability & Feasibility: establish whether states are independently observable; test early hypotheses; κ ≥ 0.65 required to proceed.

  • Phase 2 — Predictive Tests: test whether state durations and transitions outperform conventional metrics.

  • Phase 3 — Player-Level Effects: test whether specific players influence team state stability or volatility.

1.4 Reliability Requirements

  • Minimum two independent coders per match

  • Disagreements resolved by a third coder

  • Per-match κ reported

  • Calibration on three designated matches before live coding

  • Coders must pass calibration (κ ≥ 0.65)

1.5 Sample Requirements

  • Phase 1–2 minimum: 60 matches

  • Full programme target: 150 matches

  • Estimated coded minutes at 60 matches: 5,400

1.6 Analysis Workflow

  1. Preregister hypotheses and analysis plan.

  2. Lock data sources.

  3. Code matches independently.

  4. Compute inter-coder reliability.

  5. Resolve disagreements.

  6. Merge coded states with xG and event data.

  7. Run preregistered statistical models.

  8. Report results, including κ values.

  9. Revise definitions if needed.

  10. Version the manual and proceed to next phase.

2. Formal Hypotheses (v1.0)

These hypotheses reflect the updated 13-state architecture. Each is stated in falsifiable form and preregistered before analysis.

H1 — Collapse precedes measurable xG deterioration. Teams entering Collapse will show a statistically significant decline in xG difference within the subsequent 5–15 minutes, compared to periods coded as Settle, Contain, Connect, or Command.

H2 — Functional stability outperforms possession. Total time spent in Settle + Contain + Connect will correlate more strongly with match outcomes (points, xG difference) than possession percentage.

H3 — Failure states produce tactical distortions. Periods coded as Fog, Drift, Stall, Scramble, or Overheat will show significant changes in spacing variance, pass tempo, pressing trigger distance, and defensive line height variance relative to functional states.

H4 — Functional transitions precede outcome events. Transitions such as Connect → Drive, Drive → Ignite, or Ignite → Slice will occur 2–6 minutes before major outcome events (goals, big chances, defensive errors) more often than expected by chance.

H5 — Low-variability attackers outperform high-variability attackers. Attackers with low self-variability in action patterns will generate higher xG per action than high-variability attackers, controlling for team strength and role.

H6 — State disruptors increase opponent volatility. When specific players are on the pitch, opponents will show increased frequency of Fog, Drift, Scramble, or Overheat, independent of tactics or scoreline.

H7 — Youth performance depends on functional stability. Youth-heavy lineups paired with a stable functional core (Settle/Contain/Connect) will outperform youth-heavy lineups without such stability.

H8 — Substitutions in failure states underperform. Substitutions made during Fog, Drift, Overheat, or Collapse will produce smaller positive impacts on xG difference than substitutions made during Settle, Connect, or Command.

H9 — Error cascades follow destabilising events. After a destabilising event (miscommunication, near-concession, emotional spike), secondary errors will increase significantly within the next 3–7 minutes.

H10 — Half-time state outperforms scoreline. The governing state at half-time will predict second-half xG difference and goals more accurately than the half-time scoreline.

3. FSL Coding Manual (v1.0)

Inter-coder reliability target κ ≥ 0.65 · Minimum coding unit: 1 minute · Version 1.0, June 2026

3.1 State Architecture

FSL defines seven functional states and six failure states. Functional states describe coherent, purposeful behaviour. Failure states describe breakdowns, malfunctions, or destabilisation. Full definitions of each state are given on the Vocabulary page; this manual assumes that vocabulary and adds the markers and decision rules used to apply it in practice.

3.2 Behavioural Markers

Behavioural markers are the observable cues a coder uses to assign a state. They are descriptive, not diagnostic — no single marker confirms a state alone; coders weigh the combination present in the interval.

Settle

  • Passing tempo slows; more sideways/backward passes than forward

  • Players re-form base shape; gaps between lines narrow

  • Marker applies regardless of opponent behaviour

Contain

  • Compact defensive shape; reduced distance between defensive lines

  • Limited forward passing options attempted; possession recycled safely

  • Deliberate reduction of tempo by the team without the ball

Connect

  • Short-to-medium passing combinations between 3+ players

  • Movement supports passing lanes (third-man runs, rotations)

  • Ball retention with clear directional progression

Drive

  • Increased forward passing tempo and vertical progression

  • Players advance beyond the ball; forward runs in numbers

  • More duels and press resistance without loss of shape

Ignite

  • Sudden acceleration in tempo or directness

  • Increase in risk-taking actions (through balls, dribbles, long shots)

  • Distinguished from Drive by suddenness

Slice

  • Line-breaking pass or carry removing an opposition line

  • Rapid combination through a previously closed space

  • Immediate shot/cross/chance within 1–2 actions

Command

  • Sustained territorial dominance with low turnover risk

  • Opponent reactive rather than pressing

  • The team in Command dictates tempo; the opponent reacts, the team decides.

Fog

  • Miscommunication (crossed runs, unclear calls)

  • Passing options available but unused

  • No clear directional intent

Drift

  • Spacing loosens without collapse

  • Possession slows without deliberate reason

  • Momentum and initiative fade

Stall

  • Possession retained but no forward options for 2+ passes

  • Receivers face backward/sideways with no forward option

  • Unintentional failure to progress

Scramble

  • Reactive, uncoordinated defending

  • Loss of defensive shape

  • High-intensity, low-structure actions

Overheat

  • Rushed decisions (forced passes, early shots)

  • Increase in unforced errors

  • High intensity with declining execution

Collapse

  • Breakdown in shape with no coordinated response

  • Individuals act without reference to teammates

  • Opponent gains large space/time advantages

3.3 Decision Rules

These rules resolve ambiguity when multiple states could plausibly apply.

Rule 1 — Primacy of the majority window. Code the state occupying >30 seconds of the minute. If none, code the state present at the end and flag as a transition minute.

Rule 2 — Functional vs. failure precedence. If both functional and failure markers appear, failure takes precedence. Failure states describe breakdown; functional states describe intact organisation.

Rule 3 — Minimum evidence threshold. A state requires at least two independent markers. If no state meets this threshold, code Fog.

Rule 4 — Adjacent-state disambiguation

Drift vs. Stall. Stall = Low Agency × High Clarity (options visible but not reachable). Drift = Low Agency × Low Clarity (no clear options). If a visible unused option exists → Stall; otherwise → Drift.

Contain vs. Stall. Contain = deliberate slowing. Stall = involuntary failure to progress. If intent cannot be read → default to Stall.

Overheat vs. Scramble. Overheat = team still has initiative. Scramble = team reacting to lost possession/opponent attack. Possession at interval start determines assignment.

Drive vs. Ignite. Sustained tempo increase (2+ intervals) → Drive. Single-interval spike → Ignite.

Provisional rule note: the three tie-breaks above are interpretive additions. Flag as provisional; revisit after Phase 1 pilot data.

Rule 5 — No outcome-based coding. Do not infer state from goals, shots, or scoreline changes. Code behaviour, not outcome.

Rule 6 — Disagreement resolution. A third coder resolves disagreements, blind to prior codings and outcome. Their judgement is final; disagreements logged for revision.

3.4 Transition Coding

  • Record outgoing and incoming states

  • Record trigger event (if observable)

  • Record trigger confidence (H/M/L)

  • Transitions must persist for 90 seconds to count

3.5 Coding Sheet Format

Field Format Primary state SET / CON / CNT / DRV / IGN / SLC / CMD / FOG / DRF / STL / SCR / OH / COL Confidence H / M / L Transition flag 0 / 1 Trigger Text Trigger confidence H / M / L

Foundations describes the behavioural model this manual is built to test. Vocabulary gives the plain-language version of every state named here.