ECHODEPTHSPORTS
Book Demo →
Methodology Reference

Post-Match
Emotional Analysis

How EchoDepth analyses football manager press conferences using FACS Action Units to produce Confidence, Instability, and Net Confidence scores. Full methodology reference.

The pipeline

EchoDepth's post-match emotional analysis system processes publicly broadcast press conference footage through a five-stage pipeline: video ingestion, frame extraction, Action Unit detection, VAD scoring, and metric derivation. Each stage is auditable — the AU evidence that produces every score is recorded and queryable.

01

Video ingestion

Publicly broadcast footage ingested at source resolution. Timestamps anchored to ISO 8601 standard. No raw video retained after analysis.

02

Frame extraction

Up to 30fps frame extraction. Face detection and landmark normalisation. Frames with occlusion (head turns, water bottle coverage) flagged and excluded from AU scoring.

03

AU detection

44 FACS Action Units detected per frame using EchoDepth's trained convolutional model. Each AU assigned an intensity score (0–5 FACS standard scale). Cultural calibration applied from 14-cohort training data.

04

VAD scoring

AU combination patterns mapped to VAD (Valence, Arousal, Dominance) dimensional scores via learned mapping. Per-frame triples produced. Session-level aggregates calculated with confidence weighting.

05

Metric derivation

Genuine Confidence, Instability Index, and Net Confidence derived from VAD scores and AU combination evidence. Output produced as structured JSON with per-frame AU logs and session aggregates.

Key Action Units

Not all 44 Action Units are equally relevant for manager emotional analysis. The following ten AUs account for the majority of the signal in the Confidence, Instability, and Net Confidence metrics.

AUMuscleMovementSignal
AU1Medial frontalisInner brow raiserWorry, concern, surprise
AU4Corrugator superciliiBrow lowererAnger, concentration, distress
AU6Orbicularis oculi (orbital)Cheek raiserGenuine positive affect (Duchenne marker)
AU7Orbicularis oculi (palpebral)Lid tightenerScrutiny, controlled anger
AU12Zygomaticus majorLip corner pullerSmile (genuine with AU6; social without)
AU14BuccinatorDimplerContempt or suppressed amusement
AU15Depressor anguli orisLip corner depressorSuppressed distress, sadness
AU17MentalisChin raiserDoubt, suppressed emotion
AU23Orbicularis orisLip tightenerControlled determination, suppressed anger
AU24Orbicularis orisLip pressorControlled frustration, anger suppression

Metric definitions

Genuine Confidence
Range: 0–1

Derived from sustained AU6+AU12 co-activation with positive Dominance signals. Measures genuine positive affect — the Duchenne marker of authentic confidence. AU12 without AU6 contributes negatively (performed vs genuine). Higher = more authentic confidence.

Instability Index
Range: 0–1

Derived from AU1+AU4 combinations (worry/concern pattern) and AU15+AU17 co-activation (suppressed distress). Persistent activation above intensity threshold scores progressively. Higher = greater suppressed anxiety or doubt beneath the surface presentation.

Net Confidence
Range: −1 to +1

Genuine Confidence minus Instability Index, normalised. Positive values indicate the confident signal dominates; negative values indicate the instability signal dominates. The most predictive single metric for subsequent team communication quality and match outcome correlation.

Output format

Each analysis session produces a structured JSON output with per-frame AU logs, session-level aggregates, and the three composite metrics. Example structure:

{
  "session_id": "mgr_pmc_20260507_001",
  "manager": "[ANONYMISED]",
  "event": "post_match_press_conference",
  "timestamp": "2026-05-07T17:45:00.000Z",
  "duration_seconds": 847,
  "frames_analysed": 23410,
  "frames_excluded": 312,
  "metrics": {
    "genuine_confidence": 0.58,
    "instability_index": 0.22,
    "net_confidence": 0.36,
    "confidence": 0.87
  },
  "vad_session": {
    "valence_mean": 0.41,
    "arousal_mean": 0.38,
    "dominance_mean": 0.61
  },
  "top_aus": [
    { "au": "AU6", "mean_intensity": 2.1, "activation_rate": 0.34 },
    { "au": "AU12", "mean_intensity": 2.4, "activation_rate": 0.41 },
    { "au": "AU1", "mean_intensity": 0.8, "activation_rate": 0.12 }
  ]
}

FAQ

How does EchoDepth analyse post-match manager emotion?+
EchoDepth processes publicly broadcast post-match press conference footage through its FACS-based emotion recognition pipeline. Per-frame, 44 facial Action Units are detected and intensity-rated. AU combination patterns are mapped to VAD (Valence, Arousal, Dominance) dimensional scores through a learned model calibrated across 14 cultural cohorts. These scores are aggregated into three composite metrics: Genuine Confidence (AU6+AU12 with Dominance signals), Instability Index (AU1+AU4+AU15 combinations), and Net Confidence (the signed difference between the two).
What AU combinations indicate manager confidence?+
Genuine manager confidence is characterised by sustained AU6+AU12 co-activation (the Duchenne marker — simultaneous cheek raiser and lip corner puller indicating genuine positive affect) combined with high Dominance signals from the VAD model. AU12 without AU6 — the social smile — indicates performed composure rather than genuine confidence. AU23+AU24 (lip tightener and pressor) with high Arousal indicates controlled determination, also associated with confident performance.
What is the Instability Index in EchoDepth manager scoring?+
The Instability Index is a composite metric derived primarily from AU1+AU4+AU15 activation patterns. AU1 (inner brow raiser, medial frontalis) combined with AU4 (brow lowerer, corrugator supercilii) produces the classic worry/concern AU configuration. AU15 (lip corner depressor, depressor anguli oris) added to this configuration indicates suppressed distress. The Instability Index quantifies the intensity and persistence of these patterns as a normalised 0–1 score. A high Instability Index alongside a moderate Genuine Confidence score — the divergence pattern — is the most commercially significant signal in EchoDepth's manager scoring system.
Is post-match press conference analysis legal under GDPR?+
EchoDepth analyses publicly broadcast press conference footage. Processing is conducted under Legitimate Interests (Article 6(1)(f) GDPR) applied to public figures in their professional and public capacity. Processing does not produce biometric identification profiles — it produces emotional state aggregate scores. Cavefish Ltd is ICO registered (ZB915633) and maintains a Legitimate Interests Assessment. A full DPIA is available on request at legal@cavefish.ai.

See the data.
Request Access.

Manager intelligence data feeds, club deployment, and performance analytics integrations.

Book Demo

What is Post-MatchEmotional Analysis?