Update Log: v10.6 - Dynamic Team Character Analysis & Reverse Engineering

Date: January 23, 2026 Version: v10.6-character

Overview

This update introduces a sophisticated Dynamic Odds Character Analysis system to the AI Engine. Instead of static statistical analysis, the system now performs "Reverse Engineering" on historical data to understand how teams behave under specific odds conditions. This allows for nuanced insights like "Bodo/Glimt often loses as an underdog, but wins specifically in European home games against strong opponents."

Key Features

1. Dynamic Team Character (Dual Analysis)

We now analyze BOTH the Home Team and Away Team separately, based on their specific context:

Home Team Analysis:
- Finds finding closest 15 historical "Home" matches where the team had similar odds (±0.30 margin).
- Calculates Win Rate, BTTS Rate, and Over 2.5 Rate for this specific "Odds Cluster".
- Labels:
  - Banko Ev Sahibi (Win Rate >= 80%)
  - Güvensiz Favori (Win Rate <= 35% at low odds)
  - Underdog (Beklenen) (High odds, low win rate)
  - Golcü Karakter (BTTS >= 70%)
Away Team Analysis:
- Finds closest 15 historical "Away" matches where the team had similar odds.
- Generates separate insights (e.g., Man. City might be "İstikrarlı" away, while their opponent is "Underdog").

2. Match Story Integration

The match story generator (_generate_match_story) now dynamically acts on these character insights:

Narrative Injection: Adds a dedicated "Takım Karakteri" section at the start of the analysis.
Example Output:

Bodo/Glimt Karakteri (Underdog): Evinde bu oranlarda son 15 maçta %13 galibiyet, %46 KG Var. Man. City Karakteri (İstikrarlı): Deplasmanda bu oranlarda son 15 maçta %73 galibiyet, %53 KG Var.

3. Terminology Standardization

Fixed confusing terminology in the API response:

Previous: KG: Üst/Var, 2.5 Alt: Üst/Var (Confusing)
Fixed:
- KG: Var / Yok
- Over/Under: Üst / Alt

4. Reverse Engineering Script (`reverse_engineer.py`)

Added a standalone script to manually "debug" team psychology:

Can query specific conditions like "When does Man. City lose away at 1.25 odds?"
Used to discover hidden patterns (e.g., Bodo's only underdog win was vs Lazio in Europa League).

Technical Changes

Backend (`predictions.service.ts`)

Model Version: Updated to v10.6-character.
DTO: Added teamCharacter field to MatchPredictionDto.
Mapping: Now maps Python's analysis.character object to the frontend response.

AI Engine (`train_v10_full.py`)

New Method: _analyze_odds_character(home_id, away_id, home_odd, away_odd)
Logic Update: predict_match now fetches odds for both teams and triggers the dual analysis.
Fix: Resolved UnboundLocalError for officials_df in live info fetching.

Usage

The new character analysis runs automatically for every match prediction request where odds data is available.

"teamCharacter": {
    "home": {
        "label": "Underdog (Beklenen)",
        "win_rate": 13,
        "btts_rate": 46,
        "matches": 15
    },
    "away": {
        "label": "İstikrarlı",
        "win_rate": 73,
        "btts_rate": 53,
        "matches": 15
    }
}

3.3 KiB Executable File Raw Blame History Unescape Escape