Skip to content

golive_audit.md

Latest commit

 

History

History
218 lines (165 loc) · 6.8 KB

golive_audit.md

File metadata and controls

218 lines (165 loc) · 6.8 KB

🐡 NemoFish — Full System Audit & Architecture

Codebase Overview

Layer Tech Files LOC Size
Terminal (betting engine) Python 56 14,500 —
Backend (MiroFish orig.) Python/FastAPI ~20 ~5,000 —
Frontend (MiroFish orig.) Vue.js 22 469 —
Dashboard (NemoFish) React/Vite/TS 4 ~1,200 —
Data (Sackmann ATP) CSV 50+ — 1.4 GB
Trained Model XGBoost pkl 1 — 74 MB

Architecture

graph TB
    subgraph "📡 Data Layer"
        S["JeffSackmann<br/>143K matches"] 
        OA["The Odds API<br/>Live odds"]
        AT["api-tennis.com<br/>Fixtures"]
        SR["Sportradar<br/>Rankings"]
        PM["Polymarket<br/>Prediction markets"]
    end

    subgraph "🧠 Model Layer"
        ELO["WElo Engine<br/>Surface-specific + margin"]
        XGB["XGBoost Classifier<br/>30K features"]
        KELLY["Kelly Criterion<br/>25% fractional"]
    end

    subgraph "🐡 Agent Swarm (6 agents)"
        STAT["Statistical 0.32"]
        PSYCH["Psychology 0.18"]
        MKT["Market Maker 0.18"]
        NEWS["News Scout 0.13"]
        CONTR["Contrarian 0.09"]
        MIRO["MiroFish LLM 0.10"]
    end

    subgraph "🎯 Strategy Layer (8 strategies)"
        ATP["ATPConfidence 5%"]
        VAL["ValueConfirmation"]
        EDGE["EdgeThreshold"]
        SKEMP["SkempValue"]
        KELL_S["KellyOptimal"]
    end

    subgraph "💰 Execution Layer"
        PM_TRADE["Polymarket Trader"]
        RISK["Risk Manager"]
        PNL["P&L Tracker"]
        CLV["CLV Tracker"]
    end

    subgraph "📊 Presentation"
        DASH["React Dashboard<br/>:5174"]
        API_S["Dashboard API<br/>:8099"]
    end

    S --> ELO
    S --> XGB
    OA --> MKT
    AT --> STAT
    SR --> STAT

    ELO --> STAT
    XGB --> STAT

    STAT --> CONSENSUS["Weighted Consensus"]
    PSYCH --> CONSENSUS
    MKT --> CONSENSUS
    NEWS --> CONSENSUS
    CONTR --> CONSENSUS
    MIRO --> CONSENSUS

    CONSENSUS --> ATP
    CONSENSUS --> VAL
    CONSENSUS --> EDGE

    ATP --> KELLY
    KELLY --> PM_TRADE
    PM_TRADE --> RISK
    RISK --> PNL
    PNL --> CLV
    
    API_S --> DASH

    style MIRO fill:#00d4aa,color:#000
    style CLV fill:#ffd700,color:#000
    style ELO fill:#ff6b6b,color:#fff
Loading

Live Runner Pipeline

sequenceDiagram
    participant Fix as Step 1: Fixtures
    participant Rank as Step 2: Rankings
    participant Poly as Step 3: Polymarket
    participant Swarm as Step 4: Swarm
    participant Strat as Step 4b: Strategies
    participant Exec as Step 5: Execute
    participant Track as Step 6: Track

    Fix->>Rank: Today's ATP matches
    Rank->>Poly: Enriched with rankings
    Poly->>Swarm: + market odds
    Swarm->>Strat: Consensus prediction
    Strat->>Exec: Filtered bets (edge ≥ 3%)
    Exec->>Track: Place on Polymarket
    Track->>Track: P&L + CLV logging
Loading

🔴 Critical Issues

1. No Tests (Severity: CRITICAL)

terminal/tests/ is empty. Zero test coverage for the entire betting engine.

Caution

A money-generating system with zero tests = guaranteed bugs in production. One wrong prediction sign flip → catastrophic losses.

Action: Write unit tests for tennis_elo.py, strategy_base.py, all strategies, and clv_tracker.py.

2. 74MB .pkl Model in Git (Severity: HIGH)

tennis_gb.pkl (74MB) is committed to git. This bloats the repo and makes cloning slow.

Action: Add *.pkl to .gitignore, use Git LFS or store in cloud.

3. Config.yaml Has Stale LLM References (Severity: MEDIUM)

config.yaml says llm_model: "deepseek-reasoner" and fallback_model: "qwen-2.5" but actual .env uses deepseek-v3.2 and qwen3.5-32b.

Action: Sync config.yaml with actual models.

4. NHL/Hockey Code Is Dead Code (Severity: LOW)

nhl_api.py (8.5KB) and hockey XGBoost config in config.yaml are unused — the system only does tennis.

Action: Remove or move to _archive/ to reduce confusion.

5. Dashboard Is Monolithic (Severity: MEDIUM)

App.tsx is 41KB in a single file. Contains all tabs, charts, state management, styling.

Action: Split into components: OverviewTab.tsx, StrategiesTab.tsx, PnLChartComponent.tsx.

6. Backend (MiroFish) Is Disconnected (Severity: INFO)

The backend/ folder is the original MiroFish simulation engine with Zep graph memory, OASIS profiles, and report agents. It has no integration with the NemoFish betting terminal.

Action: Either integrate (use report_agent for bet analysis) or clearly separate in README.

🟡 Moderate Issues

7. Walk-Forward Not Auto-Run

Walk-forward validator exists but hasn't been run on real data yet. No baseline ROI established.

8. Common-Opponent Features Missing

Planned in implementation_plan.md but not yet coded. Expected +1-2% ROI boost.

9. No Cron/Scheduler for Live Runner

live_runner.py is manual-run only. No cron, no systemd, no container orchestration.

10. Sackmann Data May Be Stale

refresh_sackmann.sh exists but hasn't been scheduled. Data freshness depends on manual runs.

🟢 Positive Findings

Area Status
Security (no leaked secrets) ✅
.env not tracked in git ✅
.env.example has no real keys ✅
0 hardcoded API keys ✅
0 TODO/FIXMEs in NemoFish code ✅
All NemoFish Python syntax valid ✅
WElo (Tier 1 research) implemented ✅
Fractional Kelly 25% ✅
CLV tracking ✅
MiroFish LLM integrated ✅
6-agent swarm architecture ✅
8 pluggable strategies ✅
Multi-strategy live runner ✅
The Odds API connected ✅

Priority Action Plan

# Action Effort Impact Priority
1 Write unit tests for Elo, strategies, CLV 🟡 4h 🔴 Critical P0
2 Remove .pkl from git → gitignore + LFS 🟢 30m 🟡 High P0
3 Run Walk-Forward backtest → establish ROI baseline 🟡 2h 🔴 Critical P0
4 Add Common-Opponent features to XGBoost 🟡 4h 🟡 +1-2% ROI P1
5 Fix config.yaml stale refs 🟢 10m 🟢 Low P1
6 Split dashboard into components 🟡 2h 🟡 Med P2
7 Schedule live_runner.py via cron 🟢 20m 🟡 Med P1
8 Schedule Sackmann refresh via cron 🟢 10m 🟢 Low P1
9 Clean dead NHL code 🟢 10m 🟢 Low P3
10 Document backend/terminal separation 🟢 20m 🟢 Low P3