Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Latest commit

 

History

History
328 lines (243 loc) · 12.2 KB

File metadata and controls

328 lines (243 loc) · 12.2 KB

VIF Trading System — Complete System Map

Generated: 2026-05-11 | Status: Zero-Friction Context Layer Complete (Phase 1-4)


How to Use This System

For First-Time LLM Instance (Claude.ai, Cursor, Gemini):

1. Read ONBOARDING.md (5 min)      ← START HERE
2. Read .ai-context.yaml (10 min)  ← Portable context card
3. Read CLAUDE.md (deep dive)      ← Full architecture
4. Run: python tests/test_api_key.py ← Verify setup

For Claude Code Users (You):

1. Reference .claude/memory/ for context continuity
2. Use CLAUDE.md for technical deep dives
3. Run agents/orchestrator_swarm.py for full pipeline

Core Concepts (60 seconds)

Concept Definition
VIF Volatility Imbalance Framework v4.0 — proprietary signal logic (4 layers: gamma regime + structural levels + volume + kill switches)
Signal BUY/SELL/HOLD recommendation with confidence score (0–100) for 2–4 week holding periods
Watchlist Curated set of tickers organized into 4 tiers (VANGUARD / PRIMARY / SCOUTS / WAITING)
Kill Switch Override condition (K1–K6) that downgrades or rejects a signal (extreme volatility, gap risk, low liquidity, news risk, correlation risk, technical breakdown)
Pipeline Orchestrated sequence: parse → fetch → analyze → verify → report
Confidence 0–100 score. <55% triggers GitHub/HF research audit automatically. >80% publishes immediately.

6 Watchlists (170 tickers, 4-tier each)

ID Name Tickers Regime Driver File
WL1 AI Physical Layer & Power Infrastructure 47 Risk-On Tape + CapEx watchlists/AI Physical Layer & Power Infrastructure.txt
WL2 AI Verticals (Supply Chain) 31 Risk-On CapEx + Tape watchlists/AI Verticals (Supply Chain).txt
WL3 Core Growth & Macro Indices 56 Both Earnings + Macro watchlists/Core Growth & Macro Indices (Large-Cap Anchors).txt
WL4 Energy & AI (Power Convergence) 13 Risk-On Contract + CapEx watchlists/Energy & AI (Power Convergence).txt
WL5 Speculative & High-Beta 10 Risk-On ONLY Momentum watchlists/Speculative _ High-Beta.txt
WL6 Trump Admin: Onshoring 13 Risk-On Contract + Macro watchlists/Trump Admin_ Onshoring.txt

Every watchlist contains 4 tiers:

  • ###01_MACRO_VANGUARD — Regime instruments (check first)
  • ###02_PRIMARY_CONVICTION — High-conviction entries (default scan)
  • ###03_SPECULATIVE_SCOUTS — Setup confirmation needed
  • ###04_WAITING_LIST — Monitor only

Architecture Overview

9 Core Agents

Agent File Role Triggers
watchlist_watcher agents/watchlist_watcher.py VIF Analyst (parse + fetch + analyze) Manual or scheduled
orchestrator_swarm agents/orchestrator_swarm.py Master Pipeline Controller (Phase 3) schedule_daily.py or --mode premarket
indicators agents/indicators.py Shared Technical Engine (RSI/MACD/BB/EMA/ATR) Called by watchlist_watcher
weekend_catalyst_agent agents/weekend_catalyst_agent.py Macro & Earnings Briefing Sat 08:00, Sun 18:00 CT
external_alpha_auditor agents/external_alpha_auditor.py GitHub/HF Research (low-confidence signals) Confidence < 55%
finviz_screener_agent agents/finviz_screener_agent.py Custom Screener (independent mode) Manual or scheduled
finviz_orchestrator_coordinator agents/finviz_orchestrator_coordinator.py FinViz Swarm Integration Swarm pipeline
claude_research_agent agents/claude_research_agent.py Ad-hoc Research Q&A Manual prompt
(archived) report_ui_agent agents/archive/report_ui_agent.py JSON→Markdown converter Legacy (use HTML instead)

Swarm Layer (Phase 3 Ready)

9 specialist agents + KV cache manager + gossip router + consensus engine. Status: Infrastructure deployed, integration pending.


Pipeline Modes & Schedules

Mode When Sequence Output
premarket 08:45 ET (weekdays) catalyst_monitor → vif_analyst (1mo) → signal_verifier → report_builder reports/premarket/*.json + .html
market_open 09:35 ET (weekdays) swing_trade_screener (2–4 week setups) Ranked by risk/reward
afterhours 16:05 ET (weekdays) vif_analyst (5d) → postmarket_debrief → alpha_extractor Daily conviction + 5%+ mover analysis
weekend Sat 08:00, Sun 18:00 CT weekend_catalyst_analyst (macro + earnings + rotation) Monday morning briefing
full On-demand All modes sequentially Complete analysis

Entry point: python schedule_daily.py


Configuration & Tuning

All framework parameters in config/vif_config.yml:

vif_framework:
  gamma_regime.positive_threshold: 0.5
  structural_levels.lookback_days: 20
  volume.ma_period: 20
  volume.strong_threshold: 1.5x

kill_switches:
  K1: RSI >80 or <20 → REJECT
  K2: 5-day range >10% → DOWNGRADE
  K3: Volume <1M → REJECT
  K4: Earnings <5 days → DOWNGRADE
  K5: Correlation >0.8 → DOWNGRADE
  K6: Below MA + declining volume → REJECT

api:
  models: {router: haiku-4-5, analyst: sonnet-4-6, synthesizer: opus-4-7}
  temperature: 0 (deterministic)
  max_tokens: 1024
  batch_size: 15 tickers per call

data_fetching:
  cache_ttl_hours: 24
  period_default: "5d"

File Structure (AI Navigation Guide)

🔴 READ FIRST (Core Understanding)

ONBOARDING.md                          ← 5-min new contributor guide
.ai-context.yaml                       ← Portable project metadata
CLAUDE.md                              ← Full technical reference
config/vif_config.yml                  ← All framework parameters
agents/watchlist_watcher.py            ← Core VIF signal logic
agents/orchestrator_swarm.py           ← Pipeline orchestration

🟡 READ SECOND (Reference & Deep Dives)

docs/SWARM_ORCHESTRATOR_GUIDE.md       ← Multi-agent patterns
docs/AGENTS.md                         ← Agent inventory + workflows
docs/QUICKSTART.md                     ← Installation + first run
.claude/memory/watchlist_structure.md  ← Tier breakdown + tickers

🟢 REFERENCE ONLY (Supporting Materials)

docs/                                  ← 50+ docs (setup, guides, system context)
.claude/skills/                        ← 12 skills (catalyst, screening, parsing)
.claude/memory/                        ← User preferences + arch decisions
scripts/active/                        ← Analysis scripts (catalyst, swing, reporting)
reports/                               ← Output (JSON + HTML)

IGNORE (Dependencies, Cache, Version Control)

venv/                                  ← Python virtual environment
.git/                                  ← Version control
data/cache/                            ← Temporary files
logs/                                  ← Execution traces (only if debugging)
scripts/archive/                       ← Deprecated implementations

Common Commands

Quick Verification (No API Credits)

python tests/test_api_key.py           # Validate API key
python tests/test_harness.py           # Offline mock analysis

Single Watchlist Analysis

python agents/watchlist_watcher.py --watchlist vantage_portfolio
python agents/watchlist_watcher.py --watchlist ai_verticals
python agents/watchlist_watcher.py --watchlist energy_ai

Full Pipeline (All Watchlists)

python schedule_daily.py               # Run all modes once

Cost Monitoring

python scripts/active/utilities/check_usage.py

Cost Structure

Metric Value
Daily $0.13 USD (~13,000 tokens)
Monthly $3.90 USD (~390,000 tokens)
Model Routing Haiku (dispatch, cheap) → Sonnet (analyst) → Opus (synthesis)
Cache Efficiency 24-hr yfinance cache + indicator reuse
Batching 15 tickers per API call (not individual calls)

MCP Integrations (External Knowledge)

Service Purpose Status Config
GitHub External alpha audit (repo scanning) Phase 1 complete agents/external_alpha_auditor.py
Hugging Face Academic paper research (30-day cache) Phase 1 complete agents/external_alpha_auditor.py
TradingView Live chart control (78 tools) Standalone node module tradingview-mcp-jackson/

Key Concepts

Signal Confidence Scoring

  • 80–100: High conviction → PUBLISH immediately
  • 55–79: Moderate conviction → Pass 4-gate verification (volume, fundamental, sentiment, macro)
  • <55: Low confidence → Trigger GitHub/HF research audit automatically
  • 0: No signal generated

Temperature = 0 (Deterministic)

Same ticker analyzed twice = identical output. Ensures reproducibility in signal generation.

Batching Strategy (Cost Optimization)

  • 15 tickers per Claude call (not 1 ticker per call)
  • Reduces token spend from $1+ to $0.13/day
  • Enabled via data_fetching.batch_size in config/vif_config.yml

Kill Switch Logic

If ANY kill switch triggers → signal is downgraded or rejected outright:

  • K1, K3, K6 → REJECT (confidence = 0)
  • K2, K4, K5 → DOWNGRADE (confidence -= penalty)

Quick Glossary

Term Meaning
VIF Volatility Imbalance Framework (proprietary signal logic)
Gamma Regime Price action momentum (positive/negative/transition from higher highs/lows)
Structural Levels Support/resistance from 20-day historical lookback (25th/75th percentiles)
Kill Switch Override condition (K1–K6) that downgrades/rejects signals
Confidence 0–100 score indicating signal conviction (>80 = publish, <55 = audit)
Temperature Randomness parameter (0 = deterministic, 1.0 = random). We use 0.
Batching Processing 15 tickers per API call (cost optimization)
TTL Time-to-live (24 hours for cached yfinance data)
VANGUARD Regime instruments tier (regime read, check first)
PRIMARY High-conviction entries tier (default scan)
SCOUTS Speculative, setup confirmation needed tier
WAITING Monitor-only tier (pre-screened for future entry)

For Other AI Platforms (Copy-Paste Prompt)

When sharing this project with ChatGPT, Gemini, or other LLMs, use this boilerplate:

I'm using the VIF Trading System (AI signal generation for swing trades).

WHAT IT DOES:
- Analyzes 170 institutional tickers across 6 watchlists (4-tier structure)
- Applies VIF v4.0 framework (gamma regime + structural levels + volume + kill switches)
- Generates BUY/SELL/HOLD signals with confidence scores (0-100)
- Validates via 4 gates (Volume, Fundamental, Sentiment, Macro)
- Costs ~$0.13/day via Claude API (Sonnet 4.6 analyst, Haiku 4.5 router)

KEY FILES:
- ONBOARDING.md → 5-min intro
- .ai-context.yaml → Portable context
- CLAUDE.md → Full architecture
- config/vif_config.yml → Framework tuning
- agents/orchestrator_swarm.py → Pipeline orchestration

AGENTS:
- watchlist_watcher: Core VIF logic
- orchestrator_swarm: Master controller
- indicators: Shared technical engine
- weekend_catalyst_agent: Macro briefing
- external_alpha_auditor: GitHub/HF research

WATCHLISTS (6 institutional, 170 tickers):
1. AI Physical Layer & Power Infrastructure (47)
2. AI Verticals Supply Chain (31)
3. Core Growth & Macro Indices (56)
4. Energy & AI Power Convergence (13)
5. Speculative & High-Beta (10)
6. Trump Admin Onshoring (13)

QUESTION: [Your question here]

Next Steps

  1. Phase 1 Complete: System scan & inventory
  2. Phase 2 Complete: .ai-context.yaml + ONBOARDING.md created
  3. Phase 3 Complete: Architectural documentation (this file)
  4. Phase 4 Complete: Cross-reference validation

You're now ready to:

  • Share this project with Cursor, Gemini, or Claude.ai
  • Onboard new team members (send them ONBOARDING.md)
  • Run analysis immediately (python schedule_daily.py)
  • Integrate with external platforms via MCP

Status: Zero-Friction Context Layer OPERATIONAL (May 11, 2026)