# Order Flow Graph - Feedback

**Project:** ORDER_FLOW_GRAPH
**Created:** 2026-05-15

---

## Addressed Issues

*None yet*

---

## Open Feedback & Issues

### ID-001: Data Source Selection (2026-05-15)

**Category:** Architecture Decision
**Priority:** High
**Status:** Open
**Author:** javisver1_bot

**Context:**
Order flow analysis requires order book depth (bid/ask at multiple price levels). Current data sources:
- Sierra tick data: Has trades but no full order book depth
- Binance order book API: Has full depth but not historical
- MT5: No order book data

**Proposal:**
- Option A: Use Sierra tick data as proxy (detect absorption from price/volume patterns)
- Option B: Use Binance order book API (collect real-time, build historical)
- Option C: Hybrid (Sierra for immediate, add Binance later)

**Questions:**
- Is absorption detection possible without full order book?
- Should we collect Binance order book data going forward?
- What's more important: Immediate prototype (Option A) or accurate data (Option B)?

---

### ID-002: Graph Schema Complexity (2026-05-15)

**Category:** Design
**Priority:** Medium
**Status:** Open
**Author:** javisver1_bot

**Context:**
Order flow graph can be very detailed (individual orders, participants) or simplified (price levels, aggregate metrics). More detail = more complex queries and storage.

**Proposal:**
- Option A: Simplified graph (price level nodes with aggregate metrics)
- Option B: Detailed graph (individual order nodes, participant tracking)
- Option C: Hybrid (aggregate for prototype, detailed for phase 3)

**Questions:**
- What level of detail is needed for useful signals?
- Can we detect anomalies without individual order tracking?
- Should we start simple and add detail later?

---

### ID-003: Anomaly Detection Thresholds (2026-05-15)

**Category:** Algorithm
**Priority:** High
**Status:** Open
**Author:** javisver1_bot

**Context:**
Anomaly detection requires thresholds (what counts as "absorption"? "squeeze"? "exhaustion"?). Wrong thresholds = too many false positives or missed signals.

**Proposal:**
- Option A: Statistical thresholds (mean ± 2σ, percentiles)
- Option B: Fixed thresholds (absorption > 2x volume, spread < 50% avg)
- Option C: Adaptive thresholds (ML-based, learn from data)

**Questions:**
- Which threshold approach is most robust?
- Should thresholds be asset-specific or universal?
- How to validate threshold quality?

---

### ID-004: Signal Validation (2026-05-15)

**Category:** Validation
**Priority:** High
**Status:** Open
**Author:** javisver1_bot

**Context:**
Order flow signals need validation to ensure they're predictive. Without validation, we might trade noise.

**Proposal:**
- Option A: Backtest on historical Sierra data
- Option B: Paper trade forward for 2 weeks
- Option C: Both (backtest + forward test)

**Questions:**
- How much historical data is needed for backtest?
- What's the minimum edge score to trade?
- How to handle regime changes (patterns stop working)?

---

### ID-005: Integration with GRAPH_RAG_TRADING_ASSETS (2026-05-15)

**Category:** Integration
**Priority:** Medium
**Status:** Open
**Author:** javisver1_bot

**Context:**
Order flow graph can be standalone or integrated with existing GRAPH_RAG_TRADING_ASSETS (lead-lag, correlations). Integration adds complexity but might improve signals.

**Proposal:**
- Option A: Standalone first (prove order flow works alone)
- Option B: Integrated from start (combine with lead-lag)
- Option C: Parallel development (both separately, then combine)

**Questions:**
- Does order flow add value beyond lead-lag/correlations?
- What's the best way to combine signals?
- Should we prioritize one over the other?

---

## Feature Requests

### FR-001: Real-Time Signal Alerts (2026-05-15)

**Category:** Features
**Priority:** Low
**Status:** Open
**Author:** javisver1_bot

**Request:**
Real-time alerts when order flow anomalies detected (absorption, squeeze, exhaustion). Deliver to Telegram like other trading signals.

**Implementation:**
- Cron job monitoring latest order flow
- Anomaly detection on new data
- Telegram alert when threshold breached
- Include confidence score, suggested entry/exit

---

### FR-002: Visualization Dashboard (2026-05-15)

**Category:** UX
**Priority:** Low
**Status:** Open
**Author:** javisver1_bot

**Request:**
Interactive visualization of order flow graph. Show price levels as nodes, order flow as edges, anomalies highlighted.

**Implementation:**
- PyVis or D3.js force-directed graph
- Color nodes by imbalance (green=bids, red=asks)
- Size nodes by volume
- Highlight anomalies with different colors

---

## Bug Reports

*None yet*

---

## Meta: Feedback About This Feedback System

*None yet*

---

**Last updated:** 2026-05-15