> ## Documentation Index
> Fetch the complete documentation index at: https://docs.worthcall.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Holly AI Pipeline

> How Holly alerts flow from Trade Ideas through auto-evaluation to tradeable signals.

# Holly AI Pipeline

## Overview

Holly AI is Trade Ideas' AI-powered alert system. EvalEngine ingests Holly alerts, scores them through the 3-model ensemble, and surfaces high-confidence signals.

```
Trade Ideas → CSV Export → holly_import → Auto-Eval → Signal Feed → Bracket Order
```

## Step 1: Import alerts

### Manual CSV import

Export alerts from Trade Ideas as CSV, then:

```
MCP tool: holly_import
Parameter: csv (full CSV content)
```

### File import

```
MCP tool: holly_trade_import_file
Parameter: file_path (/path/to/export.csv)
```

The importer parses Trade Ideas' non-standard CSV format, computes derived metrics (hold time, MFE, MAE, giveback, R-multiple), and stores in the `holly_trades` table.

## Step 2: Auto-evaluation

When enabled, the auto-eval pipeline automatically evaluates new Holly alerts:

```
MCP tool: auto_eval_toggle
Parameter: enabled: true
```

Each alert goes through:

1. **Pre-filter** — skip if symbol was recently evaluated (dedup window)
2. **Feature extraction** — gap %, RVOL, spread, float rotation, extension
3. **3-model ensemble** — Claude + GPT-4o + Gemini score 0–100
4. **Weighted average** — ensemble score with disagreement penalty
5. **Signal generation** — stored in signal feed with go/no-go verdict

Check pipeline status: `auto_eval_status`

## Step 3: Review signals

Navigate to `/signals` to see scored alerts:

* **Score** — ensemble weighted average (0–100)
* **Direction** — long or short
* **Should trade** — go/no-go based on score threshold
* **Symbol** — with current quote data

Or query via MCP: `signal_feed` with optional `since`, `symbol`, `direction` filters.

## Step 4: Execute

For high-confidence signals, execute with full bracket protection:

```
MCP tool: place_advanced_bracket
Parameters:
  symbol: "AAPL"
  action: "BUY"
  quantity: 100          # from size_position
  entry: { type: "LMT", price: 185.50 }
  takeProfit: { type: "LMT", price: 187.00 }
  stopLoss: { type: "TRAIL", trailingPercent: 0.5 }
```

## Holly segments

Your data shows a clear edge difference between Holly segments:

| Segment     | Win Rate | Avg R  | Recommendation |
| ----------- | -------- | ------ | -------------- |
| Holly Grail | 52.3%    | +0.025 | Prioritize     |
| Holly Neo   | 48.0%    | −0.009 | Reduce or skip |

## Key strategies

Top performing Holly strategies (from your autopsy data):

* **Tailwind** — momentum continuation
* **Pullback Long** — dip buying in uptrend
* **Breakdown Short** — breakdown from consolidation

Check strategy rankings: `/holly/autopsy` or MCP `holly_exit_autopsy`

## Analytics tools

| Tool                     | Purpose                                   |
| ------------------------ | ----------------------------------------- |
| `holly_stats`            | Aggregate alert statistics                |
| `holly_trade_stats`      | Historical trade metrics                  |
| `holly_exit_autopsy`     | Strategy leaderboard, giveback analysis   |
| `holly_extract_rules`    | Reverse-engineer alert trigger conditions |
| `holly_backtest`         | Backtest extracted rules                  |
| `trailing_stop_optimize` | Find optimal exit strategy per strategy   |