Monitoring and Health¶

How to verify the system is running correctly, handle errors, and track scan performance.

Health Checks¶

Three conditions must be true for the strategy layer to produce valid scans:

1. TradingView Desktop Connected¶

The MCP server must have an active CDP connection to TradingView Desktop.

# Verify connection via MCP
tv_health_check

Expected: { connected: true, host: "localhost", port: 9222 }.

If disconnected: Verify TradingView Desktop is running with --remote-debugging-port=9222. Check that no other process is using port 9222.

2. All Four Indicators Loaded¶

All four indicators must be present on the active chart. The strategy layer checks for each indicator using study_filter substring matching.

Indicator	Verified By
Bias AI	`data_get_pine_tables` with `study_filter: "Bias AI"` returns non-empty
Session Hunt AI	`data_get_pine_lines` with `study_filter: "Session Hunt AI"` returns non-empty
OB AI	`data_get_pine_boxes` with `study_filter: "OB AI"` returns non-empty
SMT Divergences	`data_get_pine_labels` with `study_filter: "SMT"` returns non-empty

If missing: The strategy warns and skips the missing indicator's data. The scan will still run but with incomplete flags. An OB AI or Bias AI absence will likely cause setup_valid: false on every scan because core conditions cannot be evaluated.

3. MCP Server Responding¶

The MCP server must accept and return tool calls within a reasonable time.

If unresponsive: Restart the MCP server process. Check for port conflicts or TradingView Desktop crashes.

Error Handling¶

Missing Indicator¶

Behavior: Warn and skip. The scan continues with whatever indicators are available.

Log output:

{
  "level": "warn",
  "message": "Indicator not found on chart",
  "indicator": "OB AI",
  "study_filter": "OB AI",
  "impact": "structure.order_blocks will be empty"
}

Impact by indicator:

Missing Indicator	Effect on StrategyFlags
Bias AI	`bias.daily` and `bias.weekly` default to `"neutral"`. Every scan is a no-trade (`neutral_bias`).
Session Hunt AI	`session.hunt` defaults to `null` side. Every scan is a no-trade (no hunt state).
OB AI	`structure.order_blocks` empty for both sides. Every scan is a no-trade (`no_ob_in_path`).
SMT Divergences	`smt.bullish_active` and `smt.bearish_active` default to `false`. Setups lose the `smt_confirms` confluence factor (weight 2) but can still be valid if other factors meet `MIN_CONFLUENCE`.

Stale Data¶

Detection: The strategy compares the timestamp of the last successful scan against the current time. If the gap exceeds POLL_INTERVAL * 3, the data is flagged as stale.

Behavior: Log a warning. The stale flag is included in the scan output but does not block the scan.

{
  "level": "warn",
  "message": "Stale data detected",
  "last_scan": "2026-04-14T14:00:00.000Z",
  "current_time": "2026-04-14T14:00:45.000Z",
  "gap_ms": 45000,
  "threshold_ms": 30000
}

Common causes: TradingView Desktop minimized or tab backgrounded (Chrome throttles backgrounded tabs), MCP server overloaded, network issues.

Connection Drops¶

Detection: MCP tool calls return connection errors or timeouts.

Behavior: Retry with exponential backoff.

Attempt	Delay	Max Attempts
1	1 second
2	2 seconds
3	4 seconds
4	8 seconds
5	16 seconds	5 (then halt)

After max attempts, the poll loop halts and logs an error. Manual restart is required.

Logging¶

Scan Results¶

Every scan cycle logs the full StrategyFlags output as JSON:

{
  "level": "info",
  "message": "Scan complete",
  "scan_number": 42,
  "duration_ms": 1250,
  "setup_valid": true,
  "confluence_score": 7,
  "direction": "long"
}

Delta Detection¶

When running in poll mode, only changes between consecutive scans are highlighted:

{
  "level": "info",
  "message": "Delta detected",
  "changes": [
    { "path": "session.hunt.lo_swept", "from": false, "to": true },
    { "path": "session.hunt.side", "from": null, "to": "HUNT_LOW" },
    { "path": "decision.setup_valid", "from": false, "to": true }
  ]
}

Delta detection reduces log noise during quiet periods. If no fields changed between scans, only a heartbeat line is logged.

Error Context¶

Errors include full context for debugging:

{
  "level": "error",
  "message": "MCP tool call failed",
  "tool": "data_get_pine_boxes",
  "study_filter": "OB AI",
  "error": "Connection refused",
  "retry_attempt": 3,
  "scan_number": 42
}

Future: Grafana Dashboard¶

A Grafana dashboard is planned for Phase 3+ to visualize:

Scan metrics: scans per minute, average scan duration, error rate
Alert latency: time between indicator state change and strategy detection
Setup frequency: setups per session, grouped by type (hunt_long_ob_fvg, etc.)
Confluence distribution: histogram of confluence scores across scans
No-trade breakdown: pie chart of no_trade_reasons frequency

The dashboard will consume scan logs via a structured log shipper (e.g., Loki, Elasticsearch). The log format described above is designed to be directly parseable by these systems.