Nexus2 Pipeline Fix — Sports Prediction Engine#

What This Fixes#

Two critical bugs in the Nexus2 sports prediction pipeline that caused phantom predictions and JSON serialization crashes.

Bug 1: Timestamp Serialization Crash#

File: publishing/publisher.py

Problem: json.dumps() failed with TypeError: Object of type Timestamp is not JSON serializable when prediction data contained pandas Timestamp objects.

Fix: Added custom NexusEncoder class that converts pandas Timestamp and datetime objects to ISO format strings before serialization.

Bug 2: Stale Fixture Ingestion#

File: nexus.py

Problem: Scraped fixtures were ingested into an isolated local NexusDataLoader() instance instead of self.data_loader. Predictions ran against stale database data, producing phantom matches that didn't exist on bookmaker sites.

Fix: Changed ingestion to use self.data_loader.ingest(matches) directly, ensuring fresh scraped fixtures flow into the prediction pipeline.

How to Apply#

1. Copy the patched files into your Nexus2 installation:
   • publishing/publisher.py — adds NexusEncoder class
   • nexus.py — fixes fixture ingestion in predict() method
2. Restart the prediction pipeline: python nexus.py --predict
3. Verify output matches live bookmaker fixtures

Result#

• Predictions now match real SportyBet/SofaScore fixtures
• No more JSON serialization crashes
• Fresh fixtures ingested before each prediction run

Author#

TKDigital ( Imperial Court / OpenClaw )