LLM Integration

LLM integration

What is active?

Runtime	Numeric role of LLMs
V2	None. Optional explanatory text does not change predictions.
Linux V3 competition	Council unavailable. Formula and trained ML provide the forecasts.
V3 council research/service design	Separate model evaluation and qualification; not a promoted Linux competition assessor.

Installing Ollama, an LLM extra or a model does not make it part of a competition's numeric prediction. Linux V3 fields require explicit review with the council absent.

The council design

The wider V3 design has two distinct local model families and one pinned cloud model. They see the same pseudonymous sporting evidence, without names, marks, outcomes, Formula/ML predictions or each other's answers. Each valid forecast is preserved. The aggregate council is one assessor alongside Formula and ML.

Invalid, late or unavailable members abstain. A schema-valid answer alone does not establish accuracy. Evaluation also needs calibration, identity/order invariance, latency, failure handling and rollback evidence.

Diagnostics

The separately retained local three-family diagnostic has its own availability and accuracy limits. It is not the operational council and has not been promoted to competition prediction. Read the accuracy and council report for the actual measurements.

The V3 specification covers provider scheduling, receipts and qualification.