FAQ

Where do I choose V2 or V3?

During single-event setup or at multi-event tournament creation. A tournament's children inherit the choice. Nothing is selected by default, and the first numeric operation locks it. See Engine selection.

Can V3 run a full competition?

Yes, with the separately configured Linux competition runtime: STRATHEX 7.4.3 and STRATHMARK 3.0.0rc7, labeled LINUX READY. Older numeric previews cannot issue or settle, and Windows V7 production qualification remains incomplete.

Does race day need a network connection?

V2 uses local Python by default. Linux V3 runs in a separate local Python process. Both need their dependencies and data installed in advance. V2 HTTP and the V7 service are separate configured transports; neither is an automatic fallback.

Why does V3 show a time but no mark?

It is a seeding forecast. Exact heats and stands must exist before field marks are calculated. Review, approval and separate issue confirmation come afterward.

Why do marks change in a final?

Marks belong to a complete field. A final has different competitors, so the engine recalculates and rebases the field rather than copying earlier marks.

Do today's results affect later rounds?

V2 keeps its original exclusive date cutoff and excludes same-day results. Linux V3 freezes evidence within a round, then can learn from valid settled completions at the next round boundary. Neither uses the old 97/3 tournament blend.

Does wood quality change V2 predictions?

No. V2 accepts quality as compatibility context but ignores it numerically. Do not assume V2's inactive-input rules describe every V3 component; see Prediction methods.

Is there an LLM or XGBoost choosing the marks?

There is no hidden STRATHEX-side prediction selector. V2 uses its reviewed hierarchical core. Linux V3 uses Formula and trained CatBoost ML; its LLM council is unavailable. Retained local council experiments are diagnostic candidates.

Have the accuracy changes been implemented?

Yes, in the separate Accuracy Preview program. They are not enabled in competition predictions. Its historical 2.81% gain misses the 5% qualification target and has no independent future validation.

What is the difference between an interval and performance spread?

The forecast interval describes uncertainty in predicted cutting time. Performance spread describes expected race-to-race variation and drives simulation. They are not interchangeable.

What if Excel saves but the database write fails?

Keep the workbook result. Excel is canonical; ResultStore is a separate best-effort write. Preserve the warning and reconcile the derived store without duplicating rows.

Can simulation prove that a handicap is fair?

No. It shows outcomes under the supplied predictions and assumptions. Officials still review the field and apply the governing rules. See Fairness.

Why did predictions change from v6?

V7 moved numerical prediction to STRATHMARK 2's dated-history core and joint mark optimizer. The old local XGBoost/LLM selector and QAA scaling are historical.