Architecture

Architecture

STRATHEX owns the judge's workflow. STRATHMARK owns numeric predictions, mark optimization and engine eligibility. Every calculation uses the engine deliberately chosen for the competition; it cannot silently switch.

Application flow

Judge -> STRATHEX menus -> selected STRATHMARK engine
             |                   |
             |              times, marks, evidence
             |
             +-- workbook results
             +-- JSON saves and engine-selection records
             +-- local history and V3 command/receipt records

Calculation connections

Connection	Use
V2 direct Python	Default local calculation; offline-capable.
V2 HTTP	Explicit stateless calculation demo with complete field/history in the request.
Linux V3 subprocess	Separate Python 3.13 competition runtime with signed local authority.
V3 V7 service	Separate authenticated lifecycle pinned to an exact contract/source; Windows qualification remains incomplete.

V3 pre-field forecasts supply seed times without marks. Exact heat membership and stands are required before marks. Review and issue are separate steps.

Records and boundaries

The workbook is canonical for results. JSON preserves competition progress. ResultStore is a separate best-effort history write. V3 also retains durable command IDs and signed lifecycle receipts. These stores are not one distributed transaction.

V2 calculation is distinct from its optional trusted PredictionLedger. Ordinary STRATHEX V2 calculation does not write trusted ledger receipts.

Championship scenarios

Competitors can be grouped when target wood and history context match. Different wood or curated peak windows need separate calls. The simulator fixes its evidence, rejects mixed model bundles and stops rather than returning a partial field.

For the module map and detailed interfaces, see the architecture document.