Version History

Version history

For current installation choices, use Engine selection. This page records how the application changed; older limitations apply to those versions. Full dated notes remain in the repository.

7.4.3

Adds explicit shared managed CLI/workspace access while preserving the retro terminal, workbook authority and existing V2/V3 source/model pins. The first shared profile is synthetic SB/UH with real V2 predictions; official activation remains disabled. Encrypted portable profiles keep their fixed retro policy flags before managed passthrough arguments. Existing 7.4.2 profile identifiers and saved competition bindings remain historical, unchanged artifacts.

7.4.2

Coordinates STRATHMARK rc7 and the recovery helper. Encrypted recovery uses bounded GPG pipes, authenticated readback and cleanup after failure. Prospective export repairs require exact signed historical evidence and retain the original cutoff. Startup loads ID-based workbook history through roster names without dropping rows. Sixteen additional sparse-history trials were rejected; the competition model and Formula were retained.

7.4.1

Coordinates rc6 Formula improvements and regression/prospective checks. Adds optional encrypted recovery archives, verified readback before publication and retry after interruption. The installer checks hard-link support before staging an archive. Existing saved competitions keep their original source and model bindings.

7.4.0

Adds chronological V2/V3 accuracy evidence, relevant-history ML, learned Formula priors and calibrated intervals. Portable Linux profiles preserve installations, independent recovery archives and rollback. The local three-model council remains a diagnostic candidate; it was not promoted into numeric prediction.

7.3.2

Coordinates serialized Linux recovery reads and retained-head replacement. Older saved competitions still require their exact installation.

7.3.1 — 2 October 2026

Coordinates rc4 with normalized portable recovery paths. Keeps the full Linux workflow and explicit V2/V3 choice; saved 7.3.0 scopes retain their original environment.

7.3.0 — 2 October 2026

Adds the separate complete Linux V3 competition workflow: Formula/ML, local signed approval and issue, results, corrections, restart, verified recovery, championship receipts and later-round learning. Windows qualification remains separate.

7.1.0

Introduces competition-level V2/V3 selection, the ASCII Wizard, durable V3 rehearsal recovery and acknowledgments. Repairs finance, brackets, rosters, history dates and payout resume, with portable Linux/Windows launch paths and CI. At this release, V3 was rehearsal-only; the full Linux workflow arrived in 7.3.0.

7.0.0

Moves live numerical prediction to STRATHMARK 2:

• Direct V2 fields replace the old manual-override prediction bridge.
• Python and HTTP transports preserve fixed-cutoff parity and prediction metadata.
• Brackets and championship forecasts use V2.
• Same-day weighting, numeric LLM selection, local XGBoost selection and QAA scaling are retired; scikit-learn and XGBoost leave the application runtime dependencies.
• Existing ResultStore gets a pre-V2 backup; results retain stable competition IDs.

6.0.1

Repairs the STRATHMARK 0.4.1 bridge, prediction display reuse, workbook recovery, tournament replay, atomic JSON state, bracket byes, packaging and terminal rendering.

5.x and earlier

Introduces multi-event days, brackets, championship simulation, local predictor experiments, prompt tooling and early 97/3 tournament weighting. Those prediction semantics are historical, not the current v7 engine behavior.