Skip to content

Missions — strategic chapters

CVNTrade's roadmap is organized around three concurrent missions. Each mission has a clear primary KPI, a multi-track structure, and a per-track gate decision protocol. Stories under any track always trace back to one of these three chapters.

The three missions

🎯 ML Boost — F1_buy 75 mission

Primary KPI : f1_buy ≥ 0.75 median across cryptos Status : active (Sprint F1B-S0 → S6) Epic : CVN-N001-EE (wp#39) Plan : F1_BUY_BOOST_PLAN.md — 13 tracks across 6 tiers (data / label / loss / architecture / calibration / semi-supervised)

The current focus. Track 5 (label smoothing + cleanlab) is closing — see Track 5 results. Track 6 (focal loss) is in implementation. Track-by-track gate decisions per F1_BUY_BOOST_PLAN.md §6.

🎯 Filter tuning — win ratio improvement

Primary KPI : win_ratio lift via post-inference filter chain optimization Status : not started Scope : the 9-filter chain (CUSUM → trend → confidence → meta-label → regime → cost → kelly → confirmation → quality). Tune individual filter parameters + chain ordering to improve trade quality without re-training the underlying model.

Stories will land here as the F1_buy mission delivers ; filter tuning is the natural next lever once the model itself is at a higher floor.

🎯 Economic value growth

Primary KPI : expectancy_net per fold + portfolio-level Sortino Status : not started Scope : everything that converts a calibrated signal into realized PnL — sizing (Kelly), portfolio allocation across cryptos, dynamic SL/TP based on regime, kill-switch refinements, fee/slippage modeling.

This mission consumes the outputs of the first two and validates them in trading-cost space.

Cross-cutting hardening

The CVN-N011-EA — Pipeline Contract Hardening Epic is orthogonal to the three missions — it captures the systemic robustness work that protects all of them (data contracts, integration testing parity, smoke gates, etc.). Stories here are typically reactive (response to incidents §17.1-§17.3) but increasingly proactive as the platform matures.

Process per Story

Every Story under any mission follows the standard development process :

  1. Pick from active OP version (per ADR-69)
  2. Issue + plan dossier under documentation/reviews/
  3. Committee plan_review (per ADR-68)
  4. Implementation + tests (see tests index for the test catalogue)
  5. PR + CR + committee pr_review + merge
  6. FTF sweep (for ML/FTF Track Stories) + per-track gate decision (per F1 plan §6)
  7. Results dossier in this mission's chapter + Story closure

The MLOps readiness template (ADR-70) is mandatory before any merge to production-touching ML code.