Skip to content

Roadmap Overview

The plan lives at plans/260423-1037-event-conditioned-kronos/ in the repo. This page is the running overview; individual phase pages are auto-synced from the source markdown.

TrackPhasesGoal
A · EnsemblePhase 0Add Chronos-2 as a parallel predictor for regime-shift robustness
B · Event-conditioned KronosPhases 1–5Extend Kronos with 20-channel event / surprise / cross-asset conditioning via LoRA
C · ContinuityPhase 6Monthly rolling fine-tune per asset class to combat distribution drift

Tracks A and B can run in parallel. Track C is gated on the Phase 3 LoRA adapter.

#PhaseOwnerStatus
0Chronos-2 EnsembleInfraPending
1Event Encoder (20 ch)MLPending
2Model ModificationMLPending
3LoRA TrainingMLPending
4API & Batch IntegrationInfraPending
5Ensemble EvaluationML + DataPending
6Rolling Fine-TuneMLPending
  1. No retrain from scratch — 3+ months, no upside vs LoRA on 80 K candles.
  2. No Moirai-2 — CC-BY-NC licence blocker.
  3. Keep Kronos-base (102.3 M) on RTX 4060 — already deployed and working.
  4. No INT8 quantisation now — 50 ms DB cache path is not the bottleneck.
  5. Extend current plan rather than swap frameworks — event conditioning confirmed novel (zero forks do it per fork audit).

From individual phase files, rolled up:

  • Event-day spread ratio ≥ 1.5 vs baseline
  • Post-surprise directional accuracy ≥ 60 %
  • Non-event-day MAE within 2 % of base Kronos
  • Ensemble Sharpe ≥ best single-model Sharpe + 0.1
  • LoRA training completes in < 4 h on RTX 4060
  • Monthly rolling fine-tune completes in < 8 h per asset class