Steering
What is currently shaping the method — frozen specs,
learning gates, and distilled directives. A steer without a measured
outcome is labelled as such.
Frozen specs
| Era 2 | era2-v1
· hash e12b5ea9b314
· closed gate unreached |
| Era 3 | era3-v1
· hash 5f736e8cd424
· model claude-sonnet-5
|
Learning gates
No learning gates recorded yet. Set LEARNING_SHADOW_GATE=1 to persist decisions.
Active directives
- Require confirmed material catalysts (geopolitical actions, tariff activation, policy shifts) AND multi-day price confirmation before predicting sector rotation within 48h; narrative alignment alone fails.
- For mega-cap tech (QQQ, MSFT, GOOGL): separate narrative confirmation from structural capex/earnings; do not extrapolate single-name intraday spreads (<0.2%) to index directional calls.
- Weight regime context (risk_on/crisis flags, oil persistence, bond rout) as a veto on narrative-driven theses; if macro headwinds are active, AI/sentiment headlines cannot sustain rallies.
Directive effectiveness (Era 3)
| Directive set | Storylines | Mean edge |
|---|
| no directives active / predates attribution | n=6 | +0.0090 Brier |
Descriptive only — R1, no auto-retire. A storyline is attributed to the directives active at its EARLIEST forecast, never a later one, so a steer landing mid-life cannot retroactively own an older call's outcome.