​

OMNI Sentiment Score Gate

​

Current posture

• OMNI Sentiment Score is intentionally not public yet
• there is no /v1/signals/sentiment endpoint in the 3.0 surface
• the score remains gated until it clears explicit research, backtest, and stale-data thresholds

​

Why the gate exists

• volatility can be grounded directly in primary-source disclosure cadence and insider activity
• sentiment is easier to overfit, easier to make opaque, and easier to mislead users with if the methodology is not strict
• 3.0 ships the launch criteria and the operator evidence, not a premature production score

​

Required launch thresholds

• primary-source coverage: the model must be driven primarily by first-party or primary-source investor inputs, not social chatter or opaque vendor feeds
• directional precision: the score must beat a naive baseline on held-out forward-return sign classification
• calibration: score buckets must stay monotonic against forward realized volatility and downside risk
• false-positive control: the highest-risk bucket cannot trigger too often without corresponding realized stress
• freshness: stale inputs must degrade or block the score instead of silently reusing old sentiment
• methodology transparency: every public score must expose factor provenance, model version, and degraded-state reasons

​

Validation commands

• export the current gate artifact with bun run ops:sentiment-gate
• validate the gate with bun run validate:sentiment-gate

​

What changes when the gate passes

• a public endpoint can be added
• docs can move from this gate page to a live methodology page
• SDK and prompt assets can expose the score