Skip to content

signal-priority.md

Latest commit

 

History

History
119 lines (89 loc) · 4.13 KB

signal-priority.md

File metadata and controls

119 lines (89 loc) · 4.13 KB

Deterministic Signal Review Priority

Signal priority scoring persists a deterministic review priority score for stored signals. It helps operators decide which signals to review first. It is not trade advice, not directional confidence, not broker execution, not auto-trading, and not a signal classifier.

The module lives under app.modules.signal_priority and writes only to signal_priority_scores. Existing signals, classifier outputs, strategy profiles, outcomes, and context artifacts remain unchanged.

Score Components

Each component returns a score from 0 to 1, a reason, a source artifact, and a missing flag:

  • confidence_component: stored signal confidence score.
  • setup_quality_component: latest setup context quality score when available.
  • freshness_component: latest market memory freshness label when available.
  • data_quality_component: latest data quality run score when available.
  • timeframe_agreement_component: multi-timeframe agreement score or setup context agreement.
  • cross_asset_component: cross-asset alignment and data-quality context.
  • historical_reliability_component: historical case similarity and outcome summary context.
  • readiness_component: latest decision readiness label and score.
  • quality_gate_component: intelligence quality run, findings, and shadow classification agreement.
  • outcome_reliability_component: stored outcomes, confidence calibration, and cohort drift context.

Optional context degrades gracefully. Missing artifacts add missing_context warnings and do not fail scoring.

Penalties And Boosters

Penalties are deterministic and persisted in penalties_json:

  • stale data
  • low data quality
  • conflicting evidence
  • no directional signal, neutral, or unclear signal context
  • unresolved critical review blockers
  • failed grounding or safety quality findings
  • pending data recovery
  • high or critical risk notes

Boosters are deterministic and persisted in boosters_json:

  • high confidence
  • strong setup context
  • fresh data
  • aligned multi-timeframe context
  • acceptable data quality
  • historically stable profile or pattern context when available
  • recent observed follow-through when available
  • pending human review urgency

The final score is bounded to 0..1 and stored in priority_score.

Priority Labels

priority_label can be:

  • urgent_review
  • high
  • medium
  • low
  • avoid
  • stale

Labels are review-priority labels only. They do not express order intent, expected account result, or directional recommendation.

Review Buckets

review_bucket can be:

  • high_quality_context
  • needs_confirmation
  • conflicted
  • avoid_or_no_directional_signal
  • stale_or_data_issue
  • review_required

Buckets are intended for dashboard and triage grouping. They use safe review language such as review first, needs confirmation, conflicted, avoid condition, stale data, and review recommended.

Settings

Environment-backed settings:

  • SIGNAL_PRIORITY_VERSION, default v1
  • SIGNAL_PRIORITY_HIGH_THRESHOLD, default 0.75
  • SIGNAL_PRIORITY_MEDIUM_THRESHOLD, default 0.55
  • SIGNAL_PRIORITY_STALE_PENALTY, default 0.30
  • SIGNAL_PRIORITY_CONFLICT_PENALTY, default 0.25
  • SIGNAL_PRIORITY_REVIEW_REQUIRED_THRESHOLD, default 0.50

APIs

POST /signals/{signal_id}/priority-score?forceRecompute=false
GET /signals/{signal_id}/priority-score
GET /signal-priorities?workspaceId={workspace_id}&limit=100&offset=0
POST /signal-priorities/workspaces/{workspace_id}/score-recent?limit=500&forceRecompute=false

List filters:

  • priorityLabel
  • reviewBucket
  • symbolId
  • timeframe

Web Workflow Use

The command center and triage board read GET /signals/{signal_id}/priority-score and sort review cards by priority_score when available. If the endpoint or score row is unavailable, the board keeps its existing deterministic triage classification and time-based fallback ordering.

This module should remain backend-first, deterministic, and non-advisory. Future work can add workspace-level batch scoring jobs, richer dashboard filters, and operator review analytics without mutating existing signals or changing classifier behavior.