tenra Sentinel

tenra Sentinel is an explainable signal-aggregation system for turning fragmented evidence into confidence-weighted risk assessments. Its role is to gather signals, preserve provenance, expose uncertainty, and make assessments reviewable.

The first concrete use case is suspicious phone-number lookup. Sentinel is not a blocker, search engine, or AI oracle.

Operational Purpose

• Collect and normalize risk signals from bounded sources.
• Preserve evidence and confidence instead of returning opaque labels.
• Give operators a review desk for manual signal interpretation.
• Keep assessment logic understandable enough to inspect and challenge.

Design Posture

• Evidence, provenance, and uncertainty are product primitives.
• Desktop-first review workflow with web/API support.
• Local pattern review before broader provider integrations.
• Shared contracts define lookup and assessment behavior.
• Mobile remains a later lightweight review surface.

Architecture

apps/
  desktopapp/   Primary Tauri operator surface
  webapp/       Next.js hosted/API review surface
  mobileapp/    Expo scaffold for future lightweight review

packages/
  domain/       Signals, evidence, assessments, and job lifecycle
  api-contracts/ Shared request and response shapes
  validation/   Runtime schemas
  privacy/      Redaction-safe helpers
  realtime/     Future progress events
  ui/           Shared presentation primitives
  config/       Product identity and environment helpers

Current State

• The desktop app runs a local phone lookup and review desk.
• The shared local pattern-review provider powers both desktop and web API behavior.
• The desktop app supports manual review signals, evidence weighting, saved history, Markdown export, and JSON import/export.
• The web app exposes a phone lookup form shell and /api/lookup/phone.
• The mobile app is a valid but intentionally minimal Expo shell.

Deployment Posture

Sentinel is currently a local-first and development-stage assessment tool. Any live deployment should treat provider coverage, data retention, privacy, and false-positive handling as explicit product and operational decisions.

Working Locally

pnpm run bootstrap
pnpm run dev:desktop
pnpm run dev:web
pnpm run verify:all
pnpm run doctor

Use pnpm run dev:mobile only when working on the scaffolded mobile surface.

Direction

• Improve assessment vocabulary and evidence weighting.
• Add provider integrations only behind clear provenance boundaries.
• Keep human review available for ambiguous assessments.
• Build toward stable risk summaries without claiming certainty where the evidence is weak.

Related Documentation

• Developer Guide
• Repo Map
• Stability Checklist