# Realtime.Earth — Component Map (high-level)

Realtime.Earth is a geospatial “sense → align → fuse → replay” system for time-based events. Its core is pose estimation + georectification so imagery and observations become spatially anchored, time-indexed layers that can be explored in 2D/3D.

---

## 1) Data Ingest Layer
**Purpose:** Accept heterogeneous streams and files and normalize them into an event-time model.

- **Connectors / Importers**
  - Phone photos/videos (EXIF, device pose when available)
  - Drones / aircraft telemetry (GPS, IMU, camera parameters)
  - Field sensors / citizen reports (forms, messages, attachments)
  - GIS layers (vectors/rasters/COGs/tiles/point clouds)
  - External feeds (incident perimeters, weather-derived products, etc.)

- **Staging + Validation**
  - File hashing / dedup
  - Metadata extraction (EXIF/XMP, timestamps, lens hints)
  - Time normalization (device time vs UTC vs server time)
  - Coordinate normalization (WGS84, local projections)

**Outputs:** canonical “Observation” objects + raw assets in a content store.

---

## 2) Identity + Event Graph
**Purpose:** Keep track of “who/what/where/when” across messy field data.

- **Entity registry**
  - Devices (camera IDs), users/roles, platforms (aircraft/drone), incidents/events
- **Observation graph**
  - Links assets ↔ telemetry ↔ derived products
  - Provenance chain (what produced what, with parameters)

**Outputs:** an event-centric graph enabling auditability and replay.

---

## 3) Pose Estimation + Calibration (AnySurface core)
**Purpose:** Turn imagery into spatially valid measurements even when telemetry is missing or unreliable.

- **Intrinsics**
  - FOV/focal length estimation (from EXIF or solved)
  - Lens distortion model (when needed)
- **Extrinsics (camera pose)**
  - Solve orientation/position from:
    - telemetry (IMU/GPS)
    - tie points / manual correspondences
    - known landmarks / map alignment
    - multi-image constraints (bundle-like consistency, if used)
- **Quality model**
  - Confidence scores, residuals, outlier tie-point rejection
  - “pose versioning” (editable, revertible)

**Outputs:** calibrated camera models + pose solutions per asset.

---

## 4) Georectification + Measurement
**Purpose:** Convert pixels into coordinates and measurements.

- **Ray/terrain intersection**
  - DEM/DTM intersection for per-pixel or per-feature projection
- **Georectified products**
  - Image-to-map overlays (orthorectified approximations where appropriate)
  - Annotated features (points/lines/polygons) derived from imagery
- **Quantification**
  - Distances/areas, plume height proxies, motion/flow vectors (where data supports)

**Outputs:** geo-anchored layers ready for fusion and visualization.

---

## 5) Fusion + Temporal Compositing
**Purpose:** Combine many partial truths into a coherent operational picture over time.

- **Layer fusion**
  - Multiple photos/videos into a consistent spatial frame
  - Sensor tracks + perimeters + annotations
- **Time-indexing**
  - Event timeline (delta-t playback)
  - Snapshots / states / keyframes
- **Conflict handling**
  - competing geometries, differing timestamps, differing confidence
  - “best-known state” vs “alternatives” retained

**Outputs:** time-resolved scene graph of the event.

---

## 6) Storage + Tile/Asset Serving
**Purpose:** Serve large geospatial assets efficiently.

- **Asset store**
  - originals + derived (thumbnails, pyramids, tiles)
- **Spatial indexes**
  - bounding boxes, time ranges, entity tags
- **Serving formats**
  - vector tiles, raster tiles, 3D tiles / glTF / point clouds (as used)
  - range requests / chunking for large files

**Outputs:** fast retrieval for interactive viewers.

---

## 7) Viewers (2D + 3D) + Playback
**Purpose:** Human exploration, analysis, and briefing.

- **2D map viewer**
  - layers, opacity, time scrubber
- **3D globe / terrain viewer**
  - camera frustums, georectified overlays on terrain, 3D annotations
- **Timeline playback**
  - scrub, step, compare states, “then vs now”
- **Analyst tools**
  - measure, annotate, validate tie points, flag uncertain areas

**Outputs:** interactive situational awareness + storytelling.

---

## 8) Authoring + Collaboration
**Purpose:** Turn viewers into a production workflow.

- **Annotation authoring**
  - controlled vocabularies, tags, confidence levels
- **Review + QA**
  - pose approval, derived-layer approval, change history
- **Roles + permissions**
  - analysts, PIOs, responders, citizen contributors
- **Collaboration artifacts**
  - shareable links, “briefing packages,” exported layers/media

**Outputs:** structured products usable by teams.

---

## 9) Security + Provenance + Trust
**Purpose:** Maintain integrity and control in contentious/critical contexts.

- **Access control**
  - org/team spaces, incident-based permissions
- **Provenance ledger (conceptual)**
  - signed/hashed assets, parameter capture, reproducibility
- **Redaction**
  - privacy blur/strip metadata, restricted layers

**Outputs:** defensible, auditable operational picture.

---

## 10) Deployment + Edge Modes
**Purpose:** Operate in poor connectivity and sync later.

- **Edge caching**
  - local replicas of incident packages (tiles + key assets)
- **Intermittent sync**
  - pull/push derived products and annotations
- **Portable “incident kit”**
  - minimal footprint viewer + asset bundle for camps/ICP

**Outputs:** field-usable system that survives bandwidth constraints.

---

# Core objects (canonical nouns)
- **Event / Incident**
- **Observation** (asset + metadata + time)
- **PoseSolution** (intrinsics + extrinsics + confidence)
- **TiePoint** (correspondence constraints)
- **DerivedLayer** (georectified overlay, vector features, tiles)
- **TimelineState** (best-known scene at time t)
- **Annotation** (human/AI-authored, versioned)

---

# Minimal dataflow (one-liner)
Ingest → normalize → solve pose → georectify → fuse over time → serve tiles/assets → view/playback → author/review → sync/secure.