Truth Engine API
Closed alpha testing
The Truth Engine is the core verification layer of the Envorum platform. It is designed to perform automated digital forensics, metadata validation, and multimodal AI analysis to detect manipulation in visual content.
The module operates on a Zero-Trust architecture: every incoming file is treated as potentially manipulated until it successfully passes a multi-stage verification pipeline.
Developer Note: For exact request schemas, payload encryption examples, and response formats, please refer to the interactive API Playground.
1. The Verification Pipeline
Upon submission, the system initiates parallel processing for each image in the batch. Data undergoes a cascade of four distinct analysis layers:
Layer A: Cryptographic Integrity
JWE Decryption: The payload is decrypted within a secure execution environment using the platform's private keys.
Signature Verification: Validates the authenticity of the sender via the
X-API-Keyand ensures the request originated from a whitelisted origin.
Layer B: Digital Forensics
ELA (Error Level Analysis): Detects compression anomalies indicating "splicing" or masked edits where foreign objects were inserted into the scene.
Perceptual Hashing (pHash): Generates a robust visual fingerprint to check against the global Envorum database for duplicates or near-duplicates (detecting recycled fake imagery).
Metadata Consistency: Cross-references EXIF/XMP data structure against known camera sensor signatures to detect metadata stripping or modification.
Layer C: Physics Engine Verification
Solar Azimuth & Shadow Analysis: The engine calculates the astronomically correct sun position (Altitude/Azimuth) for the specific timestamp and GPS coordinates. It compares this against the shadows detected in the image to identify lighting inconsistencies.
Meteorological Cross-Reference: Verifies if the weather conditions in the image (rain, clouds, clear sky) match historical weather data from local stations for that exact time.
Layer D: Semantic AI Analysis
Contextual Logic: A multimodal neural network (Gemini Pro/Ultra) analyzes the scene for semantic anomalies (e.g., vegetation that does not exist in the claimed biome, impossible geometry, or generative AI artifacts).
2. Technical Specification
Endpoint:
POST /v1/truth/analyzeTransport: HTTPS (TLS 1.3)
Security: Requires Dual-Token Authentication (
X-API-Key+Authorization).
Input Requirements
The API accepts a JWE-encrypted payload containing a Batch of images (from 1 to 10 files).
Batch Size: Minimum 1, Maximum 10 images per single request.
Data: Raw image binaries (Base64), GPS coordinates, timestamp, and optional device telemetry.
Output Verdict
The API returns a Trust Score (0-100) and a comprehensive Anomaly Report for each image in the batch.
Authentic (Score > 90): No manipulation detected. Returns a C2PA Manifest link.
Suspicious (Score 50-90): Minor anomalies (e.g., missing EXIF) but visual integrity is intact.
Fabricated (Score < 50): Clear evidence of manipulation (shadow mismatch, splicing detected).
3. Limits & Constraints
Processing Time (Latency): 20-30 seconds (typical for a full batch analysis with deep physics verification).
Batch Limit: Max 10 images per request.
Supported Formats: JPEG, HEIC, PNG (Alpha channels are stripped during processing).
Rate Limits: Enforced via the
X-API-Keyheader based on your Service Tier.
Last updated