Changelog

Added custom trust scoring weights and thresholds endpoint for Business and Enterprise plans.

Introduced batch verification support — submit up to 100 items in a single request.

Agent SDK: New `on_threshold_fail: "return_evidence"` mode for CrewAI and LangChain.

Reduced median latency for claim verification from 210ms to 142ms.

Python SDK: Added `AsyncSignalStack` client for concurrent verification.

Fixed a bug where the `language` parameter in claim verification was ignored for some sources.

Go SDK: Fixed context cancellation not propagating to upstream requests.

Added Media Provenance endpoint (`POST /v1/verify/media`) with deepfake detection and C2PA watermark verification.

Released Rust SDK (`signalstack` crate) with async support via tokio.

Agent SDK: Added AutoGen integration package (`signalstack-autogen`).

Webhook delivery rate increased from 5 to 10 events per second per endpoint.

Python SDK: Added `signalstack[cli]` extra with a command-line verification tool.

Document analysis now correctly detects AI-generated content in scanned PDFs (OCR pass-through).

Rate limit headers (`X-RateLimit-*`) now consistently included on all error responses.

Evidence items now include `retrieved_at` field (ISO-8601). Previously this was only available on some endpoints.

Initial release of the SignalStack API v1.

Business Verification endpoint (`POST /v1/verify/business`) with Companies House, SEC EDGAR, GLEIF, OpenCorporates, and Crunchbase.

Document Analysis endpoint (`POST /v1/verify/document`) with AI generation detection and entropy analysis.

Claim Verification endpoint (`POST /v1/verify/claim`) with Wikipedia, Google Fact Check Tools, Semantic Scholar, and GDELT.

Python SDK (`signalstack`), TypeScript SDK (`@signalstack/sdk`), and Go SDK (`github.com/signalstack/sdk-go`) released.

Agent SDK preview: LangChain (`@signalstack/langchain`) and CrewAI (`signalstack-crewai`) integrations.

Webhook support with HMAC-SHA256 signature verification and exponential backoff retries.

Rate limiting with token bucket algorithm — Starter (10/hr), Business (1,000/hr), Enterprise (custom).

Trust scoring with four dimensions: source authority (35%), cross-reference (30%), recency (15%), LLM analysis (20%).