v0.2.0 - Docs Site, Scoring Engine, LLM Integrity

TL;DR

TrustSight now has a full documentation site, a mature scoring engine with FATAL severity and verification evidence, LLM verdict integrity assertions, and a redesigned docs experience with nav flow.

Added

• Full documentation site with Diataxis layout (getting-started, guides, reference, explanation, contributing)
• Scoring engine: FATAL severity (hard-stop at 100), verification evidence (checksum/PGP/gpg subtractions), pinning weights, maturity-gated novelty
• LLM verdict integrity assertions (minimum length, no score leakage, FATAL content requirement, alarmist word suppression)
• Rule scope matching: function_body context for R009/R010/R011, resolved vs raw_line match targets
• R012 prompt injection and R013 unicode bidi override rules
• URL normalization for novelty dedup and observation_count maturity tracking
• TrustSight banner and modern for-the-badge badges

Updated

• All explanation pages expanded with deep reasoning (trust model, cold start, corpus and priors, limits, benchmarks, scoring philosophy)
• Pipeline section expanded with per-stage detail and design decisions
• Home page now redirects to Getting Started with nav footer flow
• README example table aligned with real output format
• Site build artifacts moved to .gitignore

Removed

• Sandbox security model page (not yet implemented)
• Codecov badge from README
• AUR (dogfooding) install section from docs

Stats

• 48 commits since v0.1.0
• 267 tests, all passing
• 81.5% benign zero-rate, 100% CRITICAL recall
• CRITICAL p5 = 40, benign p95 = 20