[danoszz]

• docs/MANIFESTO.md:
  • what Penthera is today and what it is not, including that the benchmark has only 22 scored findings, with wide 95% intervals;
  • seven design decisions, each with its cost and what would change it;
  • an FAQ (AI, autonomy, models, accuracy, hosting).
    Readers are invited to challenge any decision in Discussions using the new challenge label.
• docs/vision-2.0.md: what "the best" means, as measurable goals, and how 2.0 closes the biggest gaps. Models propose and deterministic checks decide; local models come first; hosted models (including fast structured-decision models such as Typesafe's Jev) are opt-in only.
• Research program (inside the 2.0 vision):
  • pre-registered experiments E1–E6, with α, power and sample sizes;
  • two labellers with κ ≥ 0.70;
  • safety and ethics rules: no live testing of others' systems, responsible disclosure, ethics review, malware hygiene.
• Linked from the README, the ROADMAP and the docs index.

🤖 Generated with Claude Code