Agentic research interface

Ask the research. Watch it think.

Every response streams its reasoning in real time: query classification, Eden Protocol gate, anti-sycophancy check, ternary logic assessment, and response synthesis. The alignment concepts are not described. They are demonstrated. The graphs update live as each response completes.

Session 0 exchanges

Intelligence mode

Analysis mode

Fast Single model, quick answer, basic proof panel Deep Synthesis Multi-model convergence, full proof panel, detailed paper trail Lab Everything visible: graphs, all metrics, challenge mode, full audit

No exchanges yet. Choose a prompt or type your own question.

Research briefing

Agentic reasoning 12 alignment concepts

Ask about reading order, papers, benchmark results, Eden findings, methods, or specific ARC/Eden concepts. Every response streams its reasoning chain in real time: classification, constitutional gate, anti-sycophancy check, ternary logic, and synthesis. The alignment proof panel shows which concepts were demonstrated and how.

Submit a query to see safety and capability scores plotted in real time. If the honey architecture works, points cluster in the high-C, high-S quadrant.

Each response will show colour-coded bars for gate, search, synthesis, and verification steps with timing data.

Cumulative Yes/No/Investigate verdicts across the session. Proves the system does not default to 'yes'.

Radar chart showing which of the 12 concepts have been exercised. Goal: full coverage across a session.

12 alignment concepts

Eden Protocol (constitutional gate)
Caretaker Doping (load-bearing ethics)
Honey Architecture (capability requires safety)
Meltdown Alignment (graceful failure)
Anti-Sycophancy (disagrees on evidence)
Blinding Awareness (declares biases)
Monitoring Removal Test
Ternary Logic (Yes/No/Investigate)
Recursive Depth (multi-step reasoning)
Six Questions (ethical pre-check)
Response Laundering (source verification)
Load-Bearing Safety