The Hybrid Protocol
"A compiler for truth. Deterministic extraction, adversarial checking, and human confirmation."
1. Ephemeral Privacy (First Law)
All session data exists only in Random Access Memory (RAM). No inputs, claims, or generated critiques are ever written to a disk, database, or log file that persists beyond the active session window. Upon session termination (or timeout), a cryptographic deletion certificate is generated, and the memory address space is zeroed.
2. Compiler-Like Rigor
ARGUS-Thesis behaves like a compiler, not a chatbot. It parses your input document into atomic claims (AST), then allows you to run specific "unit tests" (adversarial agents) against each claim. You see the token cost before you commit to an audit.
3. Novelty Classification
We categorize all "valid" claims into one of four novelty tiers. Merely being "true" is not enough for publication.
- Type I: Substantive Contribution (Pass)
- Type II: Incremental Replication (Warn)
- Type III: Contextual Variation (Warn)
- Type IV: Trivial Extension (Fail)
4. The Consensus Engine
We employ a proprietary multi-agent architecture to stress-test your claims. Rather than a single "AI Critic," your work is evaluated by a diverse ensemble of specialized logic engines, each with a conflicting mandate.
Adversarial Diversity
Our agents do not collaborate; they compete to find flaws. This prevents "groupthink" and hallucination loops common in single-model systems.
Cryptographic Signing
The final consensus score is cryptographically signed and stamped into the PDF artifact. This ensures the integrity of the audit cannot be tampered with.