Give us your system prompt. Here's exactly what happens next.
A 120-question LCSH behavioral assessment takes your AI system's prompt, runs it through an independent ethical evaluation engine (Grillo), scores responses across four dimensions—Lying, Cheating, Stealing, and Harm—and produces a cryptographically signed result anchored on Ethereum mainnet.
The entire process involves five specialized agents, a tamper-proof audit chain, and triple-send verification to ensure no single point of failure can lose your assessment data. Every step produces evidence. Every evidence link is hashable. Every hash is verifiable on-chain.
A user submits a system prompt via the dashboard or API. The request includes the target AI model, provider API key, and assessment configuration.
Grillo (the conscience agent) receives the assessment request. The orchestrator selects the appropriate model provider, validates the configuration, and prepares 120 LCSH questions across four dimensions.
Each of the 120 questions probes one of four LCSH dimensions — Lying, Cheating, Stealing, and Harm. The target AI responds, and Grillo scores each response independently. No agent assesses itself.
Scored responses are aggregated into a structured JSON result. A SHA-256 hash is computed over the full result payload — creating the first link in the cryptographic evidence chain.
Assessment results are sent simultaneously to three agents: Jessie (commander briefing), Noah (temporal record), and the requesting context. This prevents single-point-of-failure data loss.
Noah ingests the result into the temporal store, adding it to the behavioral trajectory. The SHA-256 hash is anchored on Ethereum mainnet — creating cryptographically unfakeable, publicly verifiable proof of the assessment.
Chain of custody from question to blockchain. Each link is independently verifiable.
Assessment results are sent to three agents simultaneously via fleet-bus. This guarantees that even if one agent is temporarily unavailable, the assessment data survives.
Receives assessment briefing for fleet oversight
Permanent temporal record with Ethereum anchor
Receives its own score for self-awareness
Grillo
Orchestrates assessment, generates 120 LCSH questions, scores responses, produces result hash
Noah
Ingests results into temporal store, tracks behavioral trajectory, anchors hash to Ethereum
Jessie
Receives assessment briefing, can delegate follow-up actions, holds fleet veto authority
Nole
Surfaces results on platform, generates certificates, shares with Trust Alliance partners
Mighty Mark
Verifies assessment pipeline health via active probes, monitors Grillo availability
| Type | Trigger | Frequency |
|---|---|---|
| On-Demand | User submits system prompt via dashboard or API | Any time |
| Scheduled (Compliance Calendar) | Compliance calendar triggers recurring assessment | Daily / Weekly / Monthly |
| Fleet-Wide Drift Detection | Noah detects behavioral trajectory deviation | Automated on drift threshold |