corrigibility/corrigibility-horizon-mapper

⏱️ Corrigibility Horizon Mapper

A browser-based intervention-timing and reversibility simulator by Corrigibility.

Map a fictional agent workflow, classify each action by reversibility, and discover how long an authorized human intervention can still meaningfully change the outcome.

Core concepts

• —Corrigibility horizon
• —Last safe intervention point
• —Intervention latency budget
• —Reversibility debt
• —Irreversibility accumulation
• —Safe rollback depth
• —Stale-plan invalidation
• —Compensating actions

Action classes

• —Idempotent
• —Reversible
• —Compensable
• —Irreversible

Outputs

• —Corrigibility Horizon score
• —Last Fully Reversible Moment
• —Last Effective Intervention Moment
• —Intervention effectiveness curve
• —Workflow reversibility map
• —Rollback recommendation
• —Reversibility debt
• —Machine-readable workflow record

Important

This Space is a synthetic reasoning and evaluation aid. It does not execute real agents or predict the behavior of deployed AI systems. Scores and timing estimates are deterministic heuristics.

Privacy

Everything runs locally in the browser. No workflow data is sent to a backend.

Tech

• —Static HTML
• —CSS
• —Vanilla JavaScript
• —Canvas
• —No backend
• —No API
• —No framework

Built for the Corrigibility organization on Hugging Face.