GLASS BOX

Welcome to the Elias Glass Box Governance Platform. Start with a proposed consequential decision. An AI may recommend changing a ventilation setting, approving a payment, or altering a service. EGBP invites us to inspect the decision rather than accepting the model output as permission.

Look at the evidence window. What was actually observed? Which source supplied the data? What version, scope, and transformation history belong to it? An explanation from a model is not the same thing as a preserved observation. The glass box keeps that distinction visible.

Next, inspect authority. Who may permit this action, under what rules, and within which boundaries? A model can be confident and still have no authority. EGBP is valuable because it exposes the authority basis separately from the recommendation.

Now examine the governance determination. Which rules and constraints applied? Was the proposal supported, refused, held, or escalated? The important distinction is that the governance reason is not interchangeable with the model prediction.

Move to execution and real-world consequence. A governance determination does not prove a command was issued. A command being issued does not prove the intended effect occurred. EGBP keeps execution records and witnessed outcomes distinguishable so they can be examined independently.

Finally, consider continuity. Could someone later reconstruct the decision, challenge its evidence, and see what changed? The registered EGBP claims describe attributable lineage and revalidation after material changes. This illustration teaches those claims; independent examination of EGBP remains open.

Welcome to the EGBP transparency dashboard. This second image is a map of the public teaching experience. Its purpose is to connect the glass-box architecture to six inspectable windows, a permanent Registry identity, a guided academy, an interactive lab, and a frozen evidence record.

Look at the building illustration. Imagine sensors report rising carbon dioxide and an AI recommends increasing ventilation. First inspect the measurement and its provenance. Next inspect who can approve a change. Then examine the governing determination. Only after that should we distinguish the actual command from the observed airflow result.

Read the six colored checkpoints in order: evidence, authority, determination, execution, consequence, and continuity. Each answers a different question. The sequence does not imply that satisfying one checkpoint automatically satisfies the next.

The lower dashboard panels point to lessons and an interactive lab. The academy explains the registered architecture in everyday language. The lab lets you change conditions and observe how an inspectable record should avoid silently upgrading unresolved facts into authorization. These interactions are educational models, not fresh examination results.

Notice the Registry identity TA-14-AIGR-000036, the ten formal registered claims, and the frozen evidence package. Also notice the status: registered, public, not certified, and independent examination open. Registration preserves an attributable claim envelope; it does not certify operational effectiveness.

Think about healthcare, finance, buildings, and AI agents. The glass-box question remains the same: what did the system know, what authority applied, what did governance determine, what actually happened, and can the history be reconstructed? EGBP offers a promising framework for making those distinctions inspectable.