AI Alignment Pacing Simulator

Balanced Pacing uses capability rate four point five, verification depth three, compute cap seventy, and governance fifty. The executed capability formula gives one hundred thirty three point eight. The initial HTML fixture says one hundred forty two point six, and comments describe an intended calibration, but the code does not implement those intended numbers. Use the executed calculation rather than assuming every displayed fixture is authoritative. These are assigned indices, not measured frontier model capability. The dashboard margin is control capacity minus capability pressure plus forty seven point six five. Balanced settings give one hundred one point nine, which selects High Assurance Margin. Aggressive settings eight point five, one point five, ninety five, and twenty give point six five, rounded to point seven. It is positive, so the label is Elevated, not Critical Uncontrolled. The old explainer’s negative dashboard claim does not match this source formula. The twenty four month chart has different equations. At the aggressive endpoint, capability is two hundred four point five and control sixty two. Adding its fifty point offset leaves negative ninety two point five. So the chart can be negative while dashboard margin is point seven. These are distinct authored scores, not two measurements of one quantity. Balanced chart margin ends at thirty six rather than the dashboard’s one hundred one point nine. The chart includes ten extra control points at its endpoint only when compute cap is below eighty. Moving cap from seventy nine to eighty removes that bonus in one jump. The dashboard formula has no matching jump. Such choices can dominate conclusions drawn from a toy simulator. Presets and export operate locally; they do not constrain actual compute or audit a model. Native checks preserve the original dependency errors and selected controls with remote traffic blocked, rather than certify the real D three chart or policy efficacy.