Benchmark Latent Capability
Operational Enterprise Adoption
Human Expert Threshold
🧩 Agentic Task-Horizon Breakdown Matrix
The saved scaling formula starts at log base ten compute twenty five point two in year twenty twenty four. Each year adds log ten of four point two plus half log ten of three, about point eight six two logarithmic units. That equals an effective annual factor about seven point two seven, not four point two alone. At thirty pixels per annual multiplier compute factor four point two measures one hundred twenty six, effective rounded seven point two seven two hundred eighteen point one and difference three point zero seven ninety two point one. The algorithmic multiplier is applied by elapsed time, not once at initialization. Adoption repeats the same curve with a fixed one point five year lag. These assigned growth and historical benchmark labels are not independently observed scaling evidence or a forecast of future AI capability. The task threshold formula is twenty four plus one point eight times log ten of step count. Twelve steps gives about twenty five point nine four, and one hundred steps gives twenty seven point six. At ten pixels per rounded logarithmic unit twelve-step threshold twenty five point nine four measures two hundred fifty nine point four, one hundred-step threshold twenty seven point six two hundred seventy six and difference one point six six sixteen point six. The threshold crossing searches assigned trajectory points spaced point one years apart, converts the first match to a quarter and calls it a ninety percent autonomous-success horizon. Step accuracy never enters this compute-threshold formula. A trajectory crossing and a high whole-task success probability are separate conditions, so the ninety percent wording is not established by that crossing. Whole-task success is computed separately as per-step probability raised to the number of steps. At ninety eight point five percent per step, twelve steps gives about eighty three point four percent, and eighty steps about twenty nine point nine. At two pixels per rounded percent twelve-step success measures one hundred sixty six point eight, eighty-step fifty nine point eight and difference fifty three point five one hundred seven. This assumes independent identical step probabilities, without real trials. Preset and native input listeners register before missing D3 initialization, so some labels and selections can change before the chart fails. Export listeners register afterward and remain unavailable offline. Preserve that original behavior, disclose the absent chart and download, and do not claim actual AI models, legal decisions, trades, scientific research or future benchmarks were run.