Read the explanation
Capability Horizon computes average velocity across three capability vectors and average weight across four analogy settings. Its general fidelity equation is one hundred minus twenty-eight point five times average velocity minus nineteen point five times average weight, rounded and clipped between twelve and ninety-four. But an early fixture branch replaces this with forty-two point eight when the selected preset is the labor analogy, publication lag is twenty-four and agentic velocity is one point five five. The same branch assigns half-life eight point six. Its condition does not check reasoning velocity, synthesis velocity or analogy weights. At the function level, those untested inputs could differ while the branch still returns fixed values. The actual slider handlers set the selected preset to custom before updating, so ordinary slider changes bypass this fixture branch. At the initial settings, the general fidelity would round to forty-four point three instead of forty-two point eight. These are authored scenario scores, not measured frontier capabilities. The risk label first divides publication lag by the computed half-life and rounds to one decimal. A ratio above two is high, above one point two is moderate, and the remaining values are low. A separate override then replaces this label whenever the labor preset and twenty-four-month lag are selected. It always prints a high label containing two point eight times. This second condition does not check agentic velocity. The function conditions differ, but the actual slider handler sets the preset to custom. A normal agentic-slider change therefore also disables this risk override; it must not be described as leaving the old label stuck in the interface. The diagram shows two distinct decisions, not a single consistent measurement pipeline. Half-life is normally twelve divided by point nine eight times average velocity, rounded with a two-month minimum. None of these rules is independently validated forecasting evidence. The projection starts at a fixed score of sixty-five, regardless of the radar chart's current-level inputs. It samples months zero through thirty-six in three-month steps, thirteen points. Each score is sixty-five times average velocity raised to month divided by twelve, capped at two hundred fifty. The blue curve uses the initial assigned velocities. Both axes have disclosed numeric scales. The orange boundaries use score times month divided by thirty-six times average uncertainty as the spread. The lower boundary is clipped at zero; the upper boundary is not clipped to the score cap. This is a constructed band, not a calibrated confidence interval. Publication lag, analogy weights and the fixed dashboard fidelity do not enter the projection equation. The video describes how this saved tool draws its assumptions, not actual research progress or a prediction of future systems.