When does the race move to hardware?

This lab assigns model capability seventy eight and hardware capacity thirty eight. Its governing rule is deployment equals the minimum, so usable capacity is thirty eight. Bars compare seventy eight and thirty eight at two pixels per conceptual unit. Raising model capability alone to one hundred leaves deployment at thirty eight because hardware remains lower. These scores are abstract controls, not observed processor benchmarks, model quality evaluations or market shares. The page derives a bottleneck lesson from a supplied post but expressly makes no legal claim about the companies or lawsuit it mentions. No current company allegation or market forecast is established by the minimum formula. Holding the model at seventy eight and raising hardware to ninety moves the bottleneck to the model. Deployment rises from thirty eight to seventy eight, an increase of forty conceptual units or about one hundred five point two six percent relative to the initial value. Bars use two pixels per deployment unit. Once hardware is above seventy eight, further hardware increases produce no deployment gain until the model control rises too. Native prediction selection and actual run feedback can test this consequence. The transfer completion caption has the additional exact condition that model stays seventy eight and hardware exceeds it. This is a local teaching condition, not proof of a real hardware investment outcome. At equal seventy eight controls both capacities limit equally mathematically, yet the source strict less than branch chooses Hardware for ties. Bars show two equal seventy eight capacities at two pixels per unit. The canvas maps each score to available width minus sixty four, multiplied by score divided by one hundred. The orange gate sits at that same minimum endpoint and is redrawn on native resize, with positive drawing buffers preserved. Despite its animated scene label there is no animation loop; state changes and resize redraw static geometry. Prediction and inputs persist in browser local storage, but exports can be produced without running a scenario and still label the minimum as observed. Actual text downloads therefore document conceptual state, not independently collected execution evidence, exhaustive functions or public publication.