AI Scenario Explorer

Every cell is a full monthly solve of the model from 2024 to 2030, computed by an independent Python reimplementation (aiscen) that reproduces all 169 cells of the paper's Tables 3, 5 and 6 within the test suite's tolerance, 153 of them to the printed digit; the sixteen that round to a different last digit are listed cell by cell in the repository README. The three named scenarios sit at the corners of this grid. One convention to note: the gain a follows the paper's own explorer. At 0.30 it is held flat; at 0.45 and 0.80 it rises in a straight line from the substantial scenario's mid-2026 anchor (0.35) to the dial's 2030 value, so each level lands exactly on its label. The paper keeps the extreme scenario's own anchor (0.45), so the extreme corner lands within 0.03 points of this reproduction's own run of that scenario rather than exactly on it; the modest corner matches its run. Even the substantial corner, which shares the grid's 0.35 anchor, misses slightly: Table 1's literal slope for that scenario, 0.028 per year, only reaches a 2030 gain of 0.448 (which Table 1 itself rounds to "0.45" in prose), while the grid solves backward from the axis label to land on 0.45 exactly, so its GDP gap reads 8.30 against 8.28 for the scenario's own run, and its average wage 2.16 against 2.14, which rounds to 2.2 rather than the paper's 2.1. GDP growth is the exception, because it is a change over the preceding twelve months rather than a 2030 level and so depends on the whole path: at the extreme corner it reads 15.67 here against that scenario's own 15.46 and the paper's 15.4. Reinstatement runs over six values here, three of them (0.75, 1.00 and 1.25) outside the range the paper uses; at the two of those at or above 1, the labor share rises above its 60 percent starting point, territory the paper's own scenarios never enter (at 0.75 it can go either way: it ends above 60 percent in 162 of that level's 729 cells).