Seven trees and dozens of talents, with effects that don't compare: one adds speed, another a chance, another drones. Playtests show how a talent feels, but not whether it is worth as much as its neighbors. So the model gives every point one price, in the unit players actually feel: time.
A point is worth how much sooner the next three terraforming levels arrive, in the base that starts right after it is bought. The band is 6 to 12% per point.
A second measure looks at the whole game: remove the talent and replay, and the game should end 10 to 30% later, per point spent. It is bigger than the first because gains compound: a faster base reaches every later reset sooner.
The rule: no talent over the top of the band in either measure, and at least one of the two measures inside it. Both are measured in a simulator: a script built from the game's own numbers and code that plays the game with a model player.
Production into time
The 6 to 12% band, read as production, is a multiplier of about ×1.07 to ×1.15 per point.
Why a bonus on one building fades
A building that already has +150% speed runs at 2.5×. A +20% bonus takes it to 2.7×: only 8% more for that building. And the base moves at the pace of all four buildings together, so 8% on one of them is about 2% for the base. Move the sliders. Illustrative numbers.
Each talent gets two grades, one per measure. B is inside the band. A is over the top (over 12% for the next levels, over 30% for the whole game) and S far over it. C is under the band and D far under it (under 2% and 3%). The table shows the talents before tuning. Early and late mean Speed opened as the first tree or a later one. Measured for a good human player (acts every 5 seconds, holds the main building 70% of the time) over a projected 4-hour game, with the tree opened early and late. Results rounded.
BeforeAn earlier model, from September, rated Life Is Not a Race as fine, and its values had been tuned to fit.
FoundWhile rebuilding the model, the AI compared each formula with the game's code. In the model, the talent multiplied production by speed already cut in half. In the game, it multiplies by the full speed from talents.
FixedThe simulator was corrected to follow the code, and now matches it level by level within 1 to 3 percentage points.
ResultThe talent was much stronger than the old model said. That is why it shows up above as failing, far over the band.
One talent only needed a bigger number. The three that fail need more than that: for all of them, a bigger or smaller number wouldn't help, and the formula itself has to change. Two of them are the talents built for the active or idle goal, so the goal stands and their shape doesn't. This is the critique step of the process: the model shows where a talent breaks, design decides what it becomes.
Unchained Reactors: a bigger step
At first it sat under the band on the next levels. A higher value per reactor puts it inside both measures, at about 24% per point for the whole game. One step more reaches about 32%, over the top, so the tuning stops there.
Tested in the simulator: inside both bands.
Life Is Not a Race: a new shape
It halves every speed bonus, then multiplies production by the speed the building gets from talents. Cutting its value doesn't work: the halving would make the first rank slow the player down by about 10%, and the late game would still explode, because those talent bonuses keep growing.
Proposed: a cap, or growth slower than those bonuses.
Greater Incentive and The Drones Remember
Even five times the value would give about 3% per point: the holding bonus lives on one building of four and is diluted by the speed that building already has.
Proposed: reach more buildings, or multiply instead of add.
