I don't have time to write much right now, but as I've mentioned everywhere each category of production (manufacturing, research, services*, food, etc.) should produce a certain amount of revenue for each unit produced. Applying total production to produce these outputs, however, should result in a quadratic cost, and building and maintaining the structures should carry a fixed cost every turn. Moreover, wealth should not be a category in which to inject total production, but should instead be a byproduct of the economic system.
As a result, the profitability curves for every industry should be parabolas of the form -ax^2 + bx - c, where a, b and c are positive numbers. The factor a, which represents the monetary cost for each additional unit of production plowed into final production, will be a certain value for each yield but may be altered by various technologies (or ideological choices reducing, holding stable, or increasing wage rates), while factor b would be equal to (yield category + yield bonuses)(wealth + wealth bonuses). Given a certain capacity for total production, the computer should optimize its use such that maximum possible profit is produced. If there is not enough total production to get to the peak of each profitability parabola then additional total production capacity should be installed, if not then total production (and in turn, population) will be underutilized, leading to unemployment, lower approval, and discontent.
As naselus mentioned either here or in another thread, the level of percentage bonuses in all categories should be reduced dramatically. Otherwise levels of sustainable production will simply rise too far, too fast, leading to the game becoming excessively fast-paced by the end. My idea further helps to keep a lid on this by creating a maximum profitability level over which private industry will simply not go. In a sense, this creates a floor performance for the AI because to a major extent production will be done purely by algorithm.
Production will generate profits, and the player (and AIs) should be able to use this profit in order to generate tax revenue. This tax revenue may subsequently be spent on a variety of fields, from fleet maintenance to making use of unused (unprofitable) total production capacity for one's own purposes (research, manufacturing, etc.). However, collecting tax on a planet will reduce approval there. On the other hand, unemployment (unused production capacity) will also reduce happiness, so there is a good optimization point to be found between taxation and unemployment as well.
There are many places that one could go with this (stock and bond markets, wage rates, etc.), but as I mentioned I don't have much time to write so I'll leave them for a later post.
*Note: services will produce a small amount of approval.
In this new system a revamped model of approval would also be produced in which entertainment centers would play a much smaller role. Approval would instead be determined by ((some factor * services output) + entertainment center morale + other morale) * (employment rate) * (1-tax rate)) / population. Combined with rebellions and other features this could make approval a real factor in determining the governance of an empire.