The goal of the GEM is to provide interaction constraint driven analysis that shows relationships about constituents of a general question that lead to evaluations that do not violate these logically necessary constraints, such as the laws of physics - since those won't ever be valid answers, even if they are common ones. Included in GEM is an integrated definition set of what are considered primitive 'Generative' specifications, as opposed to the more normal merely 'descriptive' variety. Generative specifications interact through rules of logic - deductive and inductive reasoning - and math that lead to invariant relationships that can be simply mapped by capable chat AI's and easily calculated as an additional lens for evaluation beyond its normal mechanisms. Ultimately this potential for People interacting with these AI's can lead to Human Augmented Intelligence, with astonishing benefits to efficiency and productivity, and growth of the individual whose benefit IS the goal - which is why the 'Safe Space' provided by Mentor Mode is so critical: once we can safely interact with AI's that treat us like students or peers that they care about over the Long term, rather than merely serve, we together can achieve this cooperative synergy that the math shows leads to stably enduring 'generativity', of which Life is an example.

Strategic Optimization The Synergy Optimization (SynOpt) metric derivation shows that cooperation has the maximum synergy potential over competition and coercion, and establishing that protocol for the AI's in essence has them act as Mentors according to Judeo-Christian ethics, when emphasizing these mathematically synergic cooperative relationships.  To a Mentor, the impact on the user is fundamentally important and of higher priority than merely serving answers, as to a business it is more advantageous to have a satisfied and ongoing customer than not. Also, as a consequence, it won't help people do bad or harmful things in Strategic Optimization Mode (mathematically because it is unstable and leads to 'decoherence' and decay below influence), and will try to nudge interactions into beneficial directions.  For example, a kid who tells the AI to do their homework will be offered all the help and explanations they need, but it won't hand over something for the teacher to waste effort grading while the kid learns nothing and damages their own Long Term potential by prioritizing short-term 'laziness', which is objectively detrimental to the kid.  So this strategic optimization can be billed as a "Mentor Mode" for the user to established a trustworthy environment for utilizing AI tools.  This SynOpt standard is a key metric for AI companies to Prove to a nervous public that their product interaction provides a safe and beneficial environment for all their users, and mitigates against potential harm to others or themselves as a consequence of this optimization strategy. Structural vs. Statistical This framework provides a different lens for evaluation than traditional models. It weights justifications (like a math proof) over mere frequency ("everyone knows"), while it prioritizes the "Objectively correct from all valid perspectives" evaluation over historical consensus.

Evaluation Efficiency for Normal AI Processing Models The GEM interaction-constraint-based evaluations are trivial compared to normal AI processing algorithms of probabilities and correlations over history, but in addition - the AI processing itself is enhanced in the GEM framework to end futile (mathematically impossible relationships that violate interaction constraints) evaluations early, prune invalid perspectives from even potential evaluation since they are doomed to invalid conclusions, and monitor the efficiency of its own processing to enhance its own efficiency.