[Roadmap/SR] Current Limitations and Improvement Directions for IU-SR

[Concode0]

Overview: Moving Toward a Robust Geometric Logic

The Iterative Geometric Unbending (IU-SR) pipeline introduced in Versor 1.0.0 serves as a structural proposal for deterministic symbolic regression. While it has proven the potential of GA-based formula extraction, several technical challenges and "residual heuristics" must be addressed to reach full mathematical maturity. This roadmap outlines the critical areas where we seek smarter geometric solutions.

Known Challenges & Areas for Improvement

1. Adaptive Entry for Implicit Mode
   The current criteria for entering Implicit Mode are relatively static. We need a more "intelligent" and adaptive definition that can analyze the geometric structure of the data or the residual energy to determine exactly when an implicit representation is required. ( system now chose explicit for all first_principles datasets )

2. Numerical Stability in Deep Formula Chains
   For complex or long equations, the accumulation of rotor operations can lead to numerical instability. We have observed cases of "gradient explosion" or symbolic translation failures when the rotor chain grows too deep. Improving the robustness of long-form symbolic extraction is a high priority.

3. From Distance Correlation to Coherence in Grouper
   The current grouper module utilizes Distance Correlation to identify non-linear relationships between variables. We plan to replace this with a Coherence-based approach. Utilizing GA-native phasor relationships and phase-coherence metrics would align the grouping logic more closely with the underlying geometric algebra.

4. Elimination of Residual Heuristics
   Parts of the current pipeline still rely on non-geometric numerical "tricks" or standard least-squares refinements. We aim to replace these with purely geometric alternatives—such as isometries, geometric projections, and blade-based rejection—to ensure the framework remains "Purely Geometric" from end to end.

5. Strengthening Organic Coupling Between Phases
   The transition between Phase 0 (Data Prep) and Phase 3 (Refinement) needs to be more fluid. We are looking for ways to pass geometric constraints and information more organically between phases, ensuring that the insights gained in early stages are mathematically preserved throughout the entire extraction process.

Call for Collaboration

We invite researchers and developers with expertise in Clifford Algebra, Signal Processing (Coherence), and Symbolic Computing to contribute.

• Logic Proposals: Suggestions for smarter implicit-mode entry triggers.
• Numerical Optimization: Ideas for stabilizing long-chain rotor products.
• Geometric Refactoring: Replacing current lstsq or correlation-based steps with GA-native logic.

If you have a radical geometric intuition for any of these points, please share your thoughts in this issue or submit a Pull Request. Let's refine the "Unbending" paradigm together!