JOINT SR ARTICLE ANALYSIS — REVIEW SUMMARY

Method
1. Seven trained cases were recollected under one causal contract; all seven PPO policy replays matched exactly.
2. Broad per-case searches used raw_complete or symmetry-complete variables with three seeds.
3. R2/Pareto fronts identified recurring variables and topologies only.
4. Karman and Illusion were fitted separately as within-objective joint laws.
5. Frozen topologies were refitted with all eligible training-case rows using case-equal weighting.
6. Candidate laws were screened by short CFD, standard 200-step legacy DTW, term deletion, and coefficient-scaling shadow tests.

Karman recommended candidate
  alpha_F(x) = odd_projection[-0.381391 Cd_rear_a]
  alpha_U(x) = 1.307782 Cl_rear_s - 3.431209
  alpha_L(x) = -alpha_U(Gx)

Standard 200-step legacy DTW
  Re50  : 0.9543
  Re100 : 0.9427
  Re200 : 0.8560
  Re400 : 0.7827

Term importance
  1. Rear constant: dominant; deleting it caused the largest degradation, especially at Re200/400.
  2. Rear lift feedback Cl_rear_s: secondary; deleting it caused moderate degradation.
  3. Front Cd_rear_a feedback: weak; deleting it changed short-window DTW only slightly.

Scientific interpretation
The result supports the theory that steady rear counter-rotation supplies the principal downstream-deficit compensation. Rear lift feedback adjusts the oscillatory/asymmetric response. The tested front feedback is not strongly supported as necessary. The current law is not yet uniformly strong at Re400, so it should not be called the final universal Karman law.

Illusion recommended numerical candidate
  alpha_F(x) = odd_projection[-1.826604 Cd_rear_a + 2.064493 Cl_F]
  alpha_U(x) = 1.254440 Cd_rear_a - 1.528074 Cl_F
  alpha_L(x) = -alpha_U(Gx)

Standard 200-step legacy DTW
  0.75L : 0.8749
  1.0L  : 0.9217
  1.5L  : 0.8306

Term importance
  - Cl_F terms dominate front and rear action magnitude in scaling shadow tests.
  - Cd_rear_a terms have weaker magnitude influence and are partly replaceable.
  - Every one-term deletion remained stable and had relatively small DTW differences.
  - Therefore the present four-term law is not demonstrated to be a uniquely necessary mechanism.

Scientific interpretation
The result supports a front-lift/rear-asymmetry control family, but it does not establish explicit target tracking: target/error variables were available yet were not selected stably by the static low-complexity joint searches. The 1.5L case behaves differently and remains the weakest case. This candidate is a numerical joint reference, not a final article mechanism law.

Conclusion
The complete workflow is now clear and operational. It produced substantive Karman mechanism evidence and a stable but non-unique Illusion numerical family. It did not honestly produce a single final law satisfying high DTW, exact symmetry, and complete theoretical support across every case. Further work should focus narrowly on Re400 Karman and target-aware/dynamic Illusion structures rather than adding more infrastructure.
