|
|
7fcfde3ca7
|
feat(polyglot): add GEPA infrastructure for Python shim optimization
- Add .atlas/ with benchmark.sh, score.py, gate.sh for GEPA campaigns
- Add atlas.json project config (auto.capture=suggest, reflection_lm=openai/auto)
- Fix test_search.py receipt path from /mnt/agents/ to /tmp/
Metric targets:
dna_qubo_sort.py: avg |Rank corr| (higher=better, baseline=0.8547)
dna_qubo_nn.py: NN Tm correlation (higher=better, baseline=0.4649)
dna_lut.py: avg |Rank corr| (higher=better, baseline=0.5808)
test_search.py: pass count (ceiling=43, for stability verification)
|
2026-07-07 01:08:49 -05:00 |
|
SilverSight Agent
|
3c35fe50c2
|
Initial SilverSight: deterministic equation search via Fisher geometry
Core components:
- ChentsovFinite.lean (883 lines, 0 sorry): Fisher metric uniqueness on 8-state simplex
- HachimojiCodec.lean: Deterministic E=mc^2 -> Hachimoji state pipeline
- PVGS_DQ_Bridge (8 sections, ~6,150 lines): Photon-Varied Gaussian to Dual Quaternion
- UniversalMathEncoding.lean: 50-token math address space (~10^15 addresses)
- ChiralitySpace.lean: 4D descriptor (phase x chirality x direction x regime) ~2x10^25
- BindingSite (3 files): Amino acid vocabulary, entropy-based bindability
- Python: chaos game, Sidon addressing, Q16.16 canonical, Finsler metric, QUBO/QAOA
- CI: Lean check, Python check, Q16 roundtrip workflows
Papers: Giani-Win-Conti 2025, Chabaud-Mehraban 2022, Pizzimenti 2024, Wassner 2025
|
2026-06-21 18:02:05 +08:00 |
|
