allaun/SilverSight
1
0
Fork 0
mirror of https://github.com/allaunthefox/SilverSight.git synced 2026-07-31 01:25:21 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

docs: update glossary with new terms from this session

Browse source 
Added 12 draft terms:
- break-glass, fusion of fusions, dual quaternion model selector
- chiral ratio (χ), theorem attack, degenerate sector
- Gemma4-12B, mass semantic numbers, crossInputGap
- cleanMerge_preservesGap, lbi, warm mode

All marked as draft (not yet bound to authoritative modules).
This commit is contained in:
allaun 2026-06-23 00:05:31 -05:00
parent 8e72cec9ef
commit 4ea8e7d09f
1 changed files with 12 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
docs/GLOSSARY.md
View file

@ -245,4 +245,15 @@ authoritative module definition.
| Term | Notes | Proposed owner |
|------|-------|----------------|
| *none* | — | — |
| **break-glass** | Emergency model selection skill. Only for unsolvable problems where all free models have failed. Triggers multi-model panel via OpenRouter Fusion. | `.opencode/skills/break-glass/SKILL.md` |
| **fusion of fusions** | 4D × 2 model panel architecture. Each dimension (Math, Proof, Code, Diversity) has 2 models. Dimensions fuse independently, then results fuse together. Maps to OpenRouter Fusion 8-model panel. | `6-Documentation/docs/specs/break_glass_test_list.md` |
| **dual quaternion model selector** | Model selection using dual quaternion chiral ratio. Each model is a dual quaternion (real = compressive, dual = anti-compressive). χ = \|real\|² / (\|real\|² + \|dual\|²) determines selection. | `4-Infrastructure/shim/dual_quat_model_selector.py` |
| **chiral ratio (χ)** | Ratio of compressive to total semantic mass. χ > 0.5 = compressive (keep). χ < 0.5 = anti-compressive (drop). Used for model selection. | `4-Infrastructure/shim/dual_quat_model_selector.py` |
| **theorem attack** | Compensation strategy for a model's degenerate sector. E.g., `delegate_to_opus` for models that can't do formal proofs. | `4-Infrastructure/shim/model_theorem_matrix.py` |
| **degenerate sector** | A problem sector where a model scores < 0.4. Each model has specific degenerate sectors that require theorem attacks. | `4-Infrastructure/shim/model_theorem_matrix.py` |
| **Gemma4-12B** | Local LLM running on qfox-1 RTX 4070. Endpoint: `http://127.0.0.1:8081/v1`. Model ID: `gemma4-12b`. Free, ~54 tok/s. Primary aid for questions. | `python/gemma4_mcp.py` (SilverSight) |
| **mass semantic numbers** | Z-transform compressed model selection history. μ[k] = a₁·μ[k-1] + a₂·μ[k-2] + b₀·u[k]. Stores coefficients + state vector, not full history. | `4-Infrastructure/shim/dual_quat_model_selector.py` |
| **crossInputGap** | Spectral signature predicate: no active bin in one signature is adjacent to an active bin in another. Required for cleanMerge_preservesGap. | `Semantics.Spectrum` (Research Stack) |
| **cleanMerge_preservesGap** | Theorem: merging two gap-valid signatures with zero resonance degeneracy and cross-input gap preserves the spectral gap. Currently has 1 sorry (Q16_16↔Bool bridge). | `Semantics.GraphRank` (Research Stack) |
| **lbi** | Lake Build + Ingest wrapper. Runs `lake build` and ingests results into ENE on neon-64gb. Use instead of bare `lake build` to keep databases in sync. | `4-Infrastructure/shim/lbi.sh` |
| **warm mode** | Gemma4-12B server configuration with quantized KV cache (q8_0), full GPU offload, and auto-restart. Model stays in GPU memory between requests. | `~/.config/systemd/user/gemma4-server.service` |