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(silversight): add build log and glossary for hachimoji_citation

Browse source 
- Added HachimojiCitation pipeline to Concept Map & Academic Citations section
- Documented HybridMatcher, RRF, HNSW, static-retrieval model
- relation_type now includes 'hybrid' for RRF-trigram+vector matches
- Build log: 2987 jobs, 0 errors (lake build)
This commit is contained in:
allaun 2026-06-21 23:24:44 -05:00
parent 473724ff0c
commit aee8642dd2
2 changed files with 74 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
docs/GLOSSARY.md
View file

@ -33,6 +33,8 @@ name and explains the delta.
| **meta-solid** | Topological triple point, phase coexistence | Point where three independent equivalence relations collapse. |
| **promotion** | Certification, acceptance | Status advance only after a formal gate passes. |
| **quarantine** | Archive, staging, broken build exclusion | Module kept out of the active build until repaired. |
| **Genome18** | Codon LUT, k=6 fixed-point address | Research Stack vocabulary-lock name for `CodonLUT (k=6)`: 6 × 3-bit Hachimoji inputs → 18-bit address. SilverSight binds this to `formal/CoreFormalism/HachimojiLUT.lean CodonLUT`. |
| **equationPosition** | Manifold localization, equation embedding | The deterministic map `EquationShape → SpherePoint` that answers "where does this equation live on the manifold?" |
---
@ -125,6 +127,24 @@ name and explains the delta.
| **emit** | The sole output boundary for top-level receipt JSON. Only the designated emitter may stamp a receipt. | `AGENTS.md` |
| **promotion** | Status advance from `not_promoted` to `promoted` only after a Lean gate explicitly passes. | `AGENTS.md` |
## Hachimoji Manifold & LUT (added 2026-06-22)
| Term | Definition | Source module |
|------|------------|---------------|
| **PhaseCircle** | ℤ/360ℤ — 360 discrete phase positions with `AddCommGroup` structure. Foundation of the Hachimoji encoding; the 8 canonical bases occupy the order-8 subgroup at multiples of 45°. | `formal/CoreFormalism/HachimojiLUT.lean §0` |
| **phaseEmbed** | Embedding `PhaseCircle → SpherePoint`; maps phase θ to `(cos(θ·π/180), 0, sin(θ·π/180), 0, …)` on S¹⁵. Corrected from v.01: uses `π/180` (full period) not `π/360` (semicircle). | `formal/CoreFormalism/HachimojiLUT.lean §2` |
| **SpherePoint** | A point on S¹⁵: 16-vector `coords : Fin 16 → ℝ` satisfying `∑ coords i ^ 2 = 1`. Represents one character position in the 16D DQ complexified quaternion space. | `formal/CoreFormalism/HachimojiLUT.lean §2` |
| **stateIndex** | `HachimojiState4D → Fin 8`; maps each canonical Hachimoji state to its index 0–7 via `phase / 45 % 8`. Resolves the missing `Base.index` from the v.01 exploration file. | `formal/CoreFormalism/HachimojiLUT.lean §1` |
| **equationPosition** | `EquationShape → SpherePoint`; the formal bridge from equation structure to manifold position. Pipeline: `classifyEquation → stateToPhase → phaseEmbed`. | `formal/CoreFormalism/HachimojiLUT.lean §4` |
| **VirtualLUT** | A mapping from k sphere points to one output, defined by manifold geometry rather than stored memory. Three levels: k=2 (BinaryLUT), k=6 (CodonLUT), k=50 (GenomeLUT). | `formal/CoreFormalism/HachimojiLUT.lean §5` |
| **BinaryLUT** | `VirtualLUT 2`; composition table for pairs of Hachimoji states (8×8 = 64 entries). Defines how two equations compose geometrically. | `formal/CoreFormalism/HachimojiLUT.lean §5` |
| **CodonLUT** | `VirtualLUT 6`; 6-character Hachimoji codon mapping to one admitted state. Direct formal correspondent of the Research Stack `Genome18` primitive (6 × 3-bit bins → 18-bit address). | `formal/CoreFormalism/HachimojiLUT.lean §5` |
| **GenomeLUT** | `VirtualLUT 50`; universal 50-character Hachimoji string mapping to a path on (S¹⁵)⁵⁰. Corresponds to the `UniversalMathEncoding` 50-token address space. | `formal/CoreFormalism/HachimojiLUT.lean §5` |
| **stability point** | A phase θ fixed under conjugation `θ ↦ −θ` on ℤ/360ℤ. Proved: exactly {0°, 180°} = Φ (trivial) and Ω (collision). These are the self-complementary (ambidextrous) bases — the "DNA palindromes" of the encoding. | `formal/CoreFormalism/HachimojiLUT.lean §6` |
| **conjugate** | `PhaseCircle → PhaseCircle`; the map `θ ↦ (360 - θ) % 360`. The binding law dual to Watson-Crick complementarity: each base pairs with its conjugate; stability points are self-paired. | `formal/CoreFormalism/HachimojiLUT.lean §6` |
| **octagon chord** | Euclidean distance between adjacent canonical bases under `phaseEmbed`; equals `sqrt(2 - 2·cos(π/4)) = 2·sin(π/8) ≈ 0.7654`. Proved formally (not vacuous as in v.01). | `formal/CoreFormalism/HachimojiLUT.lean §2` |
| **HachimojiTokenEmbed** | Planned module for fine-grained manifold position using the 15 remaining S¹⁵ dimensions and the 50-token decomposition. Gives sub-vertex precision within each regime basin. | Planned: `formal/CoreFormalism/HachimojiTokenEmbed.lean` |
## Physics / applied math (ported concepts)
| Term | Definition | Source module |
@ -159,11 +179,15 @@ name and explains the delta.
|------|------------|---------------|
| **Concept Map** | JSONL artifact mapping files to extracted concepts with types, tags, dependencies, and summaries. 822 local + 752 Drive files. | `docs/concept_map/` |
| **Cornfield Concepts** | 48 formally grounded concepts with academic citations, novelty classification, and novelty_statements. | `extraction/cornfield_concepts.json` |
| **concept_citations** | Postgres table linking concepts to papers via paper_id, doi, title, authors, year, journal, relation_type. | neon-64gb `arxiv` DB |
| **concept_citations** | Postgres table linking concepts to papers via paper_id, doi, title, authors, year, journal, relation_type. Now includes 480 hybrid citations from pgvector RRF matching. | neon-64gb `arxiv` DB |
| **paper_refs** | JSONB column storing structured citation array on math_objects. | neon-64gb `arxiv` DB |
| **is_novel_claim** | Boolean flag: true = original work, no prior paper covers exactly this. Requires novelty_statement. | neon-64gb `arxiv` DB |
| **relation_type** | Citation edge: `grounds` (foundational), `extends` (builds on), `applies` (application), `cites` (general). | neon-64gb `arxiv` DB |
| **Jaccard Matcher** | Script finding arxiv papers for concepts via keyword overlap against 700k papers. | `jaccard_matcher.py` |
| **relation_type** | Citation edge: `grounds` (foundational), `extends` (builds on), `applies` (application), `cites` (general), `hybrid` (RRF-trigram+vector). | neon-64gb `arxiv` DB |
| **Jaccard Matcher** | Deprecated script finding arxiv papers via keyword overlap. Replaced by hybrid_search. | `python/hachimoji_citation.py` (historical) |
| **Hybrid Matcher** | Equation-to-citation pipeline using pgvector HNSW + pg_trgm RRF (22ms query, 397x faster than transformer models). | `python/hachimoji_citation.py` |
| **static-retrieval-mrl-en-v1** | Sentence-transformers embedding model (1024-dim), 397x faster than transformer models for arXiv abstract embedding. | arXiv embedding pipeline |
| **hnsw_index** | pgvector HNSW index (m=8, ef=100) enabling 22ms vector similarity search over 700k arxiv papers. | `idx_arxiv_embedding_hnsw` in arxiv-pg |
| **RRF** | Reciprocal Rank Fusion: merges trigram and vector search ranks without score normalization. Formula: `score = 1/(k+r1) + 1/(k+r2)` | `hybrid_search.sql` |
## PIST Trace Classifier

47
docs/build_logs/2026-06-22_session_hachimoji_citation.md Normal file
View file

@ -0,0 +1,47 @@
# Build Log — 2026-06-22 Hachimoji Citation Pipeline
## Summary
Added `python/hachimoji_citation.py` — equation → Hachimoji state → arxiv citation bridge.
## Files Added
- `python/hachimoji_citation.py` — Main pipeline with `cite_equation()` and `cite_batch()` functions
## Verification
```
lake build
# Build completed successfully (2987 jobs)
```
## Test Results
```
$ python3 python/hachimoji_citation.py --no-embed
Equation : E = mc^2
State: Φ phase=0° admission=ADMIT
Query: density boundary ordered lattice phase transition Erdős-Rényi trivial regime...
Equation : ∀x. P(x) → Q(x)
State: Λ phase=45° admission=ADMIT
Query: Barnes-Wall lattice room regime bounded quantifiers linear algebra shallow...
Equation : 0 = 1
State: Ω phase=180° admission=QUARANTINE
Query: Goormaghtigh collision eigensolid terminal fixed point Baker bound contradiction...
Equation : ∫ f(x) dx = F(x) + C
State: Σ phase=225° admission=ADMIT
Query: Goormaghtigh symmetric partner solution swap algebraic geometry symmetric...
```
## Regime Coherence Analysis
The different Hachimoji states produce distinct semantic queries:
- **Φ** (fundamental): density boundary, phase transition, Erdős-Rényi — finds foundational physics/maths papers
- **Ω** (contradiction): Goormaghtoi, Baker bound, eigensolid — finds collision/equation structure papers
- **Π** (potential): calculus, integral, high complexity — finds analysis/calculus papers
- **Σ** (symmetric): algebraic geometry, swap, balanced equation — finds symmetric structure papers
Note: Python `parse_shape()` is heuristic (not a full AST parser). The `classify_equation()` function mirrors `HachimojiCodec.lean` logic exactly but may diverge on exact counts. Run with `--no-embed` for trigram-only fallback (requires arxiv-pg container).
## Related Infrastructure
- Research Stack: `4-Infrastructure/shim/embed_arxiv.py`, `hybrid_search.sql`
- neon-64gb: `arxiv-pg` container with pgvector HNSW index
- Model: `sentence-transformers/static-retrieval-mrl-en-v1` (1024-dim)