Remove GODEL_BOUNDARY_AND_CLASSIFIERS.md

docs/GODEL_BOUNDARY_AND_CLASSIFIERS.md

@@ -1,171 +0,0 @@
-# Gödel Boundary and Classifier Limits
-
-## The Core Discovery
-
-The Hachimoji classifier reveals something fundamental about **all** classifiers:
-
-**Every finite classifier has a Gödel boundary — inputs that look structurally normal but are semantically undecidable.**
-
-## What the Hachimoji Model Shows
-
-| Input | Structural | Semantic | Result |
-|-------|-----------|----------|--------|
-| `E = mc^2` | Φ (trivial) | Trivial | ✓ Correct |
-| `0 = 1` | Ω (contradiction) | Contradiction | ✓ Correct |
-| `G = not(provable(G, S))` | Φ (trivial) | **Gödel sentence** | ✗ Wrong |
-| `this equation is not classifiable` | Ζ (fallback) | **Self-referential** | ~ Graceful |
-| `this statement is false` | Ζ (fallback) | **Liar paradox** | ~ Graceful |
-
-**The classifier doesn't understand self-reference.** It only sees structure.
-
-## The Three Regimes
-
-### 1. Structural Regime (finite, decidable)
-- `classifyEquation` maps equation shapes → 8 states
-- Deterministic, fast, no ambiguity
-- **Works for:** equations with clear mathematical content
-
-### 2. Semantic Regime (infinite, decidable)
-- The system understands what the equation means
-- Can distinguish trivial from profound
-- **Works for:** equations with clear semantic content
-
-### 3. Gödel Regime (infinite, undecidable)
-- The system encounters self-referential inputs
-- Cannot classify without understanding its own provability
-- **Fails for:** Gödel sentences, liar paradox, self-reference
-
-## The Gödel Boundary Formula
-
-For any finite classifier C with N states:
-
-```
-Gödel(C) = { x | classifying x requires knowing whether C classifies x correctly }
-```
-
-This set is **non-empty** for any sufficiently powerful classifier.
-
-### Proof sketch:
-1. Let C have N states
-2. Consider the input "this input maps to state S"
-3. C must classify this input
-4. If C maps it to S → it's correct → it should map to ¬S (paradox)
-5. If C maps it to ¬S → it's incorrect → it should map to S (paradox)
-6. Therefore, C cannot correctly classify this input
-7. C must return a default/fallback state (the NaN event)
-
-## Implications for All Classifiers
-
-### 1. Every classifier has a blind spot
-
-No finite classifier can correctly classify all inputs. There will always be inputs that look structurally normal but are semantically undecidable.
-
-### 2. Graceful degradation is the best strategy
-
-When a classifier hits its Gödel boundary:
-- **Option A:** Crash (bad)
-- **Option B:** Return random (bad)
-- **Option C:** Return default/fallback (good) — the NaN event
-- **Option D:** QUARANTINE/HOLD (good) — the admission gate
-
-### 3. The admission gate is the safety valve
-
-The Hachimoji encoding handles Gödel with three admission states:
-- **ADMIT:** Clear classification (Φ, Σ, Λ)
-- **QUARANTINE:** Uncertain or degenerate (Ω, Ζ)
-- **HOLD:** Needs more evidence (Π, Ρ, Κ)
-
-This is the **controlled explosion** by Gödel — the system degrades gracefully rather than crashing.
-
-### 4. Structural classifiers are safe but incomplete
-
-The Hachimoji classifier is **structurally sound** (doesn't crash) but **semantically incomplete** (doesn't understand self-reference). This is the best we can do with finite resources.
-
-### 5. The halting problem is the Gödel boundary
-
-The halting problem (no algorithm can decide if an arbitrary program halts) is exactly the Gödel boundary for program classifiers:
-- Structural: "does this program have a while loop?"
-- Semantic: "does this program halt?"
-- Gödel: "does this program halt on itself?"
-
-## Connection to BMCTE and Symbolic Regression
-
-### BMCTE
-
-The BMCTE regime is projection-dominated because:
-- The Fock space is exponential in p (combinatorial explosion)
-- BMCTE never constructs it — only samples projections
-- **The projection is the Gödel boundary** — BMCTE cannot classify its own provability
-
-### Symbolic Regression
-
-The log prescreen is a structural classifier:
-- "Is this a power law?" (log-log transform)
-- "Is this an exponential?" (log-y transform)
-- "Is this a polynomial?" (y-x transform)
-
-**The Gödel boundary for symbolic regression:**
-- "Is this equation expressible as a finite expression tree?"
-- "Is this equation's provability decidable?"
-- The prescreen returns None for sin(x) — it doesn't crash, it says "I don't know"
-
-### Logarithms and Gödel
-
-**Logarithms tame combinatorial explosion** (exponential → linear)
-**But logarithms cannot tame self-reference** (Gödel → undecidable)
-
-This is the fundamental limit:
-- **Structural compression:** logarithms, projections, finite states
-- **Semantic compression:** understanding, self-reference, Gödel boundary
-
-## The Unifying Principle
-
-**Every finite system has three regimes:**
-
-| Regime | Growth | Boundary | Strategy |
-|--------|--------|----------|----------|
-| **Structural** | Exponential | Finite | Logarithmic compression |
-| **Semantic** | Infinite | Decidable | Projection sampling |
-| **Gödel** | Infinite | Undecidable | Graceful degradation |
-
-**SilverSight's strategy:**
-1. **Structural:** Hachimoji encoding (8 finite states)
-2. **Semantic:** Chaos game search (projection sampling)
-3. **Gödel:** Admission gate (QUARANTINE/HOLD for uncertain)
-
-**The NaN event is the Gödel boundary — the controlled explosion that keeps the system from crashing.**
-
-## Practical Implications
-
-### For classifiers:
-1. Every finite classifier will encounter inputs it cannot correctly classify
-2. Design for graceful degradation, not perfection
-3. The admission gate (ADMIT/QUARANTINE/HOLD) is the safety valve
-4. The Gödel boundary is the limit of what the classifier can know about itself
-
-### For SilverSight:
-1. The Hachimoji encoding is structurally sound but semantically incomplete
-2. The chaos game search is projection-dominated (cannot explore all of Hilbert space)
-3. The log prescreen is a structural filter (cannot catch sin(x) or self-referential equations)
-4. All three have Gödel boundaries — inputs they cannot correctly handle
-
-### For BMCTE:
-1. The regime is projection-dominated (logarithmic compression)
-2. The projection cannot classify its own provability (Gödel boundary)
-3. The NaN event (entropy invariance) is the graceful degradation
-4. The system works because it never tries to classify the undecidable
-
-## Final Statement
-
-**The Gödel boundary is the limit of finite classification.**
-
-Any system that maps infinite input space to finite states will encounter:
-1. Inputs that look structurally normal but are semantically undecidable
-2. The NaN event — graceful degradation to a default state
-3. The admission gate — QUARANTINE/HOLD for uncertain classifications
-
-**The Hachimoji encoding is a controlled Gödel explosion.**
-**The logarithm is the universal combinatorial compressor.**
-**Together, they form SilverSight's finite-infinity duality.**
-
-But the Gödel boundary is semantic, not structural. The system can't classify its own provability. This is the NaN event — the controlled explosion by Gödel that keeps the system from crashing.