Files
LithosAnanake/docs/working/papers/ANTI_CLAIMS.md
T

13 KiB
Raw Blame History

Anti-Claims: What This Work Does NOT Claim

Version: 1.0 Date: 2025-12-14 Purpose: Explicit boundaries to prevent strawman attacks and over-interpretation


PURPOSE OF THIS DOCUMENT

This document explicitly states what StarForth does NOT claim to prevent:

  1. Strawman arguments ("You claim X" when we never claimed X)
  2. Over-interpretation by enthusiasts
  3. Scope creep in peer review
  4. Misleading comparisons with unrelated work

Principle: What is not forbidden is compulsory. By stating what we DON'T claim, we clarify what we DO claim.


I. PHYSICS & THERMODYNAMICS

NOT CLAIMED: "This is a physical theory"

We DO claim:

  • Thermodynamic metaphor for execution frequency dynamics
  • Mathematical similarity between heat equations and decay models
  • Useful conceptual framework for reasoning about adaptive systems

We DO NOT claim:

  • Actual thermodynamic processes occur in the CPU
  • Physical laws govern code execution
  • Quantum effects are involved
  • Energy conservation applies to execution frequency

Why This Matters: The thermodynamic framework is a conceptual tool, not a physics paper. Reviewers from physics should not evaluate this as a physical theory.


NOT CLAIMED: "Execution frequency is thermal energy"

We DO claim:

  • Execution frequency behaves analogously to thermal energy
  • Exponential decay resembles heat dissipation
  • Equilibrium convergence mirrors thermodynamic equilibrium

We DO NOT claim:

  • Frequency counters measure actual heat
  • Temperature sensors are involved
  • Joules or calories are relevant units
  • The Second Law of Thermodynamics applies

Precise Language:

  • "Frequency evolves like heat in a cooling system"
  • "Frequency is heat"

II. OPTIMALITY & PERFORMANCE

NOT CLAIMED: "This is optimal"

We DO claim:

  • 25.4% performance improvement over baseline
  • Convergence toward better configurations
  • Adaptive tuning based on statistical inference

We DO NOT claim:

  • This is the best possible adaptive runtime
  • No other approach could perform better
  • Optimality is proven mathematically
  • This beats all other VMs

Why This Matters: We show a working adaptive system, not the globally optimal one.


NOT CLAIMED: "This outperforms JIT compilers"

We DO claim:

  • Comparable optimization strategy (frequency-based specialization)
  • Deterministic behavior (JITs typically aren't)
  • Formal verification potential (JITs lack this)

We DO NOT claim:

  • Faster runtime than PyPy, LuaJIT, or HotSpot
  • Better code generation
  • Superior optimization heuristics

Why This Matters: Our contribution is deterministic adaptation, not raw speed.


NOT CLAIMED: "This works for all workloads"

We DO claim:

  • Works for CPU-bound, deterministic FORTH programs
  • Validated on recursive algorithms (Fibonacci, Ackermann)
  • Generalizes across workload shapes (Zipf exponent 0.81.5)

We DO NOT claim:

  • Works for I/O-bound workloads
  • Works for non-deterministic programs (e.g., random number generators)
  • Works for adversarial execution patterns
  • Works for very short-lived processes (<1000 iterations)

Why This Matters: See NEGATIVE_RESULTS.md for explicit failure modes.


III. MACHINE LEARNING & AI

NOT CLAIMED: "This is AI" or "This is machine learning"

We DO claim:

  • Statistical inference (ANOVA, Levene's test, exponential regression)
  • Adaptive parameter tuning
  • Data-driven optimization

We DO NOT claim:

  • Neural networks are involved
  • Gradient descent or backpropagation
  • Training on labeled datasets
  • Deep learning or reinforcement learning

Precise Language:

  • "Statistically-inferred adaptive tuning"
  • "AI-driven optimization"

NOT CLAIMED: "The system learns"

We DO claim:

  • The system adapts to execution patterns
  • Parameters converge to steady state
  • Feedback loops stabilize metrics

We DO NOT claim:

  • Supervised learning occurs
  • The VM generalizes across programs
  • Knowledge transfer between workloads

Precise Language:

  • "Adaptive inference"
  • "Learning algorithm"

IV. NOVELTY & PRIOR ART

NOT CLAIMED: "This is the first adaptive VM"

We DO claim:

  • First adaptive VM with 0% algorithmic variance
  • First to combine adaptation + determinism + formal verification potential
  • Novel application of statistical inference to VM tuning

We DO NOT claim:

  • First to use execution frequency tracking (profilers do this)
  • First to cache hot code (JITs do this)
  • First to use exponential decay (LRU caches do this)

Why This Matters: Our novelty is the combination and verification approach, not individual techniques.


NOT CLAIMED: "This replaces JIT compilers"

We DO claim:

  • Alternative approach with different trade-offs
  • Determinism at the cost of potential peak performance
  • Suitable for safety-critical systems requiring verification

We DO NOT claim:

  • JITs are obsolete
  • This should replace LLVM or V8
  • Compilation is unnecessary

Why This Matters: We target a different niche (verifiable adaptive systems), not mainstream VMs.


V. FORMAL VERIFICATION

NOT CLAIMED: "The entire system is formally verified"

We DO claim:

  • Deterministic behavior is empirically validated (0% CV)
  • Convergence theorems are stated (not fully proven)

We DO NOT claim:

  • Complete proof of correctness in Coq/Isabelle
  • Verified compiler toolchain (CompCert-style)
  • Mechanized proofs for all 7 feedback loops

Current Status: Partial formal verification (proofs in progress, see docs/src/internal/formal/)


NOT CLAIMED: "Zero bugs" or "Provably correct"

We DO claim:

  • 780+ tests pass
  • FORTH-79 compliance validated
  • No known correctness bugs in core interpreter

We DO NOT claim:

  • Bug-free implementation
  • Exhaustive testing
  • Formal proof of absence of errors

Why This Matters: We provide high assurance, not mathematical proof of perfection.


VI. CAUSATION & INTERPRETATION

NOT CLAIMED: "We prove causation"

We DO claim:

  • Strong correlation between adaptive mechanisms and performance
  • Functional relationship fits exponential decay model (R² > 0.95)
  • Configuration-dependent convergence (only C_FULL improves)

We DO NOT claim:

  • Causal proof via randomized controlled trial
  • Mechanistic explanation of why it works
  • Guaranteed causation (only correlation + functional fit)

Precise Language:

  • "Adaptive mechanisms are associated with 25.4% improvement"
  • "Adaptive mechanisms cause 25.4% improvement"

NOT CLAIMED: "This explains computation"

We DO claim:

  • Useful model for VM behavior
  • Predictive framework for performance
  • Mathematical characterization of adaptation

We DO NOT claim:

  • Fundamental theory of computation
  • Replacement for computational complexity theory
  • New model of computation (Turing-equivalent)

Why This Matters: This is a VM optimization technique, not a theory of computation.


VII. GENERALIZATION & APPLICABILITY

NOT CLAIMED: "This generalizes to all languages"

We DO claim:

  • Principles may generalize to other interpreters
  • Hypothesize applicability to Lua, Python, JavaScript
  • Conceptual framework is language-agnostic

We DO NOT claim:

  • Proven to work for compiled languages
  • Applicable to GPU or quantum computing
  • Works for non-stack-based architectures

Future Work: Cross-language validation needed (see SCIENTIFIC_DEFENSE_CHECKLIST.md)


NOT CLAIMED: "This scales to arbitrary program sizes"

We DO claim:

  • Validated on programs up to ~2.1M word executions
  • Dictionary sizes up to ~500 entries
  • Workloads with moderate complexity

We DO NOT claim:

  • Scales to million-line codebases
  • Handles programs with 100K+ dictionary entries
  • Tested on real-world production systems

Known Limitation: Scalability to very large programs unvalidated


VIII. SYSTEMS & DEPLOYMENT

NOT CLAIMED: "This is production-ready"

We DO claim:

  • Research prototype demonstrating feasibility
  • Suitable for experimental validation
  • Platform for exploring adaptive runtime ideas

We DO NOT claim:

  • Production-grade implementation
  • Enterprise support or SLA
  • Hardened against all security threats

Current Status: Research prototype (not production system)


NOT CLAIMED: "This replaces operating systems"

We DO claim:

  • StarForth → StarKernel → StarshipOS is a roadmap
  • Vision for FORTH-based kernel
  • Exploration of alternative OS architecture

We DO NOT claim:

  • StarshipOS is complete or functional
  • This replaces Linux/Windows
  • Production OS deployment imminent

Why This Matters: The OS vision is aspirational, not a current product claim.


IX. STATISTICAL CLAIMS

NOT CLAIMED: "Zero variance in all metrics"

We DO claim:

  • 0.00% CV in algorithmic decisions (cache hits, dictionary lookups)
  • Deterministic execution of adaptive mechanisms
  • Variance decomposition separates algorithm from environment

We DO NOT claim:

  • Zero variance in wall-clock runtime (measured 60-70% CV due to OS)
  • Zero variance in power consumption
  • Zero variance in memory allocation

Precise Language:

  • "Algorithmic variance: 0.00% CV"
  • "Total system variance: 0.00% CV"

NOT CLAIMED: "100% confidence in results"

We DO claim:

  • 95% confidence intervals for performance
  • p < 10⁻³⁰ statistical significance for determinism
  • High empirical confidence based on 90 runs

We DO NOT claim:

  • Absolute certainty
  • 100% confidence (Bayesian posterior ≈ 1 - 10⁻³⁰, not 1.0)
  • Impossibility of replication failure

Why This Matters: Science deals in probabilities, not certainties.


X. SCOPE LIMITATIONS

NOT CLAIMED: "This solves the halting problem" (or other impossible claims)

We DO claim:

  • Adaptive runtime can converge to steady state for terminating programs
  • Performance prediction for analyzable workloads

We DO NOT claim:

  • Decidability of convergence for all programs
  • Solving any computationally undecidable problem

NOT CLAIMED: "This is quantum computing" or "This uses blockchain"

We DO NOT use:

  • Quantum superposition or entanglement
  • Blockchain or distributed ledger
  • Cryptocurrency or smart contracts
  • Neuromorphic computing

Why This Matters: Buzzword bingo is not our game. We use classical algorithms on classical hardware.


XI. COMPARATIVE CLAIMS

NOT CLAIMED: "Better than [specific system X]"

We DO claim:

  • Different trade-offs than JIT compilers (determinism vs. peak speed)
  • Novel combination of techniques

We DO NOT claim:

  • Faster than PyPy
  • Better than LuaJIT
  • Superior to HotSpot
  • Replacement for any specific system

Why This Matters: We avoid direct performance shootouts. Our contribution is deterministic adaptation, not raw speed records.


XII. USAGE IN PEER REVIEW

How to Use This Document

When a reviewer says: "You claim that [X]"

Your response:

  1. Check if X is in this Anti-Claims document
  2. If yes: "We explicitly do NOT claim X. See ANTI_CLAIMS.md, Section Y."
  3. If no: Point to actual claim in FORMAL_CLAIMS_FOR_REVIEWERS.txt

Example:

Reviewer: "You claim this is the first adaptive VM, but PyPy exists."

Response: "We do NOT claim to be the first adaptive VM. See ANTI_CLAIMS.md, Section IV. Our claim is: first adaptive VM with 0% algorithmic variance and formal verification potential. See FORMAL_CLAIMS_FOR_REVIEWERS.txt, Claim 1."


XIII. SUMMARY TABLE

Category What We Claim What We DON'T Claim
Physics Thermodynamic metaphor Actual physical theory
Performance 25.4% improvement Global optimality
Novelty Deterministic adaptation First adaptive system
Verification Empirical validation Complete formal proof
ML/AI Statistical inference Neural networks or learning
Causation Strong correlation Proven causation
Generality Works for FORTH Works for all languages
Variance Algorithmic: 0% CV Total system: 0% CV

XIV. PRINCIPLE: INTELLECTUAL HONESTY

Our Commitment:

  • We state limitations explicitly
  • We acknowledge prior work
  • We avoid over-claiming
  • We invite falsification

What We Ask from Critics:

  • Criticize what we actually claim (see FORMAL_CLAIMS_FOR_REVIEWERS.txt)
  • Don't attack claims we never made (see this document)
  • Attempt independent replication (see REPLICATION_INVITE.md)
  • Engage with the evidence, not caricatures

XV. VERSION CONTROL

This document is versioned alongside code. If our claims evolve:

  • Document changes in git history
  • Update this file to reflect new boundaries
  • Never silently delete anti-claims

Transparency: Past versions remain in git history.


Conclusion: By explicitly stating what we do NOT claim, we establish clear intellectual boundaries and prevent misinterpretation. This is defensive honesty.

License: See ./LICENSE