Research-Stack/4-Infrastructure/shim/VERIFICATION_PROTOCOL.md
Brandon Schneider 0cf775c80e collapse: prover orchestration layers, FAMM verilator harness, swarm topological prober, spec sheets, virtual FPGA system tests, merge conflict resolution
- Prover-Integrated Orchestration Layers (L0-L3): Goedel-Prover-V2 watchdog, BFS-Prover-V2 swarm consensus, bf4prover topology adaptation
- FAMM Verilator benchmark: uniform vs preshaped delay comparison (4.4x speedup)
- Swarm topological device prober: 11 agents probing traces, caps, delays, errors, vias, PDN
- Spec sheet puller: 10 components with key params and topological relevance
- Virtual FPGA system tests: 6/6 passed, 134K ops/s throughput
- Fixed merge conflicts in AI-Newton test_experiment.ipynb
2026-05-06 23:42:01 -05:00

5.1 KiB
Raw Blame History

F01-F12 Verification Protocol

Layer 1: Mathematical Verification (Wolfram Alpha)

Every equation verified against Wolfram Alpha:

Input: 121.567 * 65536
Wolfram: 7,967,421.952
Rounded: 7,967,422
Q16.16: 0x0079.9120

Required for each F01-F12:

  • Closed-form solution where available
  • Numerical test with 3+ input sets
  • Edge cases (zero, infinity, singularities)
  • Document Wolfram query in comment

Layer 2: Computational Verification (Lean 4)

Every def requires:

def encodeWavelength (λ : Q16_16) : Q16_16 := ...

#eval encodeWavelength (Q16_16.ofFloat 121.567)
-- Expected: 0x0079.9120
-- Wolfram verified: 121.567 * 65536 = 7,967,422

Every def requires theorem:

theorem encodeWavelength_total (λ : Q16_16) : 
  ∃ result, encodeWavelength λ = result := by 
  simp [encodeWavelength]

Build must pass:

cd 0-Core-Formalism/lean/Semantics && lake build

Layer 3: Determinism Verification

Test: Same inputs → Same outputs (always)

def test_determinism():
    for i in range(1000):
        result = f01_encode(121.567)
        assert result == 0x0079.9120

Failure modes:

  • Non-deterministic: Reject
  • Undefined behavior: Reject
  • Platform-dependent: Reject

Layer 4: Symbolic Stripping Verification

Test: Remove all semantic content, verify computation

Original: encode_hydrogen_wavelength(λ)
Stripped: N_7 = floor(N_0 * 65536 + 32768) / 65536

Test: N_0 = 121.567 → N_7 = 0x0079.9120
Pass: Computation identical with/without names

Required: 100% pass rate on adversarial stripping test.


Layer 5: Property-Based Testing

QuickCheck-style random testing:

theorem encode_roundtrip (λ : Q16_16) (h : λ ≥ 0) :
  |decode (encode λ) - λ| < 0.0001 := by 
  -- Property: encode then decode ≈ original

Test properties:

  • Roundtrip (encode→decode ≈ original)
  • Monotonicity (x < y → f(x) < f(y))
  • Boundedness (output in valid range)
  • Idempotence (f(f(x)) = f(x) where applicable)

Layer 6: Cross-Implementation Verification

Two independent implementations:

  1. Lean 4 (canonical): 0-Core-Formalism/lean/Semantics/
  2. Python (reference): 4-Infrastructure/shim/validation/

Verify bit-identical outputs:

def test_cross_impl():
    for input_val in test_vectors:
        lean_result = lean_call(input_val)
        py_result = python_call(input_val)
        assert lean_result == py_result  # Bit-identical

Layer 7: Statistical Verification (6.5σ)

For claims with statistical component:

Claim: Compression ratio > baseline
Test: 1000 random inputs
Mean: μ, Std: σ
Achieved: x = 6.5σ above baseline?
Required: Yes for 6.5σ claim

Standard: 6.5σ = 99.999999992% confidence
Minimum: 5σ = 99.9999427% confidence
Below 5σ: REJECT


Verification Checklist per FXX

Layer Test Pass Criteria Status
1 Wolfram Alpha 3+ numerical cases verified
2 Lean #eval All public functions have #eval
3 Totality theorem Every def has totality proof
4 Determinism 1000 iterations, identical outputs
5 Symbolic stripping Passes adversarial test
6 Property tests Roundtrip, monotonicity, bounds
7 Cross-impl Bit-identical with Python shim
8 Statistical 6.5σ where applicable
9 Lake build lake build passes
10 No sorry Zero sorry in code

All 10 must pass for VALIDATED status.


Automated Verification Script

#!/bin/bash
# verify_fxx.sh

FXX=$1

echo "Verifying $FXX..."

# Layer 2: Lean build
cd 0-Core-Formalism/lean/Semantics
if ! lake build; then
    echo "FAIL: lake build"
    exit 1
fi

# Layer 4: Adversarial stripping
if ! python3 4-Infrastructure/shim/adversarial_symbolic_stripping.py --test $FXX; then
    echo "FAIL: symbolic stripping"
    exit 1
fi

# Layer 6: Cross-implementation
if ! python3 4-Infrastructure/shim/cross_verify.py --lean $FXX; then
    echo "FAIL: cross-implementation"
    exit 1
fi

echo "PASS: $FXX verified"

Best Verification: Compositional

Verify components, compose to system:

F01 verified + F02 verified + ... + F12 verified
    ↓
Compose: F01∘F02∘...∘F12
    ↓
Verify composition: Master equation

Do not: Verify only final output
Do: Verify each FXX independently, then verify composition


Summary

Best verification = 10-layer compositional checking:

  1. Wolfram Alpha (math)
  2. Lean #eval (compute)
  3. Totality theorems (total functions)
  4. Determinism (1000 iterations)
  5. Symbolic stripping (no semantic deps)
  6. Property-based (random testing)
  7. Cross-implementation (bit-identical)
  8. Statistical (6.5σ)
  9. Lake build (compiles)
  10. No sorry (proven correct)

10/10 required for publication.

Current F01-F12 status: 0/10


Document ID: VERIFICATION-PROTOCOL-2026-05-06
Standard: 10-layer compositional verification
Target: 100% pass rate for all F01-F12
Current: 0% — blocked on F01-F12 formalization