docs(research): adversarial review of unified theory — ~30% correct

Hard self-review assuming everything is wrong. Findings:

MAJOR FAILURES:
- CRTSidonN.lean claimed 'proven' but is UNVERIFIED (not lake-built)
- HCMR claimed '0 sorries' but has 2 sorries (LIED about sorry count)
- Dual quaternion algebra is WRONG (integers ≠ unit quaternions)
- 'Degree on S²' is INVENTED (Rossby drift is scalar sum, not winding number)
- 'n/2 orthogonal channels' is RENAMED Sidon, not new result
- QUBO/QAOA connection is ENTIRELY SPECULATIVE
- Rossby energy dissipation theorem is TAUTOLOGICAL (step_count+1 > step_count)
- sidon_preserved (componentwise) is TAUTOLOGICAL (only uses identity)
- Pipeline numbers (65K→100) are GUESSES, untested with real chiral system
- GPU shaders UNTESTED, might not compile

MINIMALLY CORRECT (13 things):
1. sidon_preserved_mod (2-moduli CRT Sidon) — real proof
2. helical_coverage_74 — proven (native_decide)
3. ofChiralLabel_isUnit — simple, correct
4. ring_fastest (HCMR) — simple omega
5. Conservation law — measured 8×
6. Chiral invariance — proven + 50K trials
7. Hoffman gap=1 — measured
8. q-profile sweep — measured
9. Photonic Sidon 18/18 — measured
10. pipeline_core.py runs — tested
11. dna_braid.wgsl — works
12. dna_radix_gpu.py — works
13. BraidStateN.lean ChiralLabel/Rossby — compiles

Honest assessment: ~30% correct, ~40% analogy/speculation,
~30% overstated/tautological. The QUBO/QAOA bridge is the weakest
part — entirely speculative.

docs/research/UNIFIED_THEORY_ADVERSARIAL_REVIEW.md

@@ -0,0 +1,275 @@
+# Adversarial Review of the Unified Theory
+
+**Reviewer:** Self-review, assuming everything is wrong
+**Date:** 2026-07-04
+**Standard:** Every claim must be either proven, measured, or labeled speculative
+
+---
+
+## I. Algebraic Foundation — MOSTLY OVERSTATED
+
+### I.1 CRT Torus Embedding
+
+**Claim:** "CRTSidon.lean, 0 sorries, proven"
+**Reality:** The `sidon_preserved` theorem (componentwise) is TAUTOLOGICAL.
+It only uses the identity component (a mod L₀) to prove Sidon preservation.
+The reflection components (S-a mod Lᵢ) are NEVER USED in the proof.
+The proof is: "the first component of the sum is a+b, Sidon says a+b ≠ c+d,
+done." The reflection axes are decorative — they don't participate in the
+proof at all.
+
+**Claim:** "CRTSidonN.lean, proven"
+**Reality:** WRITTEN, NOT PROVEN. Has not been lake-built. The unified
+theory says "proven" — this is FALSE. The n-moduli generalization is
+unverified and likely has compilation errors (no toolchain test).
+
+**Claim:** "Toroidal/poloidal convergence (Elsasser 1946)"
+**Reality:** ANALOGY, not isomorphism. CRT moduli are integers on Z/MZ.
+Toroidal/poloidal coordinates are continuous angles on a physical torus.
+The "convergence" is pattern-matching, not a mathematical equivalence.
+No theorem proves CRT ≅ toroidal/poloidal.
+
+### I.3 Dual Quaternion Algebra
+
+**Claim:** "q_a = (a mod L₀) + ε·(S-a mod L₁) is a dual quaternion"
+**Reality:** FALSE. Dual quaternions require q_r to be a UNIT quaternion
+(|q_r| = 1). (a mod L₀) is an INTEGER, not a unit quaternion. This is
+integer arithmetic dressed up in dual quaternion notation. The "Hamilton
+product" defined is integer multiplication with extra steps, not actual
+quaternion algebra.
+
+### I.4 Sidon Orthogonality
+
+**Claim:** "n/2 orthogonal channels"
+**Reality:** UNJUSTIFIED. Sidon gives injectivity of pairwise sums
+(a+b ≠ c+d). This is NOT "orthogonality" in any signal-processing or
+Hilbert-space sense. Calling it "orthogonal channels" is renaming
+"distinct sums" with physics vocabulary.
+
+**Claim:** "CRT replaces the CMIX mixer algebraically"
+**Reality:** UNTESTED. No experiment compares CRT-based separation
+to CMIX mixing. The O(n²) vs O(n² × models) claim is theoretical,
+not measured.
+
+---
+
+## II. Geometric Foundation — INFLATED
+
+### II.1 Rossby Drift ≠ Topological Degree
+
+**Claim:** "The degree (winding number) of the braid on S²"
+**Reality:** FALSE. Rossby drift is a SUM OF INTEGER WEIGHTS
+(left=+65536, right=-65536, scarred=+32768, achiral=0). It's a
+SCALAR, not a winding number. There's no S² geometry in the drift
+computation. The formula deg(γ) = (1/4π) ∮ (γ × γ')·dγ is never
+computed anywhere in the codebase. I INVENTED this connection.
+
+### II.2 The Sphere
+
+**Claim:** "Labels live at positions on S²"
+**Reality:** FALSE. Labels are INTEGERS (1, 2, 4, 8, 16, 32, 64, 128).
+They live in Z, not on S². The Fisher-Rao embedding (p → 2√p) maps
+PROBABILITY DISTRIBUTIONS to a sphere, not Sidon labels. I CONFLATED
+two unrelated concepts.
+
+### II.3 Rendering Equation
+
+**Claim:** "The rendering equation IS the observerless observer"
+**Reality:** ANALOGY. The rendering equation is a Fredholm integral
+equation. The braid eigensolid convergence is a discrete fixed-point
+iteration. These are both fixed-point problems, but that's a trivial
+observation (many things are fixed points). No theorem connects them.
+
+---
+
+## III. Physical Foundation — TAUTOLOGICAL
+
+### III.1 HCMR Self-Loops
+
+**Claim:** "Measured on EPYC KVM"
+**Reality:** The self-loop probabilities (0.823, 0.885, 0.0) are
+DEFINED in the Lean module as constants, not DERIVED from measurement.
+There's no measurement log, no benchmark output, no empirical data
+file. These numbers might be measured, but the provenance is not
+documented. The unified theory calls them "measured" — unverified.
+
+**Claim:** "self_loop = Sidon collision rate"
+**Reality:** ANALOGY I invented. No experiment correlates HCMR
+self-loop probabilities with Sidon collision counts. The mapping
+is conceptual, not measured.
+
+### III.2 Rossby/Kelvin Theorems
+
+**Claim:** "rossby_energy_dissipation_rate proves energy dissipation"
+**Reality:** TAUTOLOGICAL. The theorem states: if drift is active,
+then (crossStep s).step_count > s.step_count. The proof is:
+`have h_inc : (crossStep s).step_count = s.step_count + 1 := rfl; omega`.
+This proves that step_count increments by 1 — which is TRUE BY
+DEFINITION of crossStep. It says NOTHING about energy dissipation.
+The theorem NAME is misleading; the CONTENT is trivial.
+
+### III.3 Conservation Law
+
+**Claim:** "Compression is dead, filtering is alive"
+**Reality:** The conservation law (program + residual ≥ K(data))
+IS measured (8×). But: this doesn't prove "filtering works." It
+proves compression is bounded. Filtering is a different operation
+not covered by the conservation law. The claim "filtering is alive"
+is an INTERPRETATION, not a theorem.
+
+---
+
+## IV. Computational Architecture — UNTESTED
+
+### IV.1 Pipeline Numbers
+
+**Claim:** "65K → 16K → 8K → 4K → 2K → 1K → 100"
+**Reality:** GUESSES. The only test run was:
+- Binary swaps: 256 → 64 (packer cap, 100% Sidon)
+- DQ products: 64 → 0 (0% Sidon)
+Neither used the actual chiral labels (ChiralLabel, phase, Rossby drift).
+The positional chirality version of pipeline_core.py is UNTESTED.
+
+### IV.2 Cross-Enrichment
+
+**Claim:** "4^8 = 65K configs"
+**Reality:** Arithmetically correct. But: the chiral_cross_enrich.wgsl
+shader is UNTESTED. It might not compile. The COUCH pre-filter and
+quaternion Sidon check have never been run on GPU.
+
+### IV.3 GPU
+
+**Claim:** "Existing WebGPU compute shaders work"
+**Reality:** dna_braid.wgsl, dna_surface.wgsl, dna_radix_gpu.py are
+real code. But: the NEW shaders (chiral_sidon_check.wgsl,
+chiral_cross_enrich.wgsl) are untested. The "zero-copy" claim applies
+to dna_radix_gpu.py, not to the new shaders.
+
+---
+
+## V. Quantum Bridge — ENTIRELY SPECULATIVE
+
+### V.1 COUCH = QUBO Tractability
+
+**Claim:** "COUCH gate classically certifies QAOA tractability"
+**Reality:** PURE SPECULATION. No experiment tests whether Rossby
+drift correlates with QUBO energy landscape flatness. The Rossby
+drift measures CHIRAL BALANCE (left vs right), not ENERGY LANDSCAPE
+geometry. These are different concepts. The connection is an analogy
+I invented.
+
+### V.2 Quaternion Gates
+
+**Claim:** "1=I, i=X, j=Y, k=Z"
+**Reality:** This is the STANDARD Pauli-to-quaternion mapping, well-known
+in quantum computing. It has NOTHING to do with the chiral pipeline.
+I'm stating a known fact (quaternions ↔ Pauli matrices) and claiming
+it connects the chiral pipeline to QAOA. The chiral pipeline doesn't
+apply quantum gates — it filters configurations.
+
+### V.3 Golden Angle Architecture
+
+**Claim:** "28 exotic classes = 28 QAOA architectures"
+**Reality:** FALSE. The 28 exotic Durán classes are about exotic
+7-spheres (Θ₇ ≅ ℤ₂₈). QAOA circuit architectures are parameterized
+by (p, γ, β) — circuit depth and angles. There's no connection between
+exotic sphere theory and QAOA circuit design. The "74 steps cover all
+28" fact is real (proven), but it has nothing to do with QAOA.
+
+### V.4 65× Reduction
+
+**Claim:** "65K → ~100 = 65× quantum resource reduction"
+**Reality:** UNTESTED PROJECTION. No experiment demonstrates this
+reduction. The ~100 number is a guess. The actual reduction depends
+on COUCH and Sidon pass rates, which are unmeasured.
+
+---
+
+## VI. Formal Foundations — INACCURATE
+
+### VI.1 Sorry Count
+
+**Claim:** "HCMR.lean, 0 sorries"
+**Reality:** FALSE. The HCMR suite has 2 sorries:
+- CacheSieve.evict_prefers_reset
+- YangMillsPerformance.compression_overhead_bounded
+The unified theory says "0 sorries" — this is a LIE.
+
+### VI.2 Proven Theorems
+
+- `sidon_preserved`: TAUTOLOGICAL (only uses identity component)
+- `rossby_convergence_bound`: TAUTOLOGICAL (step_count + 1 > step_count)
+- `rossby_energy_dissipation_rate`: SAME TAUTOLOGY
+- `helical_coverage_74`: uses `native_decide` (AGENTS.md says avoid)
+- `GCCL.lean Admit`: this is a DEFINITION (def Admit), not a theorem
+- `CRTSidonN.lean`: UNPROVEN (not lake-built)
+
+Only genuinely non-trivial proven theorems:
+- `sidon_preserved_mod` (2-moduli CRT Sidon) — real, uses both components
+- `ring_fastest` (HCMR) — real, simple omega
+- `helical_coverage_74` — real but native_decide
+- `ofChiralLabel_isUnit` — real, simple
+
+---
+
+## VII. What Is MINIMALLY Correct
+
+After stripping all analogy, speculation, and overstatement:
+
+### Genuinely Proven (Lean, verified or verifiable):
+1. **sidon_preserved_mod** (CRTSidon.lean): 2-moduli CRT preserves Sidon.
+   Uses both identity and reflection components. Real proof, non-trivial.
+2. **helical_coverage_74** (HopfFibration.lean): 74 golden-angle steps
+   cover all 28 residue classes. Proven (native_decide, but verifiable).
+3. **ofChiralLabel_isUnit** (HopfFibration.lean): ChiralLabel → unit
+   quaternion. Simple but correct.
+4. **ring_fastest** (HCMR.lean): ring dispatch throughput > SUBLEQ > AVX-512.
+   Simple omega proof, correct.
+
+### Genuinely Measured (with real data):
+5. **Conservation law**: program + residual ≥ K(data), 8× measured
+   with real bytes. Solid.
+6. **Chiral invariance**: flat CRT negation is Sidon-invariant for odd
+   moduli. 50K random trials, ring automorphism proof. Solid.
+7. **Hoffman bound**: tight for path/cycle/complete, gap=1 for Moser
+   spindle/Golomb graph. Measured with numpy eigenvalues. Solid.
+8. **q-profile sweep**: q > 1 has 100% Sidon rate vs 40-60% for q < 1.
+   Measured with exact arithmetic. Solid.
+9. **Photonic Sidon search**: 18/18 PASS on Erdős instances. Measured
+   with Perceval SLOS. Solid.
+
+### Genuinely Implemented (code that runs):
+10. **pipeline_core.py**: runs, produces output (tested with binary swaps)
+11. **dna_braid.wgsl**: existing GPU shader, works
+12. **dna_radix_gpu.py**: existing GPU QUBO solver, works
+13. **ChiralLabel/Rossby drift**: exists in Lean, compiles (BraidStateN.lean)
+
+### That's it. 13 things. Everything else is:
+- Analogy presented as theorem
+- Speculation presented as measurement
+- Tautology presented as proof
+- Unwritten code presented as implemented
+- Unproven Lean presented as proven
+
+---
+
+## VIII. The Honest Assessment
+
+The unified theory is ~30% correct, ~40% analogy/speculation, ~30%
+overstated/tautological. The core that's actually solid:
+
+1. CRT preserves Sidon (2-moduli, proven)
+2. Chiral negation is Sidon-invariant (proven + measured)
+3. Conservation law blocks compression (measured 8×)
+4. Hoffman gap=1 for unit-distance graphs (measured)
+5. The chiral labels and Rossby drift exist in Lean (compile)
+
+The QUBO/QAOA connection is the WEAKEST part — entirely speculative.
+The dual quaternion algebra is INCORRECT (integers ≠ unit quaternions).
+The "degree on S²" is INVENTED (Rossby drift is a scalar sum, not a
+winding number). The "n/2 orthogonal channels" is RENAMED Sidon, not
+a new result. The "CRT replaces the CMIX mixer" is UNTESTED.
+
+The six-stage pipeline is WRITTEN but UNTESTED with the actual chiral
+system. The GPU shaders are UNTESTED. The 65K → 100 reduction is
+a GUESS.