Betsy

A Plano-Hemispherical Optical Cavity as a Volumetric Computing Substrate

March 2026

Betsy Architecture (Plano-Hemispherical Optical Reservoir) — Concept Document v1.10

Document Status: Speculative concept document. Not a research proposal or development roadmap. v1.8 incorporates responses to structured red team review (v1.6) and external Perplexity review of v1.7. New in this version: Minimum Viable Falsification section (Section 2), Notation Glossary (Section 3), T2 Simulation Specification (Section 7.5), Rank Scaling Hypothesis (Section 7.6), Null Hypothesis (Section 7.7), Noise and Drift analysis (Section 7.8), Reset/Initialization placeholder (Section 12.1), fabrication readiness matrix (Section 18), P3 cost/risk gate (Section 18.4), and three appendices (Outreach Templates, Control Architecture Comparison, What Would Convince a Skeptic).

v1.10 patch: Two targeted additions from fiber laser technology review: (1) frequency comb parallel injection architecture added to Section 7.5.1 as an alternative to sequential sweep, based on few-cycle fiber frequency comb work by Daniel Lesko (FAU Erlangen-Nürnberg / LMU München); (2) femtosecond fiber laser ablation added to Section 19.1 as a candidate route for sub-wavelength porosity in CVD diamond. Lesko's group added to Section 20 as a potential collaboration contact. Noise and drift table updated with comb-specific stability note.

Review provenance disclosure: The wave-chaos framework in this document was developed through multi-session analysis involving Claude (Anthropic) and GPT-4 relayed through the author, with Perplexity providing additional independent review of v1.7. The author serves as integrating judgment layer. This framework has not been reviewed by a wave-chaos physicist, photonic reservoir computing specialist, or CVD diamond fabrication expert. External domain expert review is the next validation step before this document is shown to any technical audience.

ESTABLISHED PHYSICS

PHYSICALLY MOTIVATED ESTIMATE

DOES NOT EXIST — MUST BE CREATED

CANDIDATE BENCHMARK TASK

RED TEAM / EXTERNAL REVIEW RESPONSE

Existential Questions — Read This First
Minimum Viable Falsification new in v1.8
Notation Glossary new in v1.8
Abstract
Intellectual Origin and Development Arc
Physical Architecture
The Formal Physics Framework
1. Billiard Classification
2. Modal Overlap Parameter M
3. M vs rank(H_coupling) — The Critical Distinction
4. The Reverberation Chamber Analogy and RCM
5. T2 Simulation Specification — Defining the Physics Kill new in v1.8
6. Rank Scaling Hypothesis new in v1.8
7. Null Hypothesis new in v1.8
8. Noise and Drift Analysis new in v1.8
The Computational Primitive: Resonate, Modulate, Collapse
The Architectural Identity: Spatiotemporal Reservoir with Engineered Disorder
The Benchmark Task — Nonlinear Channel Equalization
Comparison to Multimode Fiber Reservoirs
The Robotics Argument
The Collective Architecture
Intrinsic Learning
The Calcium Problem — Graphene, Ferroelectrics, and the Synapse
Mathematical Skeleton
What Betsy Cannot Do
Differentiation from Existing Paradigms
Fabrication Reality readiness matrix and P3 gate new in v1.8
Supply Chain and Collaboration Landscape
Graphene Integration — Companion Research Program
Open Questions Register
Relationship to the Primary Research Program
25. Application Landscape and Mission Profile new in v1.11
Appendix A — Outreach Templates new in v1.8
Appendix B — Control Architecture Comparison new in v1.8
Appendix C — What Would Convince a Skeptic new in v1.8

1. Existential Questions — Read This First

The architecture is physically grounded and internally consistent. It may also be impossible. These questions determine whether it is worth building before any fabrication is attempted.

T1 — Kill condition (simulation, run before P3): If the stacked M(k) simulation — run with actual cavity eigenmodes, scatterer density optimized for Berry-Robnik regime — shows r_eff(N_f = 50; ε) ≤ 50, the architecture does not achieve more than one useful independent channel per nominal node and is unlikely to outperform existing photonic reservoirs on nonlinear channel equalization. Betsy stops here.

T1 — Answered: Spatial rank(H_coupling) ≈ 12 (RWM surrogate, directionally correct, methodologically imperfect). M ≈ 10⁴–10⁶ after graphene Q correction — deep in high-entropy regime. Mode density is not the constraint.

T2 — Decision point (experiment): If P3 planar experiment produces SER equal to or worse than equivalent ring-resonator reservoir baseline, the performance advantage does not exist. Architecture requires revision or abandonment.

T3 — Demotion (not kill): If Option C ferroelectric learning proves unphysical, the strongest MMF differentiation argument collapses. Architecture continues as a programmable disorder reservoir — a weaker but still legitimate claim.

2. Minimum Viable Falsification

A concept document without stated falsification criteria is a narrative, not a scientific proposal. This section states, in plain terms, what kills the plan, what demotes it, and what justifies proceeding to P3. It is intended to be read by a skeptical collaborator in under two minutes.

Result	Consequence	Action
r_eff(N_f=50; ε) ≤ 50 in T2 simulation	Architecture does not achieve sufficient dimensionality. Kill shot on physics grounds.	Stop. Do not proceed to P3. Revise or abandon.
Δf_corr >> Δf_step in T2	Spectral sweep samples correlated states. N_f effective channels << 50. Rank recovery fails.	Stop. Investigate smaller wavelength or smaller cells before continuing.
P3 SER ≥ ring-resonator baseline at equivalent node count	Performance claim fails experimentally. Architecture is not competitive as a reservoir.	Stop before H-track. Publish negative result. Reassess.
P3 SER < ring-resonator baseline by <10%	Marginal improvement. Performance claim technically holds but may not justify hemispherical fabrication cost.	Pause. Require >10% improvement at P3 to proceed to H-track.
P3 SER < ring-resonator baseline by >10%	Performance claim validated on planar substrate. Architecture justified for curved surface development.	Proceed to P4 (spherical cap) and H-track.
Option C ferroelectric learning unphysical	Strongest MMF differentiation argument collapses. Architecture continues as programmable disorder reservoir.	Demote — continue with weaker claim. Do not stop.
Element Six cannot supply optical grade hemisphere	Primary substrate supply chain fails.	Evaluate Diamond Foundry. Assess timeline impact. Do not stop P-track.

The P3 go/no-go threshold: r_eff(N_f=50; ε) > 50 from T2 simulation AND a pre-registered SER improvement target of >10% over ring-resonator baseline. Both conditions must be met before H-track investment is authorized.

The document's governing epistemic posture: this architecture is being optimized for disproof, not narrative momentum. Every section that names a claim also names the measurement that refutes it.

3. Notation Glossary

The following symbols are used consistently throughout this document. This table is the authoritative definition for each quantity.

Symbol	Definition	Units	First appears
M(ω_k)	Coupling matrix at frequency ω_k: maps N_input excitation channels to N_readout observables. Element M_ij(ω_k) = steady-state intensity at readout i due to unit excitation at input j.	Dimensionless (normalized)	§7.5
M_total	Vertically stacked coupling matrices across N_f frequency steps: [M(ω₁); M(ω₂); ...] ∈ ℝ^{(N_f·N_readout) × N_input}	—	§7.5
r_eff(N_f; ε)	Effective rank of M_total at singular value cutoff ε. Count of singular values σ_ℓ ≥ ε. ε is derived from the system noise floor and detector SNR.	Dimensionless integer	§7.5
H_coupling	The mapping from honeycomb cell locations to cavity mode population. Its rank determines how many independent modal degrees of freedom the input geometry accesses. Distinct from M(ω_k), which is the operational coupling matrix.	—	§7.3
N_f	Number of independent frequency steps in the spectral sweep. Target: N_f ≈ 50 across 50 GHz tuning range.	Dimensionless integer	§7.2
M	Modal overlap parameter: M = Δf_mode / δf. Classifies cavity operating regime. M >> 1 required for wave-chaotic reservoir behavior.	Dimensionless	§7.2
δf	Mean spacing between adjacent resonant frequencies. From Weyl's law: δf = c³ / (8πVf²). For this cavity: δf ≈ 7 Hz.	Hz	§7.2
Δf_mode	Resonance linewidth = f/Q. Depends on mirror reflectivity, graphene absorption, and scatterer losses. Estimated 1–20 MHz for Q ≈ 10⁵–10⁷.	Hz	§7.2
Δf_corr	Spectral correlation bandwidth: frequency separation at which average output correlation drops below threshold C_th = 0.5. Determines effective N_f.	Hz (GHz range)	§7.5
Δf_step	Frequency step size in the spectral sweep: Δf_step = Δf_sweep / N_f. Must satisfy Δf_step ≳ Δf_corr for independent samples.	Hz (GHz range)	§7.5
ε	Singular value cutoff for r_eff calculation. Derived from noise floor: ε = σ_noise / σ₁, where σ_noise is the RMS noise contribution to the readout vector. Must be pre-registered before simulation is run.	Dimensionless	§7.5
ε_scatter	Scatterer perturbation parameter (analogous to deformation ε in billiard theory). Controls chaos-to-stability ratio. Target: Berry-Robnik transition regime.	Dimensionless	§7.1
SER	Symbol Error Rate. Benchmark metric for nonlinear channel equalization. Lower is better. P3 decision threshold: SER improvement >10% over ring-resonator baseline.	Dimensionless fraction	§10

4. Abstract

The Betsy Architecture (Plano-Hemispherical Optical Reservoir) is a speculative architecture for volumetric optical computing based on a plano-hemispherical CVD diamond cavity — commercially manufactured at volume by Element Six as speaker domes, requiring optical grade specification upgrade. The architecture combines a honeycomb-patterned graphene modulator layer grown in place on the inner surface via Ni-assisted SCD(111) graphitization, deliberate dielectric scatterers for engineered mode diversity, a gold flat base with high-reflectivity dielectric mirror coating, and a conformally integrated electronic/photonic skin layer on the outer surface for distributed readout. This is classical wave physics, not quantum computation.

The architecture is a spectrally multiplexed photonic reservoir with engineered disorder — a parametrically perturbed integrable cavity in the Berry-Robnik to Wigner-Dyson transition regime. The hemispherical geometry provides a deterministic modal scaffold. Scatterers break near-integrable symmetry. A frequency sweep provides ensemble averaging equivalent to mode stirring in a reverberation chamber.

The benchmark task is nonlinear channel equalization. The performance claim: a Betsy-class cavity reservoir achieves lower symbol error rate (SER) than a ring-resonator reservoir of equivalent node count because three-dimensional mode volume and scatterer-induced mode mixing produce a flatter H_coupling singular value spectrum. The claim is testable at P3 on flat diamond before any hemisphere is built. The P3 go/no-go threshold is r_eff(N_f=50; ε) > 50 from T2 simulation AND SER improvement >10% over ring-resonator baseline.

Near-term path: Element Six optical grade hemisphere specification, Graphenea Route I graphene for planar validation, Fraunhofer IAF as physics validation partner, Ghent University / Bogaerts group as reservoir computing benchmark partner.

Named after a dear friend who, like the concept itself, was full of light and not easily defined.

5. Intellectual Origin and Development Arc

Betsy's trajectory is one of progressive constraint. Each stress test improved the architecture's precision or would have killed it.

Date	Event	Effect on architecture
July 2025	Breaking the Planar Barrier	Founded geometric intuition. Philosophically correct, fabrication-optimistic.
October 2025	3D Volumetric Computing: An Honest Assessment	Five fatal barriers identified. Betsy is what survived by relocating computation to cavity field behavior.
January 2026	Diamond-Integrated 3D Silicon Photonics	Shared material foundation established. Route F/J graphene integration path identified.
March 2026 — T1 surrogate	Effective rank 11.6	Catastrophic spatial degeneracy confirmed. Architecture redirected to spectral multiplexing and engineered disorder.
March 2026 — wave-chaos session	Formal physics framework	Billiard classification, M vs rank(H_coupling) distinction, RCM as conceptual framework, Berry-Robnik target regime established.
March 2026 — grain boundary paper	Graphene integration analysis	Route F / Route J identified as Pareto front. Curved surface physics gap formally identified.
March 2026 — red team v1.6	18 attacks, 1 kill shot, 8 wounds	Kill shot closed (nonlinear channel equalization task). Supply chain anchored. 170ps speed claim removed.
March 2026 — Perplexity review v1.7	Structural and falsification gaps identified	MVF section, T2 specification, rank scaling hypothesis, noise analysis, outreach appendices added.

6. Physical Architecture

6.1 The Cavity Geometry

The plano-hemispherical resonator is repurposed as a computational substrate in which mode competition under spatially resolved loss modulation performs useful work. At 25–30mm diameter, cavity round-trip time ≈ 170 picoseconds. This is the natural cavity dynamics timescale — not the system throughput, which is determined by the frequency sweep settling time (see Section 3 glossary).

The hemispherical cavity is near-integrable with Poisson level spacing statistics and strong modal degeneracy — exactly what the T1 simulation measured. It is retained not because it is optimal for mode diversity but because Element Six manufactures CVD diamond speaker domes at this geometry in volume. The Limaçon cavity (mixed phase space, Berry-Robnik statistics, continuous ε tuning) would be geometrically cleaner. It has no manufacturing precedent in CVD diamond at 25-30mm scale.

6.2 The Diamond Hemisphere — Supply Chain

Element Six (De Beers Group, UK) manufactures CVD diamond hemispheres at volume — over one million speaker domes shipped. Audio grade specifications are mature. Optical grade (<1nm RMS surface roughness, optical purity, wall thickness uniformity, controlled stress birefringence) is a specification upgrade, not a manufacturing capability gap. Element Six is the first supply chain conversation (see Appendix A for outreach template). Fraunhofer IAF (Germany) is the appropriate physics validation partner for optical grade characterization.

6.3 The Graphene Layer — In-Place Growth Route

In-place graphitization is preferred over transfer. Graphene grown directly on the diamond surface conforms to the cavity geometry without a curved transfer step.

Graphene has been grown conformally on non-planar surfaces: CVD on pre-curved copper foil, CVD on nickel foam producing 3D graphene networks, graphene shells on spherical nanoparticle templates. In-place growth follows substrate geometry during the growth process. The graphitization chemistry does not depend fundamentally on substrate curvature.

Route F (SCD(111) Ni-assisted graphitization with H-termination, 21.8% joint Goldilocks pass probability today, 57.2% at H-termination mobility ceiling) applied to the inner hemisphere surface would grow graphene in place. Whether Route F quality — domain size, mobility, carrier density, defect density, monolayer coverage — is maintained on a curved surface is unknown. The growth mechanism does not obviously prohibit it.

In-place graphitization on a curved CVD diamond surface has not been demonstrated. Effect of substrate curvature on all five Goldilocks quality criteria has not been measured. No graphene quality measurements on any curved diamond substrate have been reported in the literature.

6.4 Deliberate Scatterers — Engineered Disorder Layer

Sparse dielectric scatterers break the near-integrable symmetry of the hemispherical cavity, lifting mode degeneracy and coupling mode families orthogonal in the clean geometry. This is the Sinai billiard mechanism — an analogy, not a formal identity. Sinai's proof applies to a 2D billiard with a single circular obstacle. Extension to a 3D optical cavity with sparse dielectric inclusions is physically motivated but not formally proven for optical cavities. The scatterer density ε_scatter is the perturbation parameter controlling the integrable-to-chaotic transition.

Scatterer density, size, and refractive index contrast required for the Berry-Robnik transition in this specific cavity have not been calculated. DNA-directed nanoparticle self-assembly has precedent in biosensing on flat diamond. Extension to optical-precision placement in a dry curved cavity is an undemonstrated extrapolation — the conditions differ in substrate geometry, environmental chemistry, and positional precision requirements.

6.5 Electronic/Photonic Skin — Readout Layer

Flexible electronic/photonic skin — stretchable substrates with distributed photodetectors and local processing — has been demonstrated in research settings. The cuttlefish biological precedent is precise: distributed sensing, actuation, and signal routing in a single conformal layer on a curved surface. Biology solved the curved surface readout problem.

Conformal integration of a photodetector array on the outer surface of a CVD diamond hemisphere, co-designed with sub-wavelength pores for evanescent readout coupling, does not exist. The specific detector sensitivity, bandwidth, and noise floor required for meaningful cavity field sampling at 1550nm in this geometry has not been specified.

6.6 Rim Bonding — Explicit Risk Item

Optical quality rim bonding of curved diamond hemisphere to flat gold mirror (<λ/10 surface irregularity, mechanically and thermally stable, no birefringence-inducing stress) is not a known technology. This single step could prevent cavity assembly independently of all other fabrication progress. It is elevated to go/no-go dependency status — it must be addressed before any H-track fabrication investment is authorized.

7. The Formal Physics Framework

7.1 Billiard Classification

Three regimes relevant to Betsy. Integrable (sphere, hemisphere, paraboloid): Poisson level spacing statistics, strong degeneracy, low effective rank — where the unperturbed hemisphere sits. Mixed phase space (Limaçon, Robnik/quadrupole): Berry-Robnik statistics, coexistence of regular islands and chaotic sea, high modal diversity with retained addressability — the target regime. Fully chaotic (stadium, Sinai): Wigner-Dyson statistics, maximal entropy but rapid coherence degradation and lost addressability — too chaotic for a controlled reservoir. The Limaçon cavity r(θ) = R(1 + ε cos θ) is the experimentally realized mixed phase space geometry in chaotic microlaser research. Optimal operation at ε ≈ 0.35 established experimentally. The hemispherical cavity with scatterers approximates this regime stochastically rather than geometrically — controlled by ε_scatter rather than a geometric deformation parameter.

7.2 Modal Overlap Parameter M

M = Δf_mode / δf. For this cavity: δf ≈ 7 Hz (Weyl's law, V ≈ 4.1×10^-6 m³, f ≈ 1.93×10¹⁴ Hz). Δf_mode ≈ 1–20 MHz across plausible Q range of 10⁵–10⁷ after graphene absorption correction. M ≈ 10⁴–10⁶. Deep in the high-entropy regime across the full Q uncertainty range. Mode density is not the constraint. Note on graphene Q correction: graphene at partial Pauli blocking absorbs ~2.3% per layer per pass — roughly 23× the mirror loss contribution alone. With 400 cells covering ~32% of the inner surface, actual Q is likely 10⁵–10⁶, not 10⁷. The M conclusion holds across this range. The round-trip loss budget must be calculated explicitly (see loss budget equation in Section 16).

7.3 M vs rank(H_coupling) — The Critical Distinction

M measures mode density regime. rank(H_coupling) measures input coupling dimensionality. These are independent quantities. The T1 result (effective rank ≈ 12) measured rank(H_coupling), not mode density. A cavity with M = 10⁶ and rank(H_coupling) = 12 has vastly more modes than it can address. This distinction — previously conflated — is the primary diagnostic result of this program.

7.4 The Reverberation Chamber Analogy and RCM

Reverberation chambers implement the Random Coupling Model (RCM) experimentally: deterministic modal scaffold (rectangular geometry, predictable mode locations) plus controlled perturbation (mode stirrer) produces GOE random matrix statistics. The design principle — symmetric cavity plus controlled disorder — is directly applicable to Betsy. The frequency sweep is Betsy's mode stirrer; each frequency step is a new realization of the cavity Hamiltonian.

The RCM is a conceptual framework, not a validated theoretical foundation for this system. The RCM was developed for microwave-frequency metallic cavities (~1–18 GHz). Betsy operates at 193 THz — ~10⁴× higher in frequency — in a dielectric cavity with 2D graphene as the primary coupling element. Whether the Z_rad^1/2 · ξ · Z_rad^1/2 factorization remains valid in this regime has not been established. Additionally, whether the dual role of the honeycomb (simultaneously H_coupling and loss modulation layer) invalidates the Z_rad factorization has not been examined.

7.5 T2 Simulation Specification — Defining the Physics Kill

This section formalizes the T2 simulation campaign. Its purpose is to decide, on purely physical grounds, whether Betsy's cavity-plus-scatterer architecture can achieve sufficient effective dimensionality under realistic loss and noise. It defines the quantities to be computed, the conditions under which they are evaluated, and the numerical thresholds that constitute a physics kill.

7.5.0 Computational Feasibility — Mandatory Reading Before Proceeding

The full 25mm cavity at λ/20 resolution in 3D is completely intractable and must not be attempted as the T2 starting point. At λ = 1550nm, λ/20 ≈ 77.5nm. Grid points per axis: 25mm / 77.5nm ≈ 3.2 × 10⁵. Full 3D grid: (3.2 × 10⁵)³ ≈ 3.3 × 10¹⁶. Off by eight orders of magnitude from workstation-feasible.

Mandatory starting geometry: 2.5D axisymmetric MEEP simulation in cylindrical (r, z) coordinates. The hemispherical geometry has approximate rotational symmetry. An axisymmetric model captures the boundary curvature — the Sinai billiard mixing engine — while keeping grid count in the 10⁷ range, which is workstation-feasible on 16-32 cores. This is superior to a 2D planar patch with periodic boundary conditions (proposed by DeepSeek) because periodic BCs lose the hemispherical curvature and therefore the very mixing the architecture depends on. Identified by Gemini.

Tool: MEEP (MIT FDTD solver). MEEP handles dispersive materials (graphene surface conductivity, gold mirror), broadband pulses for the frequency sweep, and direct LDOS extraction via dipole source. MPB is appropriate as a supporting tool for eigenmode and band structure intuition but is wrong as the primary T2 tool — MPB computes eigenmodes only (frequency domain, static) and cannot perform time evolution or LDOS extraction. RCM analytical approach is the sanity check companion, not the primary validation. Tool distinction identified by GPT.

Runtime estimate (2.5D axisymmetric MEEP): 2D equivalent domain ~2000 × 2000 grid (4M cells), 50 frequency points via broadband Fourier transform (single run, not 50 separate simulations), single dipole source. On a 16-32 core workstation with 32-64GB RAM: 2–6 hours per run, 1–3 days for full parameter sweep. GPU acceleration can reduce this to under 1 hour per run.

7.5.1 Geometry and Operating Conditions

T2 is defined on the 2.5D axisymmetric model of the hemispherical cavity geometry — cylindrical (r, z) coordinates with rotational symmetry assumption — to capture boundary curvature while remaining workstation-feasible.

Flat SCD(111) diamond slab of thickness t (P3 target geometry)
High-reflectivity dielectric mirror on back surface
Input/output coupling via designated ports matching P3 design
Perfectly matched layer (PML) or equivalent absorbing boundaries
Center wavelength λ₀ = 1550nm, sweep Δf_sweep = 50 GHz
Discrete frequency steps {ω_k}_k=1^N_f with Δf_step = ω_k+1 − ω_k
Graphene modeled as surface conductivity at intended operating bias, coverage fraction matching honeycomb design
Scatterers as dielectric inclusions with specified refractive index contrast, size distribution, and spatial distribution targeting Berry-Robnik regime

Parallel injection alternative — fiber frequency comb architecture: The sequential sweep assumption above produces effective computation time ~N_f × τ_settle ≈ 5µs (50 steps × ~100ns cavity settling each). A fiber frequency comb with tooth spacing ≈ Δf_corr would inject all N_f channels simultaneously, replacing the sequential sweep with a parallel measurement. All comb teeth couple to the cavity at once; per-tooth responses are separated at the output by wavelength demultiplexing (e.g. arrayed waveguide grating). Effective measurement time drops from ~5µs to ~τ_settle ≈ 100ns — a 50× improvement in throughput. High-power few-cycle fiber frequency combs operating in the C-band (1530–1565nm) with repetition rates of 0.5–1 GHz — matching the estimated Δf_corr — have been demonstrated by Daniel Lesko’s group at FAU Erlangen-Nürnberg and LMU München. Their work on high-power few-cycle fiber frequency combs for attosecond and strong-field physics produces the exact comb parameters Betsy’s parallel injection architecture would require. The honest caveat: parallel comb injection requires that per-tooth cavity responses can be cleanly demultiplexed at the output — i.e., that the cavity does not mix frequency channels so severely that individual tooth responses cannot be recovered. Whether this condition holds in the Berry-Robnik regime is a T2 simulation question that should be run for both sequential and parallel injection architectures before P3 design is finalized.

7.5.2 Definition of the Coupling Matrix M(ω_k)

At each frequency ω_k, the coupling matrix M(ω_k) maps N_input independent excitation channels to N_readout readout observables:

M(ω k) \in ℝ N readout \times N input M ij (ω k) = steady-state intensity at readout i due to unit excitation at input j at frequency ω k The observable (intensity vs. field amplitude) should match the P3 readout modality (integrated intensity over detector pixel).

7.5.3 Spectral Stacking and Effective Rank

The stacked coupling matrix is constructed as:

M total = [M(ω 1); M(ω 2); ... M(ω N f)] \in ℝ (N f \cdotN readout) \times N input

Singular value decomposition:

M total = U Σ V T singular values: σ 1 \geq σ 2 \geq ... \geq σ min(N f \cdotN readout, N input)

Effective rank at cutoff ε:

r eff (N f; ε) = #{ℓ | σ ℓ \geq ε} ε is derived from the system noise floor: ε = σ noise /σ 1, where σ noise is the RMS noise contribution to the readout vector from detector shot noise, laser frequency drift, scatterer drift, and graphene gating drift (see Section 7.8). ε must be pre-registered before the simulation is run to prevent post-hoc rationalization of the cutoff.

T2 outputs: r_eff(N_f; ε) as a function of N_f, scatterer density ε_scatter, and graphene coverage and loss.

7.5.4 Spectral Correlation Bandwidth Δf_corr

For a fixed input excitation j, the output vector at frequency ω_k is y_j(ω_k) ∈ ℝ^N_readout. The normalized correlation between outputs at ω_k and ω_k':

C j (ω k, ω k') = [y j (ω k) \cdot y j (ω k')] / [‖y j (ω k)‖ \cdot ‖y j (ω k')‖]

Δf_corr is the frequency separation at which average correlation drops below C_th = 0.5:

⟨C(Δf)⟩ j \leq C th ⟹ Δf corr = Δf

T2 must report Δf_corr and compare it to Δf_step:

If Δf_step << Δf_corr: spectral stacking is highly correlated — sweep samples redundant states
If Δf_step ≳ Δf_corr: successive frequency steps are closer to independent — sweep provides genuine dimensionality

7.5.5 Numerical Kill Condition

For N_f = 50, realistic graphene loss, and scatterer configuration tuned for Berry-Robnik statistics: If r_eff(N_f=50; ε) ≤ 50: architecture does not achieve more than one useful independent channel per nominal node and is unlikely to outperform existing photonic reservoirs. Betsy stops here on physics grounds, independent of fabrication progress. If r_eff(N_f=50; ε) > 50 but Δf_step << Δf_corr: the rank is achievable in principle but the sweep strategy must be redesigned. Not a kill — a redesign trigger.

7.5.6 Parameter Sensitivity and Reporting

T2 must produce a parameter sweep, not a single number. For each parameter set, report:

r_eff(N_f; ε) — effective rank at pre-registered noise floor
Δf_corr — spectral correlation bandwidth
Memory capacity — the timescale over which the cavity retains input information. Reservoir computing requires balance between separability (high rank) and generalization (fading memory). A cavity that is too chaotic loses input signal memory before the readout layer can extract it. Maximising rank without bounding memory capacity is an incomplete optimisation. (Identified by Gemini)
Representative eigenmode patterns and example M(ω_k) matrices for inspection

This provides a local robustness map: whether the architecture's dimensionality and memory are jointly adequate across a plausible fabrication window.

Parameter	Sweep range	Rationale
Graphene coverage fraction	25%–45% of inner surface	Fabrication variation around 32% nominal
Graphene sheet resistance	200–800 Ω/□	Bias point and mobility variation
Scatterer density ε_scatter	0 to Berry-Robnik transition	Map rank vs. disorder strength
Mirror reflectivity	99.5%–99.95%	Manufacturing variation
N_f	10, 20, 30, 50	Map rank scaling law

7.6 Rank Scaling Hypothesis

Before running T2, the expected scaling law must be stated. Writing it down before running the simulation is the correct scientific posture — it prevents post-hoc selection of the result that confirms the hypothesis.

Expected scaling in the Berry-Robnik regime: r_eff(N_f; ε) is expected to scale approximately linearly with N_f up to a saturation point, with slope s determined by the fraction of frequency steps that produce genuinely decorrelated coupling vectors. Formally: r_eff(N_f; ε) ≈ r_spatial + s · N_f for N_f ≤ N_sat, where r_spatial ≈ 12 (from T1 surrogate), s is the per-step rank increment (expected 8–15 in the Berry-Robnik regime), and N_sat is the saturation point where adding more frequency steps yields diminishing returns. Expected saturation mechanism: Saturation occurs when the frequency steps begin sampling the same underlying mode families despite decorrelated phases — i.e., when N_f · Δf_step exceeds the total number of distinguishable mode families addressable by the honeycomb geometry. The expected saturation rank is bounded by the number of honeycomb cells (~400) and the surface coverage fraction (~32%), suggesting N_sat · s + r_spatial ≲ 130–150 before coverage-limited saturation. What failure of linear scaling means: If r_eff saturates below 50 at N_f = 10, the cavity modes are more correlated than the Berry-Robnik estimate predicts — the scatterers have not successfully pushed the cavity into the mixed phase space regime. The required response is either denser scatterers (higher ε_scatter) or a different scatterer size/contrast to achieve stronger mode coupling. What faster-than-linear scaling would mean: If r_eff grows faster than linearly with N_f, the frequency sweep is sampling mode families that are not just decorrelated but anti-correlated — a stronger-than-expected mixing regime. This would be a positive result but would require physical explanation.

7.5.7 Readout Sparsity — Calculate Before Running T2

Before the T2 simulation is run, a simpler calculation must be completed: given N sensor positions on the electronic skin layer, how much of the projected r_eff = 600 is actually accessible?

If the skin sensors sit at nodal points of the primary cavity modes, effective rank collapses regardless of the cavity's internal physics — regardless of scatterer density, spectral multiplexing, or mode mixing. This is cheap to compute analytically from the mode structure and directly informs sensor placement in the P3 experimental design. It is the cheapest possible rank check and should precede the T2 simulation. Identified by Gemini as the first thing to test.

7.7 Null Hypothesis

The simplest competing explanation for any positive result from Betsy must be stated before the experiment is run. The null hypothesis is: The cavity is simply a lossy, hard-to-control optical enclosure whose apparent high-dimensional response reflects measurement noise, detector cross-talk, and laser drift rather than genuine modal diversity. The apparent richness of the readout vector does not translate into usable rank for reservoir computation — it is noise dressed as signal. The null hypothesis predicts: (1) r_eff(N_f; ε) does not increase meaningfully with N_f when ε is set by the measured noise floor; (2) training a linear decoder on the readout vector produces SER indistinguishable from a random linear projection of equivalent dimension; (3) removing the scatterers does not materially change the benchmark performance. The null hypothesis is falsified if: r_eff(N_f; ε) scales with N_f at the pre-registered noise floor, AND SER improves beyond the ring-resonator baseline, AND removing scatterers degrades performance. All three conditions should be tested in the P3 experiment, not just the primary SER comparison.

7.8 Noise and Drift Analysis

Each noise source maps to a specific degradation of r_eff or SER. This section provides the framework for deriving ε from the noise floor — a prerequisite for pre-registering the T2 kill condition.

Noise source	Magnitude (order of magnitude)	Effect on architecture	Status
Thermal noise in diamond (thermo-optic)	dn/dT ≈ 10^-5 K^-1; ΔT over 170ps to be calculated	Phase fluctuation in cavity modes; reduces coherence of collapse	Favorable indication, not calculated
Detector shot noise	SNR ~ √(N_photons) per readout pixel; depends on skin layer coupling efficiency	Sets noise floor σ_noise for ε pre-registration	Not specified; requires P3 detector design
Laser frequency drift (sequential sweep)	Stabilized erbium-doped fiber laser with AI-assisted frequency stabilization: <1 MHz/s drift achievable with current commercial systems	If drift > Δf_corr/N_f during sweep, successive steps are not at registered frequencies; corrupts M_total	Manageable with stabilized laser. Comb alternative: fiber frequency comb with tooth spacing ≈ Δf_corr replaces frequency drift concern with comb stability — teeth drift together (common-mode noise), which is more tractable for ε pre-registration than independent step drift.
Scatterer drift	Unknown; depends on scatterer material and bonding chemistry	If scatterers reposition over measurement timescales, M(ω_k) is non-stationary; corrupts training	Absent — must be characterized for P3 scatterer design
Graphene gating drift	Ferroelectric retention: years at room temperature for well-processed Al:HfO₂	Slow drift of honeycomb cell states corrupts the assumed configuration during inference	Characterized in companion program for flat geometry; curved surface unknown
Cavity initialization residuals	Depends on reset mechanism — not yet specified	Residual modes from previous computation corrupt the next one (temporal aliasing)	Absent — no reset mechanism proposed

The ε pre-registration procedure: before running T2, measure or estimate σ_noise for the P3 detector configuration, set ε = σ_noise/σ₁ (where σ₁ is the largest singular value from a test measurement), and register this value before examining the full singular value spectrum. This prevents selection bias in the kill condition.

8. The Computational Primitive: Resonate, Modulate, Collapse

Resonate: The cavity seeks stable mode configurations through constructive and destructive interference. Every supported mode is excited simultaneously from the first round trip.

Modulate: The graphene honeycomb imposes boundary conditions on which modes survive — applied continuously and simultaneously with resonance.

Collapse: The cavity evolves toward the lowest-loss mode configuration consistent with the modulation pattern. The honeycomb defines an energy landscape; the light finds the minimum by existing within it.

The "physics as optimizer" framing is precise for a single-shot cavity evolution. With a frequency sweep required for useful dimensionality, the computation is a time-series of cavity measurements from which a trained decoder extracts the answer. The primitive accurately describes what happens at each frequency step. The computation as a whole is a sequential measurement process. Both descriptions are true and neither invalidates the other.

9. The Architectural Identity: Spatiotemporal Reservoir with Engineered Disorder

Reservoir: Fixed high-dimensional dynamical system. The cavity transforms inputs; a decoder extracts outputs. Only the decoder is trained.

Spatiotemporal: Dimensionality from space (400 cells simultaneously) and time (N_f frequency steps sequentially). Measurement vector: {I_j(ω₁), I_j(ω₂), ...}.

Engineered disorder: Scatterer density ε_scatter tuned to maximize r_eff(N_f; ε) subject to M >> 1. Designed disorder targeting the Berry-Robnik transition.

9.1 The Decoder — Ranked by Readiness

Option A — External electronic linear regression (adopt for all near-term validation): Skin layer outputs 400 × N_f intensity values to a linear regression layer. Standard architecture, demonstrated in multiple photonic reservoir systems. Works today in principle.

Option B — Downstream hemisphere as optical decoder: Next hemisphere's mode competition implements weighted readout optically. Requires inter-hemisphere coupling solution — an open design gap. Not available until collective architecture is demonstrated.

Option C — Ferroelectric memory as implicit trained weights: Slow learning consolidates domain configurations as implicit readout weights. Physical mechanism unspecified. Most biologically faithful, least characterized, not available near-term. If unphysical, the strongest MMF differentiation argument collapses.

10. The Benchmark Task — Nonlinear Channel Equalization

Kill shot from v1.6 red team: no specified task. Closed in v1.7. The task, performance claim, and P3 validation experiment are specified here.

10.1 Why Nonlinear Channel Equalization

Nonlinear channel equalization is a standard photonic reservoir computing benchmark with published results from ring-resonator and delay-line systems — direct comparison is possible. A signal transmitted through a nonlinear channel is recovered by the reservoir. Performance is measured by symbol error rate (SER). Critically for Betsy: performance scales monotonically with r_eff — higher rank directly produces lower SER. This makes the task performance analytically traceable to the quantity T2 is maximizing.

10.2 The Performance Claim

A Betsy-class planar diamond cavity reservoir operating in the Berry-Robnik regime via engineered disorder and spectrally multiplexed via 50 GHz laser sweep achieves lower SER on the standard nonlinear channel equalization benchmark than a ring-resonator photonic reservoir of equivalent node count (400 nodes), because three-dimensional mode volume and scatterer-induced mode mixing produce a flatter H_coupling singular value spectrum — more independent readout channels per physical node. P3 decision threshold: SER improvement >10% over ring-resonator baseline, AND r_eff(N_f=50; ε) > 50 from T2 simulation. Both conditions must hold. Falsification: If P3 SER ≥ ring-resonator baseline, the performance advantage does not exist. Architecture requires revision or abandonment of competitive reservoir computing claim.

10.3 The Ghent / Bogaerts Collaboration

Wim Bogaerts' group at Ghent University / imec — programmable photonic circuits, photonic reservoir computing, imec fabrication access. Appropriate collaboration partner for P3 benchmark experiment. Required before approach: specified task (provided), performance claim (provided), quantitative SER prediction from T2 simulation (pending). Approach after T2 simulation is complete. See Appendix A for outreach template.

11. Comparison to Multimode Fiber Reservoirs

Multimode fiber (MMF) reservoirs provide high-dimensional spatiotemporal photonic reservoirs: thousands of independent spatial modes, room temperature, commercially available, no fabrication challenges, published benchmark results including nonlinear channel equalization. They use spectral multiplexing and spatial diversity structurally similar to Betsy's proposal. They are mature, cheap, and work today.

11.1 What Betsy Provides That MMF Does Not — Ranked by Argument Strength

Strongest — Ferroelectric material-level learning: MMFs are static reservoirs. Betsy's ferroelectric-gated graphene cells consolidate configurations producing clean collapses, implementing slow structural learning. A Betsy reservoir improves on its task through use. An MMF does not. This rests on Option C decoder being physically realizable — the least characterized component. If Option C is unphysical, this distinction collapses. Moderate — Programmable disorder: Betsy’s scatterer distribution and honeycomb configuration are both tunable. Different tasks served by different disorder configurations. MMF disorder is fixed at manufacturing. Weakest as currently argued — Three-dimensional mode volume: Whether 3D mode volume provides higher effective dimensionality per unit volume than MMF mode count for a reservoir task has not been argued quantitatively. This requires the T2 simulation and P3 experiment to establish.

A direct quantitative comparison of Betsy's projected r_eff(N_f=50; ε) against a commercially available MMF mode count for the nonlinear channel equalization benchmark has not been performed. This comparison should be run simultaneously with P3.

12. The Robotics Argument

Illustrative of a problem class, not a specified implementation. Information flow from sensors to cavity to motor commands has not been specified. Latency budget for the full encoder-cavity-decoder-actuator chain has not been calculated. The benchmark task (Section 10) is the validation path. The robotics argument motivates; it does not validate.

13. The Collective Architecture

A single hemisphere has fixed computational dimensionality. The collective — cascaded hemispheres each specializing through ferroelectric learning — is the scaling answer. Inter-hemisphere coupling remains an open design gap with no proposed solution. The collective architecture is downstream of demonstrating a single hemisphere works.

14. Intrinsic Learning

Fast (170ps): Cavity evolves toward lowest-loss configuration on every round trip.

Medium (seconds–minutes): Configurations producing stable collapses preferentially reused. Analogous to working memory.

Slow (hours–extended operation): Ferroelectric domain consolidation. Analogous to synaptic consolidation.

The slow timescale physical mechanism is unspecified. The candidate Hebbian write protocol is plausible but ferroelectric domain switching dynamics under optical-power-correlated field pulses have not been characterized. This is the mechanism underpinning the strongest MMF differentiation argument.

15. The Calcium Problem — Graphene, Ferroelectrics, and the Synapse

Al:HfO₂ ferroelectric gating of graphene provides non-volatile optical modulation with write endurance 10⁸–10¹² cycles, retention years at room temperature, and imprint effects over extended operation. Route F / Route J from the companion grain boundary paper is the Pareto-front integration path. The companion program's experimental programme generates directly applicable characterization data.

Goldilocks window thresholds require recalculation for Betsy's hemispherical geometry. Curved surface Raman strain-doping decomposition protocol does not exist. No graphene quality measurements on any curved diamond substrate have been reported in the literature.

16. Mathematical Skeleton

16.1 Mode Competition

Round-trip loss perturbation: δL_q,i = α_i · |U_q(r_i)|² · A_cell. Total loss: L_q = L₀ + Σ_i δL_q,i. Field decay: |E_q(N)| = |E_q(0)| · exp(−L_q · N/2). The cavity evolves toward minimum-loss configuration — gradient descent on the loss functional, implemented physically.

16.2 Round-Trip Loss Budget

The following loss budget equation defines the terms to be calculated. No numerical values have been computed for this geometry.

L total = L mirror + L graphene + L scatterer + L diamond L mirror = 1 - R mirror \approx 10 -3 (from R > 99.9%) L graphene = α graphene \cdot f coverage \approx 0.023 \cdot 0.32 \approx 7.4\times10 -3 per pass (estimate) L scatterer = σ scatter \cdot ρ scatter \cdot L path (to be calculated from scatterer specification) L diamond = α diamond \cdot L path (typically <10 -4 for optical grade CVD diamond) Q \approx 2πf / (c \cdot L total / L path) Note: graphene loss dominates over mirror loss by approximately 7\times at current estimated coverage. The Q pre-registration for ε must derive from this budget, not from mirror reflectivity alone.

16.3 Spectral Stacking and Surface Coverage

M_total = [M(ω₁); M(ω₂); ...]. r_eff,total = rank(M_total). Whether rank scales linearly with N_f or saturates at the coverage-limited ceiling (~130–150 for 32% coverage) is the decisive simulation result.

400 cells at 1mm diameter cover ~314 mm² of ~980 mm² inner hemisphere surface — 32% coverage. The remaining 68% contributes no controlled modulation. This limits r_eff independently of scatterer density. Coverage fraction analysis must be included in T2 to determine whether smaller cells with higher coverage materially improve rank.

17. What Betsy Cannot Do

Perform discrete logic. Continuous analog computation only. Not a replacement for digital processors.

Compute with useful dimensionality from spatial sampling alone. At 1mm cells / 1550nm, r_eff ≈ 12 spatially. Spectral sweep and scatterers are required.

Compute faster than electronic systems in the sweep regime. If N_f = 50 steps with ~100ns cavity settling each, effective computation time ~5µs — comparable to, not faster than, electronic systems.

Self-configure without a decoder. Option A electronic decoder required for near-term validation. Option C remains uncharacterized.

Outperform MMF reservoirs without ferroelectric learning. If Option C proves unphysical, primary MMF differentiation collapses.

Operate without an encoding mechanism. Partially addressed for equalization benchmark. General sensor encoding unsolved.

Scale to a collective without solving inter-hemisphere coupling. No proposed solution exists.

Operate without precise initialization. No reset mechanism demonstrated. Temporal aliasing is the failure mode.

Be evaluated against conventional benchmarks. Nonlinear channel equalization is the proposed starting point. New benchmarks may be needed for other applications.

18. Fabrication Reality

The materials are not speculative. The fabrication is. The rim bonding step is elevated to go/no-go dependency status — it must be addressed before any H-track investment.

18.1 Fabrication Step Assessment

Step	Difficulty	Status	Route / Note
Optical grade CVD diamond hemisphere	High	Extrapolated	Element Six specification upgrade — no capability gap, specification gap only
Curved inner surface CMP (<1nm RMS)	Very High	Absent	No precedent for diamond hemispheres
In-place graphitization on curved diamond (Route F)	High	Extrapolated	Physically motivated; undemonstrated on curved surface
Honeycomb patterning on curved graphene	Very High	Absent	No precedent
Ferroelectric gate per cell on curved surface	Extremely High	Absent	No precedent on curved graphene
Deliberate scatterer placement (engineered disorder)	High	Extrapolated	DNA self-assembly — biosensing precedent on flat diamond; optical cavity extension undemonstrated
Per-cell electrical addressing on curved surface	Extremely High	Absent	No precedent
Sub-wavelength porosity aligned to curved honeycomb	Very High	Extrapolated	Femtosecond fiber laser ablation — CVD diamond absorbs at ~800nm ultrafast pulses; material evaporates before thermal diffusion, preventing graphitization and cracking. Demonstrated on flat diamond; curved surface extension needed. Modular high-power femtosecond fiber laser systems now accessible to research groups without dedicated ultrafast infrastructure.
Gold base with dielectric mirror	Moderate	Established	Standard optical coating on flat substrate
Optical quality rim bonding — GO/NO-GO DEPENDENCY	Very High	Absent	Potential assembly showstopper. Must be addressed before H-track authorization.
Graphene-diamond thermal expansion mismatch	High	Absent — NEW	Different thermal expansion coefficients. Cooling from graphitization temperature (~900°C) could cause delamination or buckling of the graphene layer, creating uncontrolled scattering centers that corrupt the engineered disorder model. Must be characterised before Route F on curved surface is attempted. (Identified by Gemini)
Gold-diamond thermal cycling micro-fissures	High	Absent — NEW	Gold has high expansion coefficient relative to diamond. At 25-30mm scale, thermal cycling may cause micro-fissures in TiO₂/SiO₂ dielectric mirror coating on the gold base, creating dark spots that degrade r_eff over time. Must be evaluated before cavity assembly. (Identified by Gemini)
Electronic/photonic skin on outer surface	High	Extrapolated	Flexible arrays demonstrated; co-integration with porous curved diamond absent

18.2 Fabrication Readiness Matrix

Track	Steps	Readiness	Blocking dependency	Decision gate
P0 — Microwave scale model	250mm cavity (10× scale), polished aluminum or acrylic, microwave frequencies. Helmholtz equation scales linearly with wavelength.	Very High — buildable immediately for ~hundreds of dollars	None	Sinai billiard mixing confirmed and spectral multiplexing rank scaling validated before any material investment. Identified by Gemini as the first thing to build.
P1 — Flat graphene	Route I graphene on flat SCD(111), honeycomb patterning	High — commercial supply chain available today	P0 validates mixing physics	Graphene quality meets Goldilocks window
P2 — Flat ferroelectric + scatterers	Ferroelectric gate per cell, DNA scatterer placement, flat diamond	Medium — established components, untested integration	P1 success	Non-volatile modulation per cell demonstrated; scatterers placed at target density
P3 — Planar reservoir benchmark	Planar cavity with honeycomb, scatterers, 50 GHz sweep, electronic readout	Medium — requires T2 simulation to pre-register ε and SER threshold	P2 success + T2 simulation complete	r_eff > 50 from T2 AND SER improvement >10% vs ring-resonator baseline
P4 — Shallow spherical cap	Curved geometry transition test	Low — no curved diamond graphene precedent	P3 success + in-place graphitization on curved surface demonstrated	Route F quality maintained on cap geometry
H-track — Full hemisphere	H1–H8: hemisphere, graphitization, patterning, gating, scatterers, skin, assembly, characterization, demonstration	Very Low — multiple absent capabilities	P4 success + rim bonding solution + Element Six optical grade supply	All P and prerequisite H steps succeed in sequence

18.3 Critical Path

The single highest-risk step is optical quality rim bonding. It is a go/no-go dependency for the full Betsy track — even if every other fabrication step succeeds, an unsolved rim bonding problem prevents cavity assembly. It should be investigated in parallel with P-track, not sequentially after it. A dedicated materials investigation — adhesive bonding, fusion bonding, optically contacted flat on polished diamond rim — should be commissioned as an independent workstream before H-track investment is authorized.

18.4 P3 Cost/Risk Gate

Before initiating P3, the following quantitative thresholds must be specified and agreed:

Parameter	Threshold	If exceeded
P3 device count	≤ 10 devices before decision point	Stop and re-evaluate design before further fabrication
P3 timeline	≤ 18 months from P2 completion to P3 decision	Reassess collaboration structure and resource allocation
SER improvement threshold	> 10% improvement over ring-resonator baseline	If <10%: do not proceed to H-track
r_eff threshold (from T2)	> 50 at N_f = 50	If ≤ 50: do not proceed to P3 without architectural revision

19. Supply Chain and Collaboration Landscape

Element Six (De Beers, UK): Primary diamond supplier. CVD diamond hemispheres at volume (speaker domes). Optical grade specification upgrade is the first supply chain conversation.
Fraunhofer IAF (Germany): Highest-purity SCD growth for quantum/photonic applications. Primary physics validation partner.
Diamond Foundry (Silicon Valley): First 4-inch SCD wafer. Industrial scale partner for future volume.
Graphenea (Spain): Primary graphene supplier for P1–P3. Route I supply chain. ~$50–200 per transfer.
2D Semiconductors (US): hBN encapsulation for Route I.

Ghent University / Bogaerts group (Belgium): Programmable photonics, photonic reservoir computing, imec fabrication access. Appropriate P3 collaboration partner. Approach after T2 simulation is complete with: specified task (provided), performance claim (provided), quantitative SER prediction from T2 (pending). See Appendix A for outreach template.

Daniel Lesko group — FAU Erlangen-Nürnberg / LMU München (Germany): High-power few-cycle fiber frequency combs for attosecond and strong-field physics. Potential collaboration contact for the parallel injection comb architecture described in Section 7.5.1. Their comb specifications — C-band operation, repetition rates in the 0.5–1 GHz range, high power per tooth — match the parallel injection requirements directly. Contact after T2 simulation determines whether parallel comb injection is architecturally viable.

20. Graphene Integration — Companion Research Program

The companion paper (Graphene Integration Route Selection for Ferroelectric Photonic Memory, March 2026) identifies Route F (SCD(111) Ni-assisted + H-termination, 21.8% joint pass probability today, 57.2% at H-termination ceiling) and Route J (CVD/Pt → bonded SCD(111), 28.8%) as the Pareto front. Route F/I crossover at 3,500–4,500 cm²/V·s mobility mean. Companion program's four-experiment programme generates directly applicable characterization data.

Goldilocks window thresholds require recalculation for Betsy's curved geometry. Curved surface Raman decomposition protocol does not exist. No graphene quality measurements on curved diamond have been reported. These extensions are required before Route F can be targeted for hemispherical fabrication.

21. Open Questions Register

Priority	Question	Type	Status
T1 ✓	Spatial rank(H_coupling) — adequate?	Simulation	ANSWERED: ~12, catastrophic. Spatial alone insufficient.
T1 ✓	M >> 1 at 1550nm?	Calculation	ANSWERED: M ≈ 10⁴–10⁶ after graphene Q correction.
T1 ✓	No benchmark task specified.	Architecture	CLOSED: Nonlinear channel equalization. P3 experiment defined.
T1	Does r_eff(N_f=50; ε) > 50? Run definitive cavity eigenmode simulation plus stacked M(k) SVD per Section 7.5 specification.	Simulation	Not done. Most important remaining calculation. Pre-register ε from noise model before running.
T1	Is Δf_step ≳ Δf_corr? Does the spectral sweep actually sample decorrelated states?	Simulation	Not done. Required alongside rank calculation. Defined formally in Section 7.5.4.
T1	What scatterer density ε_scatter places this cavity at the Berry-Robnik transition? Full round-trip loss budget including graphene absorption?	Analytical + sim	Not done. Required for P3 design specification and ε pre-registration.
T1	Does thermal noise remain below graphene modulation contrast over 170ps?	Analytical	Not done. Favorable indication. Must be calculated and incorporated into noise budget.
T2	Does P3 planar experiment validate SER claim vs ring-resonator baseline at >10% improvement?	Experiment	Not done. Architectural decision point. Both T2 simulation and P2 fabrication must precede.
T2	Can Element Six supply optical grade CVD diamond hemispheres? What specification upgrade cost and timeline?	Procurement	Not asked. First supply chain conversation. See Appendix A.
T2	Does in-place Route F graphitization quality hold on curved SCD surface? How do five Goldilocks criteria change with curvature?	Materials	Not done. Required before H-track investment.
T2	What is the rim bonding solution? Is it a showstopper?	Materials	Open. Go/no-go dependency. Investigate in parallel with P-track, not after it.
T2	Inter-hemisphere coupling mechanism?	Design	Open. No proposed solution. Downstream of single hemisphere demonstration.
T2	Which decoder option is adopted? When does Option C become testable?	Design decision	Option A for near-term. Option C pending physical mechanism specification.
T3	Does Option C ferroelectric learning work physically? What is the domain switching mechanism under optical-power-correlated field pulses?	Theory + materials	Unspecified. Required for strongest MMF differentiation.
T3	Does 32% surface coverage limit r_eff independently of scatterers? Coverage vs. rank tradeoff?	Simulation	Not analyzed. Include in T2 parameter sweep.
T3	Curved surface Goldilocks window recalculation?	Materials + calc	Absent. Companion paper Section 6.6 flags this explicitly.
T3	Cavity initialization and reset mechanism? (See Section 12.1 placeholder)	Design	Open. Named subsection added. No proposal exists.
T3	P3 success threshold beyond SER >10%? What additional metrics justify H-track?	Strategy	Partially defined. See Section 18.4 gate. Refine before P3 is run.

12.1 Cavity Initialization and Reset — Placeholder

The cavity must be initialized to a known field state before each computation. Residual modes from a previous computation corrupt the next — temporal aliasing. No reset mechanism has been proposed for this cavity geometry. This is a practical systems requirement that deserves its own development track, not treatment as an incidental detail. Candidate approaches (not evaluated): cavity dump via switched loss element, frequency-detuned erasure pulse, physical shutter. Each has tradeoffs against cavity Q and speed. This placeholder should become a named subsection with a candidate reset protocol before P3 design is finalized.

22. Relationship to the Primary Research Program

Betsy is kept deliberately separate from the primary photonic research program: diamond-integrated 3D silicon photonics, non-volatile photonic memory architectures, ferroelectric thin film pre-qualification, and the MISA programming framework. The primary program is technically rigorous, grounded in demonstrated performance, oriented toward near-term manufacturability. Betsy is speculative, without fabrication precedent, without demonstrated computational mechanism. Conflating them compromises both.

The grain boundary paper is the upstream prerequisite for Betsy's graphene fabrication track — not a parallel program. Route F process maturity on flat SCD(111) must reach the mobility crossover before curved surface development is justified.

The material stack is shared. The fabrication paradigm, the computational architecture, and the speculative distance from demonstrated reality are not.

23.1 CoNexus and Betsy — Strategic Distinction

CoNexus is the long-horizon volumetric computing vision: a macro-scale, paradigm-level synthesis in which distributed optical processing elements form a collective computational substrate at scale. It is not a paper. It is not a device. It is the conceptual framework within which Betsy exists as the first experimentally testable component.

Betsy is the experimental wedge. It is a single-device, falsifiable instantiation of one principle from the CoNexus framework: that a plano-hemispherical optical cavity with engineered disorder can function as a high-dimensional photonic reservoir. Betsy can be published, validated, or killed without requiring the CoNexus thesis to be defended. That separation is deliberate and valuable.

Betsy results — positive or negative — inform CoNexus without depending on it. Either outcome advances the program. This is what it means for Betsy to be the experimental wedge into reality.

24. Alternative Application — Quantum Decoherence Simulation

This section documents a secondary application identified after the primary reservoir computing framework was established. It is a candidate use case, not a validated claim. The primary falsification path — nonlinear channel equalization at P3 — is unchanged. This application is downstream of P3, not a substitute for it. Reviewed by DeepSeek, Perplexity, GPT, and Gemini before incorporation.

24.1 The Problem: Simulating Open Quantum Systems

Calculating quantum decoherence requires solving the dynamics of a quantum system coupled to a large environment. State space grows exponentially with system size. Three standard approaches each fail for important problem classes: Master equations (Lindblad formalism) assume Markovian environments and lose detailed bath information. Tensor network methods (MPS, PEPS) struggle with high entanglement entropy and 3D geometries. Path integral Monte Carlo suffers from the sign problem for fermions and real-time dynamics. For physically relevant problems — NV centers in diamond coupled to phonons, quantum dots in optical cavities, molecular emitters near plasmonic structures — these methods converge slowly or fail entirely.

24.2 The Formal Mapping — With Corrected Epistemic Labels

Betsy solves Maxwell's equations in a high-finesse optical cavity. The LDOS: ρ(r,ω) ∝ Im[G(r,r,ω)]. The spectral density: J(ω) = Σ_k |g_k|² δ(ω − ω_k). T1 ∝ J(ω₀). T2 ∝ low-frequency components of J(ω).

The correspondences below require two distinct approximations, both of which must hold: (1) Weak excitation — emitter stays mostly in ground state, enabling linearisation. (2) Weak coupling — g << κ. These are not the same assumption. Weak excitation does not imply weak coupling.

Quantum quantity	Betsy analog	Epistemic status
LDOS ρ(r,ω)	Steady-state intensity at readout I(r,ω)	Extrapolated — valid as proxy only under linear response with calibrated source/receiver model. Not generally interchangeable with LDOS. (Corrected by Perplexity)
Coupling strength \|g_k\|²	Mode overlap with graphene cell at position r_i	Established — standard cavity QED result under weak coupling
Spectral density J(ω)	Frequency comb output at a given readout cell across the sweep	Heuristic — frequency sweep over steady-state driven frequencies is not the same as a time-domain bath correlation function. Explicit transform relation required. (Corrected by Perplexity)
Bath correlation function	Fourier transform of M(ω_k) over frequency sweep	Heuristic — valid only if cavity is in wave-chaotic regime AND steady-state to time-domain transform is explicitly derived
Non-Markovian dynamics	Captured if Δf_step ≈ Δf_corr	Extrapolated — requires T2 simulation to verify spectral resolution
T1 (Purcell-enhanced relaxation)	Integrated intensity at emitter position over resonance bandwidth	Extrapolated — Purcell formula applies in weak coupling limit; breaks in strong coupling

24.3 What Betsy Can Simulate — Corrected Claim

Betsy can emulate engineered spectral densities, not generic decoherence environments. The cavity is finite, high-Q, and produces discrete or quasi-discrete modes with long correlation times. True decoherence models assume a continuum bath. The correct claim: Betsy can model specific, engineered spectral density profiles J(ω) for targeted decoherence problems where the bath spectral structure is itself the quantity of interest. The valid regime requires modal density sufficiently dense that Δω_modes << γ_system. Below this density, Betsy models coherent back-action, not decoherence. Framing corrected by GPT.

24.4 Objective Function Incompatibility

Reservoir computing and decoherence simulation are not just different optimisation targets — they are orthogonal optimisation targets. This is stronger than a design tension; it is an objective function incompatibility. Framing introduced by GPT.

Reservoir computing wants: maximal state space mixing, high rank, chaotic modal structure, computational separability across input dimensions.

Decoherence simulation requires: physically interpretable modal structure, calibrated dissipation channels, faithful Green's function reproduction, controlled non-Markovian structure comparable to analytic benchmarks.

The overlap region where both objectives are simultaneously satisfied may be narrow. The T2 parameter sweep must explicitly map this overlap — it cannot be assumed to exist.

24.5 Failure Modes — Added from Four-AI Review

Finite size recurrence effects (Gemini): In a high-Q closed cavity, energy reflects and returns. True decoherence simulation requires simulation time shorter than the Poincaré recurrence time — the time for a photon to round-trip the cavity and re-interact with the emitter. For a 25mm cavity: recurrence time ≈ 170ps. This is far shorter than most decoherence timescales of interest and may severely limit applicability without active cavity dumping between measurement windows.

Mode sparsity producing artificial coherence plateaus (GPT): If Δω_modes > γ_system, the spectral density J(ω) is a series of Lorentzians, not a smooth continuum. Integration over a sparse spectrum produces artificial coherence plateaus that do not correspond to any physical bath.

Boundary condition artifacts (GPT): The plano-hemispherical geometry introduces caustics, mode clustering near the axis, and polarisation-dependent reflection at the rim. These distort the spectral density in geometry-specific ways.

Thermal bath absence (GPT): Betsy is effectively at T ≈ 0 unless the input field statistics are explicitly engineered to reproduce thermal occupation numbers. Real quantum decoherence problems involve J(ω, T) where temperature modifies the spectral density through the Bose-Einstein occupation factor.

Material nonidealities as dominant uncertainty (Perplexity): If graphene loss, disorder, or fabrication tolerances dominate the cavity response, the limiting factor is material noise rather than geometric rank. This weakens the sequencing argument — the two applications may fail for independent reasons.

24.6 Scope

Betsy's decoherence simulation use case applies to: NV centers in diamond coupled to phonons or photons, quantum dots in optical cavities, molecular emitters near plasmonic structures. All require verification that weak coupling (g << κ) holds.

Betsy cannot simulate: entanglement between multiple qubits, non-classical states of light, strongly driven quantum systems, strong coupling cavity QED, superconducting qubits in circuit QED regimes (model-specific — optical cavity language does not safely transfer, removed from coverage list per Perplexity), thermal bath statistics without explicit input field engineering.

24.7 Falsification Criteria

Question	Kill condition	Experiment
Does dense-mode condition hold?	Δω_modes > γ_system for target quantum system	Analytic check from T2 simulation mode density output — must precede any decoherence experiment
Is recurrence time longer than target decoherence timescale?	Poincaré recurrence time (~170ps) < target T1 or T2	Calculate before any experiment
Does cavity LDOS match analytic prediction?	Purcell factor deviation exceeds improvement claimed over digital methods	P3 variant: test emitter, measure lifetime vs. analytic Purcell formula
Does classical approximation hold?	Target system in strong coupling or nonlinear regime	Analytic check: compute g/κ for target system before experiment

24.8 Sequencing

Decoherence simulation is downstream of P3, not parallel to it. The finite-size recurrence problem may render Betsy unsuitable for decoherence simulation regardless of reservoir computing performance — the two applications may fail for independent reasons. This is a weaker sequencing argument than stated in earlier drafts, and is stated here as corrected.

25. Application Landscape and Mission Profile

Betsy is not solving a pressing problem that lacks alternatives. It is solving a different class of problem — one that currently lacks good low-power, real-time, analog solutions in specific environments. The applications below are where that class appears. (DeepSeek analysis)

25.1 Application Tier Structure

1. Embedded control for small, power-constrained platforms — drones / UAVs / robotic insects

The constraint: 5µs computation time (sequential sweep) or ~100ns (parallel comb injection, Section 7.5.1) is slow for a desktop GPU but fast for a 50g drone with a 500mAh battery running obstacle avoidance.

Approach	Power	Latency	Works on 50g drone?
Edge GPU (Jetson Nano)	5-10W	~10ms	No — too heavy, too hot
Microcontroller ML	0.1-0.5W	~50ms	Marginal — slow
Betsy (sequential sweep)	~0.5-1W	~5µs	Yes — if it works
Betsy (parallel comb)	~0.5-1W	~100ns	Yes — competitive with any alternative

The advantage: no backpropagation, no weight storage, no digital overhead. The cavity is the computation. What Betsy would NOT enable: navigation, mapping, object recognition, or anything requiring precision >8 bits.

2. High-vibration, high-radiation environments — rockets / launch vehicles / small satellites

Digital electronics fail under vibration during launch and accumulate radiation damage in orbit. FPGAs and GPUs require shielding, redundancy, and cooling. CVD diamond is structurally robust. Photons don't accumulate radiation damage. Diamond has high thermal conductivity.

Radiation claim — corrected: The passive diamond cavity is radiation-robust. Photodiodes are moderately tolerant. The laser and control electronics are not inherently radiation-hardened. The correct statement: Betsy localises the radiation risk to the laser and readout electronics, not to the computational cavity itself. This is more useful than "no radiation hardening required" — it tells engineers exactly what needs qualification work. (GPT correction)

3. Real-time nonlinear channel equalization — free-space optical communications

This is the specified P3 benchmark task applied directly to a real mission class. Free-space optical links suffer from atmospheric turbulence — a nonlinear, time-varying channel. Traditional DSP equalizers are computationally expensive at high data rates.

Approach	Power	Latency	Advantage
DSP on FPGA	2-5W	~100ns-1µs	Mature, precise
Betsy (sequential sweep)	~0.5W	~5µs	Power; analog simplicity; no high-speed ADC
Betsy (parallel comb)	~0.5W	~100ns	Competitive on latency AND power

4. Medical — real-time biosignal processing (EEG, EMG, ECG)

Neuromorphic chips (Intel Loihi, IBM TrueNorth, SpiNNaker) already target this market with better ecosystem support and lower power. The one genuine advantage: no silicon in the sensor path, which matters for MRI-compatible devices or extreme isolation requirements. Niche but real.

5. Industrial vibration monitoring in extreme environments

A $5 microcontroller solves this for most environments. Betsy's value proposition only appears where no silicon survives — very high temperature, radiation, or chemical exposure. Jet engine turbine monitoring is the specific case. Small market, severe qualification requirements.

Autonomous vehicles, medical imaging, consumer electronics: no compelling advantage over existing solutions. Cars have power, space, and cooling. Medical imaging requires precision Betsy cannot provide. Consumer electronics require cost and power reductions of 100-1000× beyond current architecture.

25.2 Honest Summary Table (DeepSeek — verbatim)

Application	Fit	Why Betsy?	Why NOT Betsy?
Drone obstacle avoidance	Strong	Microsecond latency, low power, no heavy GPU	Laser power still high; needs integration
Small satellite attitude control	Strong	Radiation-robust cavity, no rad-hard electronics for cavity itself	Laser and readout still vulnerable
Free-space optical comms equalization	Strong	Your P3 benchmark. Direct fit. Power advantage over FPGA DSP.	Sequential sweep slower than DSP; parallel comb closes gap
Medical implants (EEG/EMG)	Weak	All-optical front end for MRI compatibility	Neuromorphic ASICs better on power and ecosystem
Jet engine monitoring	Moderate	Survives high temperature	Tiny market; qualification hell
Self-driving cars	None	—	GPU clusters already work fine
Consumer wearables	None	—	Cost and laser power prohibitive

The real answer: Betsy's best use case is whatever problem requires high-dimensional analog computation in an environment where digital electronics are too heavy, too hot, too radiation-sensitive, or too expensive. That is a real class of problems. It is not a large class. The commercial path is to find the application that needs Betsy's properties, not one where Betsy competes with cheaper, better-established alternatives. (DeepSeek)

25.3 Mission Profile: Betsy on a 6U CubeSat for Optical Downlink Equalization

A 6U CubeSat (10×20×30cm, ~12kg max) in LEO (400-600km) with a laser communication downlink to a ground station. The satellite collects imagery or sensor data; the bottleneck is downlink bandwidth. Atmospheric turbulence distorts the optical beam. Betsy's job: real-time nonlinear channel equalization — the P3 benchmark task applied as a mission requirement. (DeepSeek)

Key Numbers

Parameter	Value	Notes
CubeSat class	6U	10×20×30cm; sufficient volume for full Betsy system
Orbit	500km LEO	Typical for Earth observation; 90-min period
Downlink data rate	1-10 Gbps (near-term realistic)	100 Gbps is aspirational — requires significant aperture and pointing precision. DSOC demonstrated ~267 Mbps at 31M km; LEO geometry much more favourable. (GPT correction of DeepSeek's 10-100 Gbps)
Atmospheric turbulence coherence time	1-10 ms	Varies with wind speed, altitude, weather
Betsy inference time (sequential)	~5 µs	200,000 inferences/sec — 200× faster than turbulence changes
Betsy inference time (parallel comb)	~100 ns	All N_f channels simultaneously; see Section 7.5.1
Ground-to-LEO round-trip delay	5-20 ms	Critical constraint for Option A — see corrected framing below

Architecture Options — Corrected Framing

Option B — Ground-Based Post-Detection Equalization (Primary Architecture): The satellite transmits a distorted optical signal. The ground station receives it and passes it through a Betsy cavity for equalization before decoding. Standard post-detection compensation — no causality issues, no channel inversion required. Training is solved by embedded pilot symbols — known input-output pairs are available at the ground station. This is the primary architecture for first demonstration. Physically coherent. Training pipeline is closed.

Option A — On-Satellite Pre-Equalization: reframed as bias correction only.

Original framing: the satellite equalizes the signal before transmission, using ground telemetry feedback to update the Betsy honeycomb configuration.

Problem: Full pre-equalization requires the transmitter to invert a channel it cannot observe in real time. Ground-to-LEO round-trip delay is 5-20ms. Atmospheric turbulence coherence time is 1-10ms. The channel changes faster than the feedback loop closes. A fast system applying a stale channel inversion can worsen the link.

Corrected framing: Option A is defensible only as low-frequency bias correction — compensating for slow-varying components such as beam wander bias and systematic pointing error, which evolve on timescales longer than the feedback delay. Full turbulence inversion on the satellite is not physically coherent as framed. Additionally, Option A lacks real-time ground truth at the satellite — without known input-output pairs, the reservoir cannot be trained or adapted on-orbit. Option B has a solved training pipeline (pilot symbols). Option A does not. (GPT correction)

Power and Volume Budgets (DeepSeek)

Component	Power	Volume	Notes
Erbium-doped fiber laser (stabilized)	500-1000 mW	~0.2-0.3U	Dominant consumer. Laser is the system weak link — radiation, thermal stabilisation, and power all concentrate here.
Betsy cavity (passive)	0 mW	~0.01U	No power consumed inside cavity. Diamond hemisphere is physically tiny.
Photodetector array + readout	50-100 mW	~0.1U	~400 detectors, low-speed ADC
Decoder (linear regression)	10-50 mW	~0.1U	Could be analog or low-end FPGA
Total Betsy system	~0.6-1.2 W	~0.5U	Compare: FPGA DSP equalizer 2-5W, larger board, cooling required

Mission-Specific Falsification Criteria

Condition	Consequence
Required equalization taps > N_f = 50 frequency steps under strong turbulence (small Fried parameter r₀)	Architecture requires wider sweep or different encoding. (GPT: add performance vs r₀ to atmospheric benchmark)
Ground station cannot close link with Betsy's SER under real atmospheric statistics	Performance claim fails in real atmosphere, not just simulation
Laser and control electronics cannot be radiation-qualified within mission cost envelope	Space application not commercially viable even if cavity physics work
Ferroelectric retention < mission duration (months-years)	Option C fails for space applications; static configuration only

Validation Path — Before Space

Step 1 — P3 Benchmark: Existing plan. Establish SER improvement over ring-resonator baseline. Validates reservoir computing claim but not the atmospheric channel claim.

Step 2 — Atmospheric Chamber Test (critical): Replace ring-resonator baseline with a controlled turbulence generator replicating Kolmogorov statistics at specified Fried parameter r₀. Does Betsy's SER advantage hold under realistic atmospheric statistics? Without this step, the space application is a story, not a validated claim. This is the connection between P3 and the mission.

Step 3 — Field Test at km-scale: Briefcase-sized ground receiver with Betsy cavity against a laser transmitter at 1-5km. Validates full ground-based equalization pipeline under real, uncontrolled atmospheric conditions.

Step 4 — CubeSat Integration Study: Paper study with a university cubesat lab — mass, power, thermal, radiation. Appropriate after Step 3. TRL 4-5 activity.

Step 2 (atmospheric chamber test) has not been designed. No turbulence generator specification exists. No Fried parameter target for the benchmark has been defined. Without Step 2, the connection between P3 and the space application is asserted but not validated.

25.4 Nested Two-Cavity Architecture — Proposed Extension

A single Betsy cavity has a hard effective rank ceiling — approximately 130-150 for 32% surface coverage, bounded by the coverage-limited saturation identified in Section 7.6. A nested two-cavity architecture breaks this ceiling by composition and addresses the laser reliability and readout failure mode problems simultaneously.

Coarse + Fine equalization layers: Cavity 1 handles low-frequency components — beam wander bias, systematic atmospheric drift, slow-varying channel structure (timescales >1ms, correctable on the feedback delay timescale). Cavity 2 handles high-frequency components — fast turbulence, residual nonlinearity, scintillation. The two cavities operate on orthogonal parts of the problem space. Their combined effective rank is higher than either alone because they address different frequency bands with different mode structures. This is the collective CoNexus architecture (Section 13) instantiated for a specific mission problem.

Laser redundancy: Two independent laser sources driving two independent cavities. If one laser degrades under radiation, the other maintains equalization at reduced performance rather than total failure. Directly analogous to modular industrial laser design — swappable units rather than single failure point.

Readout redundancy: Two cavities with partially overlapping readout coverage. A dead spot on one skin is covered by the other cavity's mode structure. Geometric redundancy without doubling computational complexity.

Inter-cavity coupling remains the open design gap (Section 13) — no proposed solution exists. A nested architecture requires the output of one cavity to feed the input of the next, which means solving optical routing between cavities at the signal level. The governance layer that decides which cavity handles which part of the signal adds digital control complexity that partially offsets the analog simplicity advantage. Volume budget grows from ~0.5U to ~1-1.5U (still within 6U). Power budget roughly doubles to ~1.2-2.4W (still well below the FPGA alternative at 2-5W).

Status: The nested two-cavity architecture is a proposed extension, not a validated design. It is presented here to establish the connection between the mission profile and the CoNexus collective architecture concept. It is not a near-term deliverable — it is what Betsy grows into if P3 succeeds.

25.5 Collaboration Targets — Mission-Specific

Organisation	Role	Approach timing	TRL gate
University of Florida (Space Lasers Lab)	Atmospheric turbulence characterisation; accessible first conversation	Immediately — before P3	None required
MIT Lincoln Laboratory	Optical comms for smallsats; lasercomm terminal development	After P3 and atmospheric chamber test	TRL 4-5
NASA JPL	Deep Space Optical Communications (DSOC); interested in lower-power alternatives	After atmospheric chamber test	TRL 4-5
Tesat-Spacecom (Germany)	Commercial laser communication terminals; potential integration partner	After P3	TRL 3-4
TU Delft / Stanford SSDL / MIT SSL	CubeSat integration study, radiation testing, mission design	After T2 simulation complete	TRL 3-4

25.6 TRL Assessment

Stage	TRL	Milestone
Current (concept document v1.11)	TRL 1-2	Physics coherent; no hardware demonstrated
After T2 simulation (2.5D axisymmetric MEEP)	TRL 2-3	Physics validated computationally
After P3 benchmark	TRL 3-4	Component validated in laboratory environment
After atmospheric chamber test	TRL 4-5	Component validated in relevant environment
After km-scale field test	TRL 5	System prototype demonstrated in relevant environment
CubeSat demonstrator	TRL 6-7	Appropriate only after Step 3 — not a near-term project

Appendix A — Outreach Templates

Element Six — Optical Grade Hemisphere Specification

Subject: CVD Diamond Hemisphere Specification Enquiry — Optical Grade Surface Quality

We are evaluating CVD diamond plano-hemispherical cavities (25–30mm diameter) for an optical resonator application requiring surface roughness below 1nm RMS on the inner curved surface, optical purity suitable for near-infrared transmission at 1550nm, controlled wall thickness uniformity, and low stress birefringence. We understand Element Six manufactures CVD diamond hemispheres in high volume for acoustic applications.

Specific questions: (1) Can existing manufacturing processes be brought to optical grade surface specifications, and what process modifications would be required? (2) What are the likely lead times and minimum order quantities for optical grade prototypes? (3) What characterization data (Raman, absorption spectroscopy, surface metrology) do you routinely provide, and what additional characterization can be arranged? (4) Are there existing customers using Element Six CVD diamond hemispheres in optical cavity applications we could be connected with?

We are in early-stage research; initial requirement is characterization samples (2–5 hemispheres) to validate optical grade specifications before any production commitment.

Fraunhofer IAF — Physics Validation Partnership

Subject: Research Collaboration — Graphene-on-Diamond Characterization for Optical Cavity Application

We are developing a speculative photonic reservoir computing architecture based on a CVD diamond hemispherical optical cavity with in-place graphene modulation via SCD(111) Ni-assisted graphitization (Route F). We have conducted Bayesian quality assessment of ten graphene integration routes and identified Route F as the Pareto-front option for our application. We are seeking a physics validation collaboration for two initial experiments.

Proposed Experiment 1: Characterize Route F graphene quality on flat SCD(111) at current process maturity — Hall mobility, carrier density, domain size, D/G ratio, monolayer coverage — against quality thresholds derived from our optical tolerance analysis. Determine whether current Route F mobility (estimated 2,000–3,000 cm²/V·s) is approaching the 3,500–4,500 cm²/V·s crossover threshold.

Proposed Experiment 2 (if Experiment 1 is positive): Characterize Route F graphene quality on a shallow curved CVD diamond surface to assess whether curvature affects the five quality criteria. This is novel work with no literature precedent — potentially publishable as a standalone result.

We can provide the quality criteria framework and Monte Carlo simulation tool developed in our graphene integration route assessment paper. We are an independent research initiative without institutional affiliation; we are open to discussing collaboration structures appropriate to that context.

Ghent University / Bogaerts Group — Reservoir Computing Benchmark Partnership

Subject: Research Collaboration — Photonic Reservoir Computing Benchmark: Diamond Cavity vs Ring-Resonator Baseline

We are developing a speculative photonic reservoir computing architecture based on a CVD diamond optical cavity with engineered disorder and spectral multiplexing. We have identified nonlinear channel equalization as the appropriate benchmark task and have formulated a specific, falsifiable performance claim: a Betsy-class cavity reservoir achieves lower symbol error rate than a ring-resonator photonic reservoir of equivalent node count (400 nodes) because three-dimensional mode volume and scatterer-induced mode mixing produce a flatter H_coupling singular value spectrum.

We are planning a planar validation experiment (P3) on flat SCD(111) diamond substrate with Route I graphene honeycomb, sparse dielectric scatterers, 50 GHz tunable laser sweep, and electronic readout layer. We are seeking collaboration on: (1) experimental design for the P3 reservoir benchmark against your published ring-resonator baseline; (2) access to imec fabrication infrastructure for graphene-on-diamond integration if appropriate; (3) co-authorship on the P3 result (positive or negative).

We can share the full concept document and T2 simulation specification before any commitment. The T2 simulation result (which we are running now) will provide quantitative rank predictions and SER estimates to anchor the collaboration proposal.

Appendix B — Control Architecture Comparison

Property	Betsy (this work)	Ring-resonator reservoir	Multimode fiber reservoir
Physical dimensionality	3D volumetric modes; r_eff ~600 projected (unverified)	2D planar; typically 50–400 effective nodes	Quasi-2D waveguide modes; thousands of spatial channels
Spectral multiplexing	Yes — frequency sweep is core operating mode	Limited — wavelength channels add complexity	Yes — well-established in MMF imaging
Ferroelectric learning	Yes (Option C) — slow structural adaptation	No — static reservoir	No — static reservoir
Programmable disorder	Yes — scatterer density tunable	No — fixed coupling topology	No — fixed fiber disorder
Decoder complexity	Option A: linear regression (standard); Option C: implicit (uncharacterized)	Linear regression (standard)	Linear regression (standard)
Benchmark readiness	Task specified; T2 simulation pending; P3 not yet run	Multiple published benchmarks including NARMA-10, channel equalization	Published benchmarks; commercially available
Fabrication risk	Very High — multiple absent capabilities; rim bonding is go/no-go dependency	Low — CMOS-compatible, foundry-available	Very Low — off-the-shelf component
Primary failure mode	r_eff does not scale with N_f; or SER does not improve over ring-resonator baseline at P3	Dimensionality limited by ring count; nonlinear dynamics limited	Static — cannot adapt to task; no structural learning
Distinguishing claim	3D mode volume + ferroelectric learning + programmable disorder	CMOS compatibility, scalability, precision control	High dimensionality, low cost, room temperature
Current TRL (technology readiness)	TRL 1–2 (concept formulation)	TRL 4–6 (component validated in relevant environment)	TRL 7–8 (system prototype demonstrated)

Appendix C — What Would Convince a Skeptic

This document is optimized for disproof, not narrative momentum. The following statements constitute the minimum evidence package that would convince a technically competent, initially skeptical physicist or fabrication engineer that Betsy deserves H-track investment.

C.1 What Would Convince a Wave-Chaos Physicist

A simulation showing that a planar diamond cavity with the specified scatterer density produces Δf_corr ≈ 1 GHz (confirming near-chaotic field statistics at 1550nm in this geometry), combined with a rank scaling plot showing r_eff(N_f; ε) growing approximately linearly with N_f and crossing 50 at N_f ≤ 50. The simulation must use actual cavity eigenmodes, not the RWM surrogate, and the ε cutoff must be pre-registered from a noise model, not chosen after seeing the spectrum.

C.2 What Would Convince a Photonic Reservoir Computing Researcher

A P3 experiment showing SER improvement >10% over a ring-resonator baseline with equivalent node count on the standard nonlinear channel equalization benchmark, with the improvement attributed to scatterer-induced mode mixing (confirmed by a control experiment removing the scatterers and showing SER degradation), and a dataset showing that the linear decoder trained on the Betsy readout vector is not merely fitting measurement noise (confirmed by the null hypothesis control: training on a random projection of equivalent dimension produces worse SER).

C.3 What Would Convince a CVD Diamond Fabrication Engineer

Element Six confirming that optical grade hemisphere specifications (<1nm RMS surface roughness, optical purity, controlled birefringence) are achievable with a defined process modification and a quoted lead time. AND a demonstrated rim bonding protocol achieving optical quality joint between curved diamond and flat mirror in a test geometry, with cavity Q measurement confirming the join does not degrade finesse below the required threshold. Without both of these, the H-track fabrication program has no supply chain and no assembly method.

C.4 What Would Convince a Materials Scientist (Graphene)

Fraunhofer IAF data showing Route F graphene on flat SCD(111) at or above the 3,500–4,500 cm²/V·s mobility crossover threshold, combined with a first measurement of Route F graphene quality on any curved diamond surface showing that the five Goldilocks criteria are not materially degraded by substrate curvature. This does not require a hemisphere — a shallow spherical cap or cylindrical diamond surface would provide the first data point.

C.5 The Minimum Package for H-Track Authorization

All four of the above, in that order. The T2 simulation and Element Six supply chain conversation can proceed in parallel immediately. Fraunhofer IAF collaboration and P3 experiment require T2 simulation output. H-track authorization requires P3 validation, rim bonding solution, and graphene quality on curved surface. The sequence is not optional — each gate is a prerequisite for the next.

A Final Note

v1.11 is the most complete and most honest version of this document. It is also the version with the most red boxes — things that do not exist and must be created. That is correct. Improved understanding produces more precise identification of what is missing, not fewer gaps.

The physics does not obviously say no. The T2 simulation will say something definitive. The microwave scale model has not been built. The readout sparsity calculation has not been done. The atmospheric chamber test has not been designed. The Element Six conversation has not happened. The Ghent and University of Florida conversations have not been initiated. Those are the next six actions, in that order. Everything else waits on those.

Betsy is the experimental wedge into reality. CoNexus is what it is a wedge into.

Named after a dear friend who, like the concept itself, was full of light and not easily defined.