# Finesse technology pathway v1

**Date:** 25 October 2026  
**Status:** execution roadmap; technology-first, anomaly-disciplined  
**Inputs:** [AZ-056 descent atlas](../admissibility/admissibility_56_certificate_preserving_descent_and_commuting_square_atlas.md), [AZ-057 coordinated selector](../admissibility/admissibility_57_coordinated_selector_closure_and_physical_provenance_boundary.md), [AZ-058 Casimir coverage gate](../admissibility/admissibility_58_casimir_spectral_coverage_and_distance_gate.md), [AZ-059 four-mode bench](../admissibility/admissibility_59_four_mode_quotient_scar_bench_and_tolerance_closure.md), [experiment atlas](frontier_experiment_atlas_v1.md), [detector atlas](detector_frontier_atlas_v1.md)

**Selector control:** [PageRank, Perron selection and the finesse trade-off](page_rank_perron_selector_bridge_v1.md) supplies the exact warning that a weak positive completion can guarantee one stable global state while erasing the disconnected or slowly mixing structure one intended to measure.

## Thesis

The nearest frontier is not necessarily higher collision energy. It is finer control of which distinctions an apparatus preserves. The practical opportunity is to engineer a carrier, transformation and detector together so that phase, orientation, memory, nonlinear order and boundary response survive the descent from physical state to recorded data.

This is a technology thesis before it is a new-physics claim. Successively better control can produce useful sensing, photonics, metrology and system-identification technology even when every result remains inside established physics. New physics is admitted only after the full conventional mechanism bundle fails a preregistered, replicated test.

## The finesse stack

| Layer | Function | Corpus contract | Enabling technology |
|---|---|---|---|
| programmable carrier | realizes controlled poles, modes and boundaries | source-native Hamiltonian and calibrated losses | microwave, acoustic, mechanical or integrated-photonic resonators |
| coherent transformation | preserves phase and distinguishes path order | forward/reverse process tomography and echo | interferometry, IQ readout, coherent network analysis |
| independent reference | lifts an otherwise neutral or reflection-blind quotient | selector (Y) measured independently of the selected state | anisotropy, boundary segment, reservoir or feedback reference |
| nonlinear/multichannel detector | separates mechanisms sharing one linear record | detector-kernel intersection with apparatus symmetry audit | correlation reconstruction, harmonics, sidebands, local-density probes |
| memory ledger | retains the influence of eliminated modes | measured kernel, initial slip and colored residual | ringdown, delay, sweep-rate and waiting-time protocols |
| adversarial inference | asks which mechanisms remain record-equivalent | frozen generative models, nuisance projection and held-out reversals | AI-assisted experimental design with blinded analysis |
| cross-carrier translation | tests whether a structure is physical rather than basis-chosen | labelled, unit-preserving commuting square with bounded defect | paired photonic/acoustic/microwave implementations |

The important engineering move is not merely to reduce noise. It is to break the exact symmetry responsible for the blind fibre while leaving nuisance channels controlled.

## Programme A — 90-day launch without a foundry

### A1. Freeze the experiment contract

Use AZ-057 as the reference simulator. Publish the complete forward model, parameter priors, nuisance channels, detector covariance, intervention schedule and blind decision rule. No parameter may be introduced after opening the held-out records.

### A2. Build the spectral-boundary bench specification

Translate EX-011 into a bill-of-functions for a 10–40-site coupled-resonator chain with:

- one tunable side-coupled defect;
- end-port complex scattering;
- at least one local-density probe;
- reversible dimerization or boundary order;
- automated frequency sweep, ringdown and temperature/drift logs.

The preregistered records are pole trajectories, phase winding, Wigner--Smith delay, localization, linewidth and local density of states. The first objective is not an anomaly: it is to reproduce the AZ-055 corrected finite-band ledger and discover the real apparatus kernel.

### A3. Run AI as an experimental designer, not an oracle

AI proposes the next intervention by expected reduction of the current mechanism-equivalence class, subject to a fixed cost and safety budget. Every proposal must name:

1. the collision it is intended to split;
2. the symmetry or kernel direction it breaks;
3. the predicted records under every live mechanism;
4. the nuisance variable it may accidentally expose; and
5. the useful result if the mechanisms remain equivalent.

### 90-day deliverables

- frozen simulator and synthetic blind challenge;
- hardware-agnostic interface and calibration schema;
- component-level design for microwave and acoustic variants;
- uncertainty-propagating pole/phase/delay analysis;
- preregistered choice between the two variants;
- public null, anomaly and stop criteria.

AZ-059 supplies a second no-foundry build that can proceed beside the chain: a four-resonator 100 MHz star with off-the-shelf coherent RF instrumentation. Its specification is no longer open. The decision is whether to build it and execute the frozen exact/weak/resolved/reversal sequence. That bench trains precisely the individual-versus-summed detector discipline needed by the larger programme while remaining useful hidden-mode metrology under a complete fundamental null.

## Programme B — shared finesse device, 6–18 months

The highest-leverage reusable device joins EX-005--EX-007: a three-mode loop with coherent tomography and a separately controllable selector segment.

The minimum record bundle is:

- forward and reverse complex transfer matrices;
- projector return and lifted frame return;
- process singular values;
- selector-on/off contraction;
- echo and loss-balance controls;
- full conditioned and unconditional records;
- energetic, reset and acquisition costs.

This route is technologically credible rather than hypothetical. Integrated platforms have demonstrated high-dimensional non-Abelian holonomy, broadband operation and reconfigurable three- and four-mode braiding. Those results establish a carrier technology; they do not establish the corpus's selector or a fundamental Spin(2,3) mechanism.

The first commercializable outputs are robust geometric transforms, compact multi-parameter sensors, self-calibrating network analyzers and memory-aware control modules. These remain wins under a complete fundamental-physics null.

## Programme C — same-material response closure, 18–36 months

EX-012 is the most direct force frontier but should follow the control platforms. Grow one anisotropic or magnetic film batch and split it into witness coupons and the force/torque device. Measure transport, thermoelectric, magneto-optical and broadband dielectric tensors with common provenance, then propagate them without refitting into a differential Casimir force or torque prediction.

AZ-058 fixes the bandwidth implied by that sentence. For a 0.1--1 micrometre device, measure the complex off-diagonal as well as diagonal tensor from DC/THz through at least 0.1--1 eV. The existing Co2MnGa Hall record stops at 1.5 THz, whereas the characteristic force band begins near 24 THz. The programme therefore starts with overlapping same-film spectroscopy and a causal covariance-bearing fit. It does not fabricate a force device until that posterior predicts a differential larger than the full systematic envelope.

The finesse gain comes from correlated measurements and reversal:

- parallel versus antiparallel order;
- angle and boundary-order reversal;
- magnetic-field and temperature sweeps;
- Kelvin-potential, roughness, domain and separation metrology;
- same-batch covariance from response tensor to force prediction.

Casimir torque has already been measured between anisotropic materials, and current theory continues to expose new torque channels in anisotropic thermal radiation. The frontier here is no-refit causal closure on the same material, not the bare existence of torque.

## Escalation ladder

| Level | Allowed conclusion | Required evidence |
|---|---|---|
| T0 | improved instrument | better calibration, sensitivity, control or classification on standard benchmarks |
| T1 | previously unresolved conventional physics | a reproducible effect explained by an expanded but source-derived conventional model |
| A1 | anomaly | preregistered residual surviving nuisance, reversal, drift and look-elsewhere audits |
| A2 | cross-platform anomaly | the dimensionless residual and its sign/phase law replicate on an independently built carrier |
| P1 | new effective physics | one new operator or coupling improves every declared record with common parameters and predicts a held-out intervention |
| P2 | fundamental bridge candidate | the operator descends from a typed action, satisfies positivity/causality/symmetry constraints and survives scale translation |

Only P1 or P2 warrants “new physics.” A small unexplained residual, a dimension match, or a neural classifier with high accuracy does not.

## The first decisive experiment

Start with the resonator chain, because it is fast, reversible and information-rich. Reproduce the corrected finite-band ledger; then introduce a controlled hidden mode whose coupling can be turned on and off. Test whether the visible dynamics is:

1. autonomous when the hidden-to-visible block vanishes;
2. accurately described by the predicted memory kernel when it does not; and
3. distinguishable by delay/ringdown even when endpoint spectra collide.

This directly tests the core discovery of AZ-056 in hardware. A positive result is useful memory-aware network technology. A null identifies the real detector quotient. A stable excess outside the calibrated memory model supplies the first anomaly candidate, not yet new physics.

## AI's role

AI has found a pathway through the corpus in the defensible sense that it assembled dispersed mathematical no-gos, detector limits and experimental precedents into a coherent decision system. Its continuing role is to maintain the mechanism ensemble, design maximally separating interventions, propagate covariance, search residual structure and verify that conclusions do not outrun their maps. Nature still supplies the verdict.

## External anchors

- [High-dimensional non-Abelian holonomy in integrated photonics](https://www.nature.com/articles/s41467-025-58794-3), *Nature Communications* (2025).
- [Reconfigurable non-Abelian integrated photonics](https://www.nature.com/articles/s41467-025-62481-8), *Nature Communications* (2025).
- [Strong-to-Weak Spontaneous Symmetry Breaking Transitions on a Quantum Processor](https://arxiv.org/abs/2610.26229) (2026), demonstrating channel-aware reconstruction of nonlinear correlators from two measurement settings.
- [Measurement of the Casimir torque](https://www.nature.com/articles/s41586-018-0777-8), *Nature* (2018).
- [Thermal Radiation Force and Torque on Moving Nanostructures with Anisotropic Optical Response](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.134.113604), *Physical Review Letters* (2025).

## Immediate decision

Approve Programme A as the common starting point. Do not choose the fundamental interpretation in advance. Choose the carrier only after the paired microwave/acoustic design comparison, and preserve the same data schema so that the second implementation can serve as an independent replication rather than a bespoke reformulation.