# Research-table dictionary

All files are UTF-8 JSON Lines. Each line is one record. Every configuration uses the split label `train` as a Hub convention; no row is designated an ML training example. Numerical examples are synthetic and dimensionless. The ten tables contain 1,165 rows in total.

## Shared provenance fields

- `row_id`: unique identifier within its configuration.
- `source_file`: path to the authoritative raw JSON file within the repository.
- `source_pointer`: RFC6901 JSON Pointer to the source record or value. The empty string points to the whole source document. Exported rows can include adjacent source fields as described below.

`docs/HF_TABLES_MANIFEST.json` records each table's path, count, and column names. No null is imputed during conversion.

## claims — 30 rows

`id`, `status`, `statement`, `hypotheses`, `evidence`, `limits`, and `worldwide_novelty_confirmed` preserve the claim registry. Lists remain lists of strings. Status distinguishes conditional proof, existing methods, numerical evidence, proposed experiments, and unestablished goals. A claim row is not evidence by itself; read its referenced proof and source data.

## reliability_bounds — 8 rows

`radius`, `kappa`, `diffusion`, and `T` specify the centered planar OU example. `diffusion` is the actual coefficient d in `dZ=-kappa Z dt+sqrt(2d)dB`, not an unmultiplied strand coefficient before applying pair covariance. `beta=kappa*radius^2/(2d)` is the dimensionless barrier.

`endpoint_lower` bounds continuous exit from below. `v5_direct_upper` and `v5_block_upper` are old model certificate values; they are not device failure measurements. `poisson_upper`, `laplace_upper`, `additive_upper`, and `upper` are evaluated analytic upper bounds; `upper` takes the smallest implemented candidate. `laplace_s` and `laplace_log_upper` retain the transform optimization parameters. `mean_exit_time` is the mean-exit identity evaluated numerically.

`pde_exit_estimate`, `pde_exit_resolved`, and `pde_cells` expose the 800-cell numerical survival result. If unresolved, the estimate is null. All numerical upper-bound decimal evaluations use floating arithmetic, not outward-rounded interval evaluation.

## grid_refinement — 32 rows

Each example has four cell counts: 100,200,400,800. `exit_estimate` is a finite-volume estimate; `exit_resolved` marks whether it exceeds the declared numerical-resolution threshold. `mean_relative_error` and `laplace_relative_error` compare the spatial solver with independent analytic mean and transform calculations.

## sampled_monitoring — 5 rows

All rows describe nested observation grids on the same 30,000 paths. `stride*fine_dt` is `observation_interval`. `detected_exits`, `sampled_exit_rate`, and Wilson interval limits concern sampled crossings, not continuous first exits. `continuous_pde_estimate` is a separate numerical comparison. `same_nested_paths` is true and `continuous_monitoring` is false. The seed is 20261013; no full trajectory array is included.

## clearance_design — 65 rows

`T`, `kappa`, `diffusion`, `error_budget`, and `events` specify the design policy. `poisson_radius` and `v5_direct_radius` are two sufficient radii, not hardware optima. Event independence is not assumed by the union budget.

## graph_budget — 1,000 rows

Each `allocation_index` identifies a saved variance outcome from the seeded allocation generator. `nodes`, `gain_budget`, `kappa`, and `noise_covariance_q0` describe the fixed linear covariance model. `worst_pair_stationary_variance` is the random allocation result. `proved_uniform_optimum` and `computed_uniform_value` give the symmetric reference.

`edge_gains_recorded` is false: no full gain vector was saved in the raw file. The seeded code can reproduce the random allocations; the table itself cannot reconstruct an allocation from its scalar variance. The source uses seed 62291, n=8, G=4, kappa=1, q0=2. Gain budget is not calibrated physical power.

## quantum_holonomy — 7 rows

`A=|alpha|^2`, `fock_dimension`, and `phase_steps` specify the ideal cat-code calculation. `mean_n_even` and `mean_n_odd` are photon means. `ideal_relative_phase_radians`, `phase_defect_from_pi_radians`, `overlap_quadrature_relative_phase_radians`, and `wrapped_absolute_quadrature_error` compare the ideal formula and overlap quadrature.

`parity_jump_rate_even_per_kappa1` and `parity_jump_rate_odd_per_kappa1` flatten the source two-element jump-rate vector. They are parity-jump diagnostics, not logical-error rates after recovery. Every row assumes ideal adiabatic code transport; no global dissipative gap is certified.

## quantum_cutoff — 5 rows

The fields retain the truncated-state boundary residual, direct floating residual, Poisson tail, minimum computed jump singular value, log absolute determinant, and exact kernel dimension. `common_jump_eigenvalue_real` and `common_jump_eigenvalue_imag` are the real and imaginary parts of the jump's common eigenvalue. The source pair is not two distinct eigenvalues.

`exact_kernel_dimension=0` is the finite nilpotent-cutoff algebraic statement when the constant polynomial term is nonzero. It does not describe the infinite oscillator's dark-state dimension. A small numerical singular value is near-dark behavior, not an exact zero eigenvalue.

## stationary_averaging — 12 rows

`lambda`, `T`, and `theta` are the positive scalar OU spectral rate, averaging horizon, and noise parameter. `closed_form_variance` and `independent_quadrature_variance` calculate the stationary time-average variance; `relative_difference` records their agreement. Stationary preparation is an assumption, not a measured hardware condition.

## calibration_coverage — 1 aggregate row

`repetitions=5000`, `scalar_innovations_per_repetition=40`, and the model parameters describe the exact Gaussian calibration experiment. `covered=4758` and `empirical_coverage=0.9516` report the one-sided confidence procedure's coverage. `wilson_95_low` and `wilson_95_high` describe Monte Carlo uncertainty in that coverage. Known kappa and exact uncensored observations are essential assumptions. The individual innovation arrays are not included.