# Data dictionary

All files in `results/` contain simulated or formula-derived data, not observations of an apparatus.

| CSV | Coordinates and units | Quantity |
|---|---|---|
| diffusion_recovery.csv | q rad/m; D m²/s; time s | Heat attenuation, inverse noise gain, bounded decoder error |
| ou_covariance.csv | Dimensionless drift and sampling interval | Covariance and inverse error across 48 repetitions, scalar effective count |
| heat_resolvent.csv | Dimensionless drift; integration steps | Time-integral approximation to inverse, truncation bound |
| parity_budget.csv | Dimensionless amplitudes; independent raw-shot variance | Equal-budget bias, variance, and RMSE |
| attainability.csv | Ordered monomial feature dictionaries | Ranks and target residuals |
| atmosphere.csv | Hz, K, kPa, RH fraction and molar percent | ISO linear absorption dB/m and relaxation frequencies |
| parametric_budget.csv | Hz, m², cycles, acquisitions | Assumed ideal aperture/feature rate and associated absorption |
| species_count.csv | Dimensionless independent binomial counts | Fraction and logit Fisher information; no kinetics |
| semilinear_mode_weights.csv | Dimensionless graph eigenvalues/time | Terminal attenuation vs accumulated kernel and encoder gain |
| semilinear_mixed_hessian.csv | Dimensionless signed amplitudes | Nonlinear ODE mixed response and quadratic-order error |

`experiments.json` stores these rows, parameters, seed, software versions and provenance. The dictionary uses the actual filenames listed by the release manifest. NumPy rank decisions use a stated numerical threshold, rather than asserting exact symbolic rank for arbitrary noisy matrices. Source and receiver coordinates must be experimentally calibrated before assigning these abstract channels to air.