Real applications
Over a hundred figures built from the data a working measurement actually produces – an instrument sweep, a simulation grid, a production log, a market tape – grouped by the field they come from.
These are not plot-type demonstrations. Each one starts from the measurement and lets the figure follow from it: what the sign of the quantity means, how many decades it spans, which samples are missing, whether the sampling grid is rectilinear at all, and which of the axis, scale and colour choices are forced by the data rather than picked for looks. The plotting call is usually the short part; the reasoning above it is the example.
Between them they exercise most of the library – lines, steps, stems, bars, error bars, histograms, box plots, event rasters, filled bands, stacks, meshes, images, contours, quiver fields, hexbins, twin and shared axes, inverted and categorical axes, log scales, logarithmic colour norms, polar projections, colorbars and figure legends – which is the other reason they exist: a gallery this broad is a regression suite that happens to be readable.
Every figure below is also live. Pick a tool on the embedded copy and zoom, pan, point-pick or annotate it; nothing is active until a tool is selected.
Earth, ocean and atmosphere
Remote sensing, weather radar, climate records and subsurface geophysics. The recurring problems here are incomplete coverage (land under an ocean sensor, a radar’s unfilled corners), quantities whose sign carries the meaning (radial velocity toward or away, an anomaly above or below a reference), and grids that are polar, curvilinear or unevenly spaced before they ever reach the plotting call.
Ice-core CO2 and temperature over four glacial cycles
The zonal-mean seasonal cycle, animated month by month
Sea-surface temperature through the seasonal cycle, animated
Astronomy and space science
Photometry, radio dynamic spectra, aperture-synthesis beams and rotation curves. Astronomical quantities routinely span many decades, are quoted in magnitudes that run backwards, and arrive with per-point uncertainties that are the whole argument – so log axes, inverted axes and error bars are the defaults rather than the decorations.
Folding a pulsar: signal-to-noise improving pulse by pulse
Medical imaging and clinical data
Reconstruction-domain data (k-space, sinograms), depth-resolved imaging, and the time series and survival curves that clinical work is actually decided on. Several of these are the raw domain rather than the picture a clinician sees, which is exactly why they need a log-compressed or high-dynamic-range colour treatment to be readable at all.
Biology, genomics and epidemiology
Expression matrices, differential-expression statistics, structural biology scatter densities, growth kinetics and compartment models. These bring a different set of plotting problems from instrument data: thousands of points that must not be drawn one by one, significance thresholds that belong on the figure as reference lines, and quantities that only behave linearly once they are logged.
Chemistry and spectroscopy
Separations, spectra, correlation maps and electrochemistry. Spectroscopy has strong plotting conventions that are worth honouring even though they look backwards – infrared runs right to left in wavenumber, NMR runs backwards in ppm, transmittance is read downward – and the examples show how to get them with axis inversion rather than by pre-flipping the data.
Infrared spectrum with the conventional reversed axes
UV-vis spectrum through a first-order reaction, animated
Materials, surfaces and microscopy
Diffraction, texture, grain structure, mechanical testing and hyperspectral mapping. Two themes run through the section: reciprocal-space and pole-figure data live on non-Cartesian grids that must be projected before plotting, and material properties are so often power laws that the fatigue and diffraction examples are naturally read on log axes.
Grain size distribution: which mean is the right mean
Spectroscopy and wavefunctions
Two-tone and multi-photon spectroscopy, Floquet sidebands, avoided crossings, and the wavefunctions and quasiprobability distributions those spectra are signatures of. The shared shape: drive a frequency (or a photon number, or a drive strength) along one axis and read a response along another.
Anharmonicity from flux-swept 0-1 and two-photon 0-2 spectroscopy
Floquet sideband spectroscopy of a flux-modulated qubit
Fluxonium level spectrum from exact numerical diagonalization
Readout and device maps
The signal chain a qubit is read out through, and the two-dimensional gate or bias maps that characterize a device before it is used: charge stability, conductance, switching probability, and the mixer and amplifier behavior that sets how much of that signal survives to the digitizer.
Active reset calibration: residual excitation vs rounds and pulse amplitude
Measurement-induced state transitions vs readout power and duration
IQ mixer calibration: suppressing the image sideband
Josephson parametric amplifier: gain versus pump power and bandwidth
Coherence and noise
How a qubit loses information, and to what: T1/T2 decay, Ramsey dephasing against a wandering environment, a two-level-system defect drifting through resonance, and the noise-bias engineering (dynamical decoupling, a Kerr-cat’s protected subspace) built to outlast it.
Qubit coherence: T1 relaxation and T2 Ramsey fringes
T1 relaxation vs frequency and time: wandering TLS defects
Ramsey chevron: frequency calibration from a fringe pattern
CPMG dynamical decoupling: coherence extension vs number of pulses
Kerr-cat qubit: exponential bit-flip protection, linear phase-flip cost
Gate calibration
Tuning up the control itself: Rabi chevrons, DRAG and AllXY amplitude calibration, two-qubit CZ and cross-resonance characterization, crosstalk nulling, and the amplitude/duration limits a gate runs into before it starts leaking out of the computational subspace.
Error-amplification gate calibration: pulse-angle error vs repetitions
DRAG coefficient calibration from amplified leakage
AllXY calibration sweep: amplitude scaling error across 21 sequences
CZ gate calibration: the 11-02 avoided crossing chevron
Static ZZ crosstalk vs tunable-coupler flux and qubit detuning
Single-qubit gate speed limit: Rabi map with a leakage threshold overlaid
Benchmarking and multi-qubit
System-level validation once individual gates are calibrated: randomized, purity and cross-entropy benchmarking, leakage-aware variants, and the multi-qubit signatures – GHZ parity, a Bell/CHSH correlation map, a process’s full Pauli transfer matrix, a repetition-code syndrome – that only show up once several qubits act together.
Single-qubit RB fidelity across the flux-tuning range
Simultaneous RB: crosstalk-limited fidelity vs spectator activity
Purity RB: separating coherent from incoherent gate error
Cross-entropy benchmarking fidelity vs circuit depth and qubit count
Leakage RB: population escaping the computational subspace
GHZ parity oscillation: N-fold fringes and their fading visibility
Quantum process tomography: the Pauli transfer matrix of a noisy gate
CHSH correlation map: where quantum mechanics beats any local theory
Semiconductors and electronic test
Wafer-level maps, device I-V and C-V characterisation, and the link-margin plots that qualify a serial interface. Semiconductor data mixes scales aggressively: the same transistor sweep spans picoamps to milliamps, and a bit-error rate is only interpretable over ten decades, so a linear axis is usually the wrong default.
C-V measurement and the doping profile extracted from it
MOSFET output characteristics, swept smoothly in gate voltage
Fluids, heat transfer and mechanical engineering
Conduction and convection fields, vortex dynamics, velocimetry, turbomachinery and the design charts engineers still size equipment from. Vector fields need thinning before they are legible, conjugate heat problems need a colour scale that is honest about sign, and the classical design charts – Moody, pump curves, airfoil polars – are log-log or twin-axis figures by construction.
Stokes’ first problem: viscous diffusion from an impulsively started plate
The decaying Taylor-Green vortex: an exact solution, animated
Acoustics, vibration and sonar
Room modes, decay curves, loudspeaker response, machinery vibration and passive sonar. Acoustics is a decibel discipline: nearly every quantity here is already logarithmic, so the axis choice is about frequency spacing and dynamic-range floors rather than about whether to take a log at all.
A pressure pulse bouncing down a rigid duct, animated
A standing wave, actually standing: room modes in time
Energy and power systems
Photovoltaic and wind characterisation, battery ageing, and the load and generation curves a grid operator plans against. The distinctive plotting need is showing two quantities that share an x axis but nothing else – current and power, capacity and efficiency – and showing composition over time without losing the total.
Transport, robotics and navigation
Drive cycles, traffic flow, automotive radar, lidar scans and flight envelopes. Vehicle data is where a plot’s native coordinate system most often is not Cartesian – a lidar returns range against bearing, a radar returns range against Doppler – and where an operating envelope rather than a curve is the thing being drawn.
Stall creeping outward as Reynolds number climbs, animated
Manufacturing, reliability and quality
Control charts, capability studies, Pareto analysis, life-test fitting and production scheduling. These are the figures where the reference lines carry the decision – control limits, specification limits, a cumulative 80% cut – so most of the work is making the limits unmistakable against the data rather than drawing the data itself.
Computing, networks and machine learning
Service latency, classifier evaluation, training diagnostics and hardware performance models. Systems data is heavy-tailed almost by definition, which is why the tail percentile rather than the mean is plotted, and why the roofline and scaling charts are log-log: the interesting behaviour is a change of slope, and only a log axis makes a slope change visible.
ROC and precision-recall for an imbalanced problem
Finance, economics and risk
Drawdowns, term structures, return distributions, portfolio frontiers and option surfaces. Financial series are cumulative and multiplicative, so levels belong on a log axis and losses belong measured from the running peak; and the tail of the return distribution – not its centre – is the part a risk figure exists to show.
Annotating a price series: a single event vs. a whole regime
