Turing

Turing is a local scientific visualization workstation. Create a project before importing data so the source, object versions, visualizations, and exports have a persistent context. Supported inputs include CSV, TSV, record and scientific dataset JSON, PDB, XYZ trajectories, and PNG, JPEG, or WebP images.

The import preview reports inferred column types, units, missing data, citation, and license. Check each field before accepting it. The application does not infer scientific meaning from a column name alone, and a declared time variable or unit should be checked against the source documentation.

Use the workbench to create visualizations and connect compatible selections. Turing supports scatter, line, area, bar, histogram, box, violin, density, matrix, contour, graph, timeline, parallel-coordinate, Pareto, spectrum, molecular, Cartesian 3D, surface, mesh, vector, image, and orthogonal-volume representations.

Point selection, region brushing, declared relationship propagation, cross-filtering, and camera bookmarks help inspect a research object from several angles. A linked selection follows declared relationships; it does not imply that two records are causally related. Review the source version and link definition when a selection surprises you.

Turing records transformations instead of rewriting the original dataset. Available analysis includes ordinary least squares, pairwise Pearson correlations, graph connectivity, standardized PCA, and deterministic k-means, among other tools. Review input variables, missing-data behavior, model settings, and task cancellation before interpreting output.

Dataset comparison can match by explicit keys, convert compatible units, and show unmatched records and absolute or relative differences. Faceted plots use shared scales by default to aid comparison, while independent scales remain an explicit option. Preserve supplied uncertainty values and units; Turing will not create missing error bars.

In Figure Studio, arrange panels, order them, define captions and methods, annotate observations, and record review notes. Set physical size and DPI for PNG output. Pin the source and visualization versions reviewed so a figure can be connected to the exact data state.

SVG and PDF preserve plot text while 3D scenes and voxel slices are embedded raster images. Inspect the exported file at its intended publication size. Include source status and version in the figure metadata and report any sampling or scale policy used for dense or faceted views.

Turing exports SVG, PNG, PDF, underlying CSV, visualization JSON, and ZIP reproducibility packages. The package manifest states whether declared input versions are present and includes archived inputs and hashes when available. Verify the manifest and retain the package with the original source files where required.

Imports are bounded to 30 MB and 250,000 table rows. Larger plots may use stated sampling limits. A project package is bounded to 30 MB compressed and 40 MB expanded, with at most 1,000 declared source versions. Check the limits displayed in the application before preparing a large handoff.

Missing numeric observations remain missing rather than being treated as zero. Error bars require supplied nonnegative magnitudes in compatible units. Numeric color scales include endpoints and units, and correlated values report their complete-pair sample size. These details help a chart remain honest about its data.

Plots do not validate scientific conclusions. Inspect whether uncertainty was supplied, whether axes share comparable scales, whether the source is current, and whether a transformation changes the interpretation. The default demo fixtures are illustrative and carry no experimental or validated scientific status.

Create a project for one visualization question before importing files. Supported inputs include delimited tables, record and scientific dataset JSON, PDB coordinates, XYZ trajectories, and common raster images. The import preview reports inferred schema, units, missing data, citation, license, and source status; check those fields rather than relying on file extensions alone.

For PDB data, verify model and atom identity. XYZ imports explicitly assume angstroms unless the source establishes another convention. Tables need declared units and stable keys if they will be compared. Accept the import only after confirming that missing numeric observations remain missing and that all source and licensing metadata are correct.

Turing supports a broad catalogue of chart, graph, molecular, image, and volume views: scatter, line, area, bar, histogram, box, violin, density, matrix, contour, graph, timeline, parallel coordinates, Pareto, spectrum, molecular, Cartesian 3D, surface, mesh, vector, image, and orthogonal volume. Pick a representation based on the variables and question rather than selecting a chart for visual effect.

Axis, color, size, and geometry encodings should retain units and meaning. Numeric color scales include endpoints and units. Missing values remain missing instead of becoming zero. If a plot samples dense data or omits labels, inspect the notice and use the underlying table or inspector for the full record names.

Point selection and region brushing can connect compatible views. Cross-filtering follows declared relationships among scientific objects and can reveal which records contribute to a selected region. Review the relationship definition and source version when a selection propagates; a link in the data model does not imply a causal or functional relationship.

Camera bookmarks and saved recipes help a collaborator return to a selected orientation, filter, or view. Save the visualization state after important changes and keep a note about the intended selection. A saved state is a view of the source, not a copy of the underlying scientific truth; if its input version changes, review the stale indication before reusing it.

The workbench records transformations as derived operations rather than rewriting original data. Available analysis includes ordinary least squares, pairwise Pearson correlations, graph connectivity, standardized principal-component analysis, and deterministic k-means. Review which variables enter each task, how missingness is handled, the method settings, and cancellation status before interpreting the output.

Correlation uses complete pairs and records the sample size. PCA standardizes variables according to its documented method, while clustering is deterministic for the stated inputs and settings. These algorithms summarize the selected data; they do not establish causality, significance, or meaningful biological classes without a scientific review. Preserve the input version and task metadata with any derived table or figure.

Dataset comparison matches records by explicit keys, converts compatible units, and reports unmatched records and absolute or relative differences. Choose the key fields deliberately and review duplicate or missing keys before comparing. A join that silently loses records can create a persuasive but misleading chart.

Unit conversion uses compatible dimensions and preserves source values. Check the conversion path and whether the requested comparison is scientifically meaningful; equal units do not guarantee equal measurement procedures. Keep unmatched observations visible, and state whether your conclusion is based on matched pairs, all records, or a filtered subset.

Faceted plots use shared axes and numeric color scales by default to support cross-panel comparison. Categorical colors retain group mapping; shared histograms use common bins, and shared density views use an attributed pooled bandwidth. Independent scales are available explicitly and should be identified in the figure because they change visual comparison.

Error bars require supplied nonnegative magnitudes in compatible units; uncertainty conversion uses scale factors without affine offsets. Turing does not invent absent uncertainty values. For graph views, arrows require declared direction, paths follow directed edges, and force layout does not encode measured distance. For images and volumes, calibration and anisotropic spacing determine physical measurements; uncalibrated coordinates remain labeled as such.

Figure Studio arranges multiple panels in a chosen order and adds captions, methods, annotations, and review notes. Pin the exact source and visualization versions inspected, then set physical dimensions and DPI for PNG output. A figure should identify its scale policy, sampling behavior, and any supplied uncertainty that informs interpretation.

SVG and PDF preserve plot text as vector elements, while 3D scenes and voxel slices are embedded raster content. Inspect the exported figure at the intended publication size for legibility and inspect the complete data separately from the visible preview. Figure composition improves communication; it does not validate the statistical or scientific conclusion represented by a plot.

Export SVG, PNG, PDF, underlying CSV, visualization JSON, or a ZIP package. The package manifest indicates whether declared input versions are present and includes archived inputs and hashes where available. Verify the manifest before handoff; a package marked incomplete should not be represented as self-contained.

Project archives have file-size, row-count, compressed-size, expanded-size, and source-version limits. The app displays those bounds for large files and packages. Keep original sources and license notes with the export when required. Visualization JSON can describe the view but does not replace the source dataset unless that dataset is included in the package.

Scientific source records retain their originating product, object ID, version, status, citation, license, and file hash. A cross-product handoff should identify the source owner and destination project. Repeated request IDs are idempotent; reuse of a key with changed content is rejected so a retry cannot silently modify an earlier transfer.

When an upstream object changes, existing visualizations can become stale until reviewed and rebound. Confirm the new dataset version and row identities before updating a view. Preserve the old saved visualization if it supports an earlier report or decision. Source ownership is not changed by an atomic visualization command.

A visualization can make a pattern easier to inspect, but it cannot establish that a pattern is real, statistically significant, causal, or generalizable. Check sample size, selection filters, missingness, uncertainty, scale range, axis policy, and data calibration. Use external domain methods where the research question requires a validated statistical model or image/volume interpretation.

Bundled example records are illustrative and explicitly retain that status in the interface and exports. Supported import formats are bounded, and unsupported scientific formats require additional decoders. For 3D and voxel views, inspect coordinate units, voxel spacing, and rendering limits. Keep the method and source context close to every published figure.