Find the structure your question needsExplore the native modalities.
Start with familiar data, or go deeper into scale, topology, geometry and order. Each entry explains the representation, an example and its supported scope.
ScaleMERA tensor networks
Represent information across resolutions, with the relationships between coarse summaries and finer detail kept explicit.
8DB’s MERA layer brings multiscale representations, reconstruction recipes and fidelity metadata into the data system. Trained time-series layers provide a concrete application of this work.
What you can askHow does this signal behave at a broad scale, and what detail does a closer inspection reveal?
Native operations & scope
Implemented components include paired scalar transforms, reconstruction and trained per-series/window layers. Approximation and stale-state metadata matter. N-way Tucker arrays are a separate capability. These implementations do not establish general 2-D/3-D MERA, lossless compression for every dataset or local I/O for every partial reconstruction.
Topological braids
Represent how strands interact over time, preserving crossing order that a static relationship graph leaves out.
Braid representations let applications compare the structure of interacting histories. In a bundle of signals, rank crossings describe how their relative positions change.
What you can askDo these groups of signals follow a similar sequence of crossings, even when individual values differ?
Native operations & scope
Implemented components include braid encoding, strand operations, topology matching and declared graph, hypergraph, vector and timeline projections. Storage codecs and test-authorized reopen controls exist. Time-series rank braids are computed from stored signals. Durable semantic-identity ingestion remains gated while its encoding profile is corrected; not every BraidQL operator is released. Fingerprints are derived summaries, and projections need not reconstruct the original braid.
Geometric manifolds
Keep coordinates together with the geometry that makes distance meaningful.
Two points can be close in straight-line distance and far apart along a surface. Stored charts and declared metrics let 8DB query the geometry relevant to the problem.
What you can askWhich observations are close along this curved surface, rather than through the empty space between them?
Native operations & scope
Local charts and metric kinds support GeoQL geodesic and scalar chart-consistency operations. Metric choices include Euclidean, Riemannian, Fisher–Rao and hyperbolic forms. The documented geodesic path uses a neighborhood graph and shortest-path computation, an approximation to continuous geometry. Full atlas transitions and arbitrary geometric solvers are separate capabilities.
Order lattices
Represent values that are partly ordered, including meaningful alternatives that cannot be ranked on one scale.
BioTwin’s measurement-design prototype uses set-lattice operators to retain competing explanations and ask which observations could distinguish them. Other uses include incomparable evidence and compartmented security labels.
What you can askWhich hypotheses could this joint observation rule out, and which would remain possible?
Native operations & scope
Typed lattice operators support dominance, joins, meets and incomparable frontiers. Carriers include interval enclosures, confidence vectors, security labels and vector clocks. Each carrier has its own rules. A confidence join is not automatically a calibrated probability, and a lattice primitive alone does not establish enforcement across every access path.
n-dimensional spatial data
Work with coordinate spaces beyond latitude and longitude. Native n-dimensional linear/grid indexes support Euclidean, Manhattan and Chebyshev metrics.
For exampleFind nearby simulation states in a declared multidimensional space. Choose coordinates, units and a metric that make “nearby” meaningful.
Scope & measurements
n-dimensional coordinate queries, geographic predicates and 3-D point-cloud operations are distinct paths. Dimensionality affects index behavior and performance. The published spatial comparison is a historical 2-D radius-query workload, not a benchmark for every dimension.
Read the spatial measurementPoint clouds
Keep measured 3-D shape as points with spatial neighborhoods. Native XYZ operations include nearest-neighbor and region queries.
For exampleFind the measured points around a component under inspection, with the scan linked to its sensor, capture time and asset.
Native operations & scope
Recorded execution controls cover XYZ k-nearest-neighbor and region queries, confidence-aware filtering and rejection of nonfinite coordinates. The inspection workflow is an application example. Supported attributes and their retention depend on the representation; a downsampled view need not retain every source attribute.
Hypergraphs
Keep a relationship with three or more participants together as one fact, with named roles and its own identity.
For exampleOne observation connects a sample, instrument, method and time. Query the participants without reconstructing the observation from separate pairwise links.
Native operations & scope
Native n-ary facts support typed participants, participant lookups and role/type intersections. This preserves the meaning of the group relationship. It does not imply that every possible hypergraph pattern has a dedicated query operator.
Time series & events
Preserve change, measurement context and missingness. A zero, a value below detection and a sensor that was not deployed mean different things.
For exampleCompare what was known at the time with a later corrected signal, then inspect its interval rollup or multiscale view.
Native operations & scope
Implemented typed samples include points, enclosures, censored values and absence. The time-series path includes as-of reads, late deltas, source-backed similarity, interval rollups, rank braids and trained multiscale layers. Durable carrier and reopen gates are recorded. Older flat-ingest throughput figures do not measure this newer durable path.
Meshes & raster grids
Use typed triangular meshes and scalar grids to retain connectivity or regular spatial layout alongside values.
For exampleA surface mesh can preserve which vertices form each triangle. A grid can preserve the layout of a measured field.
Native operations & scope
Typed TriMesh values combine spatial vertices with connectivity. Scalar raster grids have their own typed representation. These are implemented data primitives; a mesh type alone is not a full CAD, rendering or simulation package.
Tables, records & key-value
Use typed rows, point and range reads, filters, sorting and projections for the operational facts your application needs.
For exampleRetrieve an asset record, select its inspection results and retain row identity alongside linked evidence.
Measured key-value reads ↗
Documents & text
Keep passages, sections and claims connected to the documents they came from. Native lexical retrieval provides a way back to the supporting text.
For exampleFind the paragraph behind an engineering claim, then follow its relationship to the inspected asset.
Read about provenance as a query ↗
Vectors & similarity
Use dense, sparse and hybrid retrieval to find similar information, with embeddings connected to their source records.
For exampleFind reports with similar descriptions, then inspect their text, dates and relationships before accepting a match.
Native operations & scope
Native paths include cosine similarity, approximate-neighbor indexing and sparse/dense fusion. Embeddings are supplied by a model, with its version, dimension and normalization specified. Similarity is a retrieval signal; it does not establish that a retrieved statement is true.
Graphs
Follow typed relationships and dependencies in either direction, with source context available for the facts being connected.
For exampleTrace which conclusions depend on a changed observation, or follow an asset to its components and inspections.
Measured profile lookup, with scope ↗
Images, audio & video
Keep media and supported derived representations connected to subjects, events, documents and evidence.
For exampleLink a photograph or recorded inspection to the asset, capture event and findings it supports.
Native representation & scope
Typed schemas and registered decomposers preserve supported content and relationships. Format support varies. Semantic interpretation, transcription and embedding generation require the appropriate model or codec; storing media does not automatically supply those abilities.
Arrays, distributions & collections
N-way tensors, probability distributions, trees, sets and multisets provide additional typed building blocks.
For examplePreserve the axes of a scientific array, a distribution’s parameters or repeated membership in a multiset instead of erasing those distinctions.
Native representation & scope
TensorN/Tucker handles N-way array factorization separately from the scalar multiscale path. Distribution, tree, set and multiset types have recorded implementation entries. Each type’s operators and validation contract define what can be concluded from it.