Search files, editable visual templates, and live browser targets from one registry-backed directory. Filtered query views stay crawlable for links but are deliberately noindex; the stable taxonomy pages below remain the canonical search surfaces.
Page 4 of 6; 24 results per page.
JSON Schema reference that validates the HDF5 decoded-result contract, kept beside the binary source for semantic rather than byte-level comparison. Stable P8 artifact p8-convert-hdf5-schema.
A MATLAB v7.3 variable store, which is an HDF5 file with a 512-byte MATLAB user block and MATLAB_class attributes — the bytes are the same whichever extension you give it. Because MATLAB is column-major, A appears in HDF5 as its own transpose, which is the conversion bug this fixture is built around.
A Hermitian complex matrix where each line carries a real and an imaginary part and the implied mirror entry is the complex conjugate rather than a copy. Every diagonal entry has a zero imaginary part, which the Hermitian property requires and a validator should check.
A 4x3 dense matrix where values run down each column rather than across each row, which is the Matrix Market array convention and the opposite of what most readers assume. Getting it wrong rearranges every element rather than raising an error.
A matrix whose banner declares an integer field, so the values are integers and must not be widened to floats on the way in. The stored range deliberately reaches the signed 16-bit limits, so a narrow integer type overflows visibly.
An 8x8 sparse matrix in Matrix Market coordinate form with twelve stored entries and 1-based indices, the convention that catches every reader written against 0-based arrays. Comment lines beginning with a percent sign appear before the size line, where the format requires them.
A symmetric pattern matrix carrying only positions and only the lower triangle, so nine stored lines describe sixteen non-zeros. Both halves of that — no value column and an implied mirror — are places a reader can silently disagree with the file.
A valid Matrix Market file describing a 6x6 matrix with no stored entries, which is an all-zero matrix rather than an error. It separates readers that model an empty result from readers that treat a missing entry block as truncation.
A 2020-12 JSON Schema for the absorbance rows, carrying the physical unit and coverage factor as custom annotation keywords beside the standard type and range constraints. Annotations like these are ignored by validators and read by humans and tooling, which is precisely the contract being tested.
The six NaN bit patterns broken into sign bit, quiet bit and payload so the difference between a quiet and a signalling NaN is a column rather than folklore. It is the reference for testing whether a serializer preserves NaN payloads or normalises them all to one canonical pattern.
Six distinct NaN bit patterns — quiet, signalling, signed and payload-carrying — stored as float64. They are all NaN and none equals another, so any code that deduplicates, sorts or hashes them by value is doing something undefined and this array proves it.
One variable that hides absent data three different ways at once: two _FillValue cells, one legacy missing_value cell and one raw NaN, with valid_min and valid_max also declared. Averaging the raw array without honouring all three gives roughly -60 K instead of a real mean.
The same 4x5x6 temperature grid written in the 64-bit offset variant of NetCDF-3, whose only visible difference from the classic file is the fourth magic byte and the width of the header offsets. Use it to check that format sniffing looks at the version byte instead of stopping at the 'CDF' signature.
A NetCDF-3 classic file holding a 4x5x6 air-temperature field with full CF-1.8 coordinate metadata: standard names, axis attributes, units on every variable, and a declared _FillValue. Use it to check that a reader attaches units and coordinates to the array rather than returning bare numbers.
Four physical quantities on one shared (time, station) grid, each with its own CF units string including the dimensionless '1' and the UDUNITS 'm s-1' spelling. It is the fixture for a unit-aware layer that has to keep four different unit systems straight in a single file.
The same temperature field stored as int16 and recovered through the CF packing attributes scale_factor and add_offset. This is the failure that parses cleanly and is silently wrong: a reader that ignores the attributes returns values around -2000 instead of 260-292 K.
Three CF time coordinates over the same six records: hours from the Unix epoch, days from 2026-01-01, and the same days under a 365_day calendar. A decoder that assumes one epoch, or ignores the calendar attribute, dates the records wrongly without raising anything.
A NetCDF-3 file with a genuine unlimited record dimension, so the record variable is stored interleaved rather than contiguously while the fixed-size station variable is not. It exercises the striding path that a reader written only against fixed dimensions gets wrong.
A six-taxon Newick tree with branch lengths on every edge and bootstrap-style support values written as internal node labels. That support notation is ambiguous by design in Newick — the same position can hold a node name — which is precisely what a parser has to decide about.
The identical tree topology with every branch length and support value stripped, which is what a Newick writer produces when the source tree has no lengths. Comparing the two proves a parser distinguishes 'length zero' from 'no length recorded'.
A 6x10 boolean mask stored as one byte per element, because .npy does not bit-pack booleans however tempting that assumption is. The True cells follow a simple divisible-by-three rule so a mis-decode is obvious rather than plausible.
A 4x6 matrix of 0..23 stored row-major, so the byte sequence begins with the first row. It is one half of a memory-order pair that is indistinguishable from its twin unless the header's fortran_order flag is honoured.
The real FFT of a 64-sample sine stored as complex128, where each element is an interleaved pair of doubles rather than two separate planes. Readers without a complex type usually flatten it, which doubles the reported length and shifts the peak bin.
Five timestamps stored as datetime64 with second resolution, where the unit is part of the dtype descriptor and the values on disk are plain int64 epoch seconds. Dropping the unit turns 2026-01-01 into the integer 1767225600 without any complaint.
We use Google Analytics and show ads via Adsterra. Non-essential cookies and ad scripts run only after you allow the matching categories. See our cookie policy.
