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 2 of 6; 24 results per page.
A FITS binary table with four typed columns, big-endian numerics packed with no inter-field padding and TUNIT declaring milli-jansky on the two flux columns. It is the fixture for a table reader that has to build a row struct from TFORM codes rather than guess widths.
Integer FITS images mark undefined pixels with the BLANK keyword, and BLANK is compared against the stored value before BZERO and BSCALE are applied. Scale first and the four undefined pixels turn into a perfectly plausible zero, which is the ordering bug this file exists to expose.
A completely valid FITS file consisting of one 2880-byte header block and no data unit, which the standard permits whenever NAXIS is 0. It separates readers that model the data array as optional from readers that treat 'no pixels' as corruption.
FITS has no unsigned 16-bit type, so unsigned data is stored signed with BZERO = 32768 and recovered as BZERO + BSCALE * stored. Every stored value in this image is negative, so a reader that ignores the scaling returns an entirely negative frame while parsing the file perfectly.
The baseline rung of the BITPIX ladder: a 24x16 synthetic star field stored as unsigned bytes, the one integer width FITS holds without a BZERO shift. Every image in this family carries the same scene so a decoder can be scored across data types.
Single-precision FITS pixels including two NaNs, which is the only way the standard marks undefined data for floating BITPIX, plus one +Inf. Statistics computed without NaN-aware reductions come back as NaN for the entire frame.
The top rung of the BITPIX ladder, carrying the identical synthetic scene as IEEE-754 doubles with DATAMIN and DATAMAX declared. It is the reference against which the 8-, 16- and 32-bit members of the family are scored for precision loss.
A header of 47 cards, so END lands well past the first 2880-byte block, alongside repeated HISTORY and COMMENT cards that legally share a keyword. A parser that reads one block and stops never sees END and truncates the metadata.
A three-HDU FITS file whose primary unit is header-only (NAXIS = 0) with the real data in named SCI and MASK extensions. A reader that only ever looks at HDU 0 reports an empty file, which is exactly the bug this fixture is built to catch.
Every header card of the tangent-plane WCS file transcribed to keyword, value and comment columns. Diff a header parser's output against it to prove the parser split each 80-column card at the right places instead of guessing on whitespace.
A 32-bit integer image carrying a complete tangent-plane WCS: 1-based CRPIX reference pixels, a deliberately negative CDELT1 so right ascension runs the other way, and an ICRS frame. It targets the two mistakes every WCS implementation makes first, the 1-based origin and the flipped axis.
Every element of the float64 .npy array written as its shortest round-tripping decimal representation, one row per element in C order. Parsing this text back to float64 reproduces the binary array bit for bit, so it is a usable oracle rather than a lossy printout.
A ten-step ladder from the smallest representable double up across the subnormal boundary into normal range, showing precision degrading to a single significant bit. Any pipeline running with flush-to-zero enabled turns the first eight entries into plain zeros without reporting anything.
Each rung of the subnormal ladder with its bit pattern, decimal value, C99 hex float and remaining significant bits, plus the value a flush-to-zero pipeline would produce instead. It makes gradual underflow a table you can diff rather than a behaviour you have to infer.
The precision-loss cases as a two-column float64 array holding the original value beside its float32 round trip, both kept at double width so the difference itself is representable. Subtracting the columns gives the error directly without a second narrowing step.
Twelve doubles narrowed to single precision with the absolute, relative and ULP error of each, plus whether the value survived unchanged. It covers the three distinct ways narrowing fails: rounding, overflow to infinity at 1e39, and demotion to a subnormal at 1e-40.
The same 1002 numbers added five different ways, giving results from 0 to 1000 for an exact answer of 1000. Floating-point addition is not associative, and this file is the smallest complete demonstration of what that costs a naive accumulator.
Every interval written in both coordinate conventions with its length, which is identical either way. It is the lookup table for the most common bioinformatics off-by-one, and the length column is what proves a conversion did not quietly change an interval's size.
Seven features on a synthetic contig — two genes, an mRNA, two exons and two CDS records — linked by ID and Parent attributes across both strands. GFF3 coordinates are 1-based and inclusive, which is the fact its BED twin exists to contrast.
Ten weighings as tab-separated values with headers that contain spaces, brackets and parentheses — 'u(mass) [g]' — the way balance software actually writes them. Column names like these break naive slug-based header handling long before the numbers become a problem.
Every value from the CF-1.8 NetCDF grid flattened to one row per cell, with the axis values spelled out in full. Diff a NetCDF-to-table export against this file to prove the reader walked the dimensions in (time, lat, lon) order and did not silently transpose the array.
A five-source uncertainty budget laid out the way the GUM prescribes: each source with its distribution, its divisor, its sensitivity coefficient and its contribution, then the combined and expanded values. Rectangular sources are divided by the square root of three, which is the step budget spreadsheets most often get wrong.
One array stored twice in the same file: chunked with the shuffle and gzip filters applied, and contiguous with no filters at all. Comparing the two proves a reader ran the HDF5 filter pipeline rather than reading raw chunk bytes.
A compound-dtype dataset mixing an integer key, two float measurements, a fixed-width byte string and a boolean flag in a single record. It is the fixture for readers that must map an HDF5 record type onto a native struct without reordering or repadding the fields.
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.
