A container can depend on other objects
HDF5 soft links point to paths, external links can refer to another file, and virtual datasets can assemble values from multiple sources. This local tool does not fetch those dependencies or treat missing values as a successful extraction. It records the object and explains why its content is not exposed as an ordinary dataset slice.
Hard-linked group structures can also contain cycles. Hierarchy traversal has explicit depth and entry limits to avoid unbounded inspection. A bounded inventory is a practical browser workflow, not a replacement for a complete archive-management system. Use the source application for structures that exceed the documented bounds.
Compression and datatypes affect feasibility
Built-in HDF5 filters cover common compressed datasets, but optional filter plugins are not installed dynamically. A dataset requiring another plugin is marked accordingly. Large uncompressed chunks can require substantial memory even for a small logical slice, so the workbench also limits individual chunk size.
Complex datatypes need equally explicit treatment. Primitive integers, floats, strings and flat fixed-width compound fields are supported for extraction. References, opaque values, enums, nested/variable-length compound fields and variable-length arrays remain metadata-only. This avoids turning an unfamiliar byte sequence into a misleading table of guessed values.
Read omissions as part of the result
CSV, NPY and PNG cannot reproduce the complete HDF5 object graph. Attributes, compression, links and dimension-scale relationships are not silently embedded in those outputs. The manifest and slice report retain metadata and bounded attribute values, with reasons when a value is omitted or unreadable.
Keep these reports next to the converted files and retain the original source. If a downstream process needs exact application semantics, validate that process with the producing software or a domain-specific reader. The desk provides inspectable local extraction within a clear format matrix; it does not certify the scientific meaning or completeness of an external dataset collection.