Added glc_collection_plan() to create deterministic, serializable, metadata-only plans for collectable file-group units from canonical terms and matched, all, or selected declared variables, with optional stable dataset and file-group restrictions. Plans report stable unit ids, exclusions, required read/output columns, declared byte availability, and exact package/schema/revision provenance without reading measurement contents; glc_read() and glc_collect() remain the runtime authorities.
glc_explore() now uses the same declaration-compatibility engine as glc_collection_plan(), keeping Explorer selection, exclusions, device-safe partitioning, unit ids, and preferred-unit choice in parity with the public planner.
glcdp 1.0.0
Promoted schema 3.0.2 to the current default and primary stable import contract, while retaining 3.0.0 and 3.0.1 as compatible stable predecessors. Schema 3.0.2 is a corrective patch that requires datasheet_channel to contain at least one entry when present; its metadata-driven data import contract is unchanged. Schema 1.0.0 and 2.0.0 remain available as barebones legacy compatibility paths.
Declared variable types and factor levels in schema-declared order now drive import, including boolean true/false and 1/0 representations. Variable and factor-level descriptions are exposed in glc_variables(), and per-file encodings declared by a file group are applied to the corresponding files. glc_collect() rejects groups whose factor labels or level order differ.
Added a scheduled and manually dispatchable live schema 3.0.2 integration check against the validated IZTECH package. It covers exact Explorer inventories, metadata resources, factor and integer imports, selection, download, and local reopen behavior.
Metadata summaries now load tabular core resources such as participants.csv, and LightLogR standardization safely reconciles schema-declared source Id, Datetime, and file.name columns.
Collection-datetime groups can now be combined across participants when their formats agree; the participant-specific timestamp values are no longer mistaken for different datetime schemas in either glc_collect() or the Explorer.
Remote file metadata, full inventories, and package summaries are now cached for immutable revisions. Small Git blobs are inspected concurrently for LFS pointers, and glc_summary() counts variables directly from the normalized model instead of constructing the complete variable inventory.
glc_explore() now accepts IDE-provided Shiny viewer functions through shiny.launch.browser. A central, accessible status banner describes registry, package, summary, contents, selection, and preview loading work. Package contents can be loaded directly from the completed summary without leaving that view; completion offers a direct handoff to the loaded contents, and failures can be retried in place.
Explorer file groups can now be discovered by device, wearing position and type, modality, role, data state, contained source variables, and semantic terms in both Package contents and Select & hand off. These field filters narrow participant-specific repeated groups before exact group selection. The complete filtered result, rather than only its current page, can be transferred from Package contents into a preseeded handoff.
Numeric participant characteristics use inclusive range sliders. The metadata hierarchy loads complete records incrementally, keeps table paging independent, labels participant-characteristic records with participant and characteristic names, and presents singular flat child objects inline without hiding genuinely repeatable record collections. Repeated leaf fields, such as study-group inclusion, exclusion, and dataset lists, are folded with record counts; compact label/value grids remove the former wide spacing. Dense records show an in-place loading message and progress indicator while their values render. Metadata results are no longer truncated after 500 values.
Large Explorer inventories now remain responsive by paging File groups, Variables, and included handoff groups at 100 rows, abbreviating long selection-summary lists, deferring metadata hierarchy indexing until its view is opened, and serving large selectize choice sets from the Shiny session. A page-level busy pulse and output spinners provide immediate feedback during remaining reactive work.
Package contents now opens in the order Metadata, Datasets, Variables, and File groups. Metadata field labels stay on one aligned line, record nodes use a domain-neutral icon, and discovery sidebars omit the redundant Filters heading. File-group compatibility is summarized by a compact green or orange action above the filters; incompatible selections open focused refinement guidance instead of occupying the result card. The completed package-summary action opens Package contents directly on Metadata.
Explorer handoffs can now filter on schema semantic terms as well as source variable names. Variable filters automatically retain a collectable subset of file groups and report included and excluded counts instead of emitting one incompatibility error for every excluded group. The former Recommended action is now Primary, reflecting schema-declared primary variables and falling back to all variables when none are declared.
Select & hand off now starts with an explicit package-and-metadata or measurement-data choice. Metadata-only scripts pin the validated revision, download selected core resources, and load no measurement files. Data handoffs now use one full-width, seven-step wizard whose visible tabs cover package/metadata, file groups/datasets, participants/devices, variables/terms/rows, review, preview, and R export. Each step separates controls from a stable information column, fills the available window height, and uses equal-width Back/Continue navigation. Large dataset and exact-group selections scroll inside bounded inputs instead of stretching the card. Downstream filters temporarily narrow rather than erase the dataset and exact-file-group baseline selected in Step 2. Preview reads two files by default, with independent file and row limits, while an optional maximum rows per file is retained in the exported glc_read() call. Review now presents its selection and included-group tables as an accordion, with the selection open initially. Preview results sit directly below their build action, export scripts sit directly below their download action, and step content scrolls independently so Back/Continue controls remain visible. Generated data scripts clean up their temporary handoff settings, leaving only local_package and glc_data.
glcdp 0.9.3
Added extract_metadata() and add_metadata() to select requested metadata by imported-data identifiers and optionally join it onto every observation. Their default file_group_id link traverses from each file group to its dataset, participant, study, and device; dataset-level extraction remains available with by = "Id". Extracted summaries retain the input dataset’s grouping and grouping columns.
LightLogR-standardized glc_collect() output now exposes file_group_id and rejects contradictory group relationships or multiple device links within a dataset.
Added glc_explore(), a Shiny application for browsing the registry, inspecting package contents and hierarchical metadata, filtering datasets, file groups, participants, devices, and variables, previewing selections, and exporting annotated reproducible R scripts.
Added a documented Posit Connect deployment entry point and dependency manifest workflow.
glcdp 0.9.1
Corrected the package logo, removed its exterior background for transparent display, and refreshed the pkgdown logo and favicon assets.
glcdp 0.9.0
Initial package implementation for registry discovery, metadata inspection, selective downloads, Git LFS retrieval, and metadata-driven data import.
Added a pkgdown website and workflow vignettes covering discovery, metadata inspection, selective import, collection, and reproducible downloads.
glc_read() now reports file-level progress by default in interactive sessions; set progress = FALSE to suppress it.
Filtered glc_read() calls now use datetime source columns internally without retaining them unless the filters explicitly select them.
LightLogR-standardized glc_collect() output now maps dataset ids to Id, participant ids to participant_Id, and removes internal .glc_* columns.
Fixed remote loading of small Git blobs and metadata searches through nested data-frame fields, resolving the metadata examples in the vignettes.
glc_search_metadata() can now search metadata values, field paths, or both with the new search_in argument.
Unauthenticated ordinary Git file transfers now use immutable raw-content URLs, preventing per-file GitHub API rate limits during reads and downloads.
Summaries of local subsets now distinguish available from declared datasets, file groups, and files; glc_read(dataset_id = "all") skips files that are intentionally absent from a local subset and reports the discrepancy.
