TidyPanel

After installing the dependencies, run this in an interactive R session and select DESCRIPTION in the extracted package folder:

install.packages(dirname(file.choose()), repos = NULL, type = "source")

install.packages("TidyPanel_0.2.133.tar.gz", repos = NULL, type = "source")

A built source archive also includes the example guide:

vignette("TidyPanel_introduction", package = "TidyPanel")

To locate the launcher after installation, without relying on the source folder:

system.file("exec", "tidypanel.R", package = "TidyPanel", mustWork = TRUE)

Pass that returned path to Rscript --vanilla. R callers can alternatively use TidyPanel::tidypanel_cli(); it returns an exit status without terminating R. The bundled launcher exits with that status.

Requests are limited to 65,536 bytes. Only options listed by describe are accepted. na_strings: [] declares no caller missing tokens. sheet selects one name or positive index, not "ALL". auto_pivot defaults to false. preview_rows defaults to 0; preview_columns defaults to 8 and preview_chars to 80. Preview cells are text, with null for missing positions; their column indices map to the complete schema in the metadata artifact.

The output directory must be new, have an existing parent and stay outside the installed package and launcher package folders. Existing files are never overwritten by the job interface. A failed write may leave a partial new directory; retain it for inspection and choose a new directory for a retry. The interface inherits the caller's filesystem permissions and is not a sandbox.

The RDS artifact retains the full cleaned R table and audit for a compatible trusted R environment. CSV is a convenience export with empty fields for missing values; it cannot preserve every type or distinguish every literal empty string from missingness. Consult column metadata before cross-language reading. The audit records supported operations and may be empty; it is not a complete cell-by-cell history. Preserve the source and review the result before analysis. This interface adds no new cleaning rules, unit inference or analysis model, and sends no files to an AI service.

Automatic inference: infer_data_types() is a separate, heuristic operation. It can convert categorical text to logical values and turn nonmatching values into missing values when a threshold is met. A successful run or a stricter threshold is not proof that category meanings were preserved. Use recognized _code or _id names for identifier fields, retain the source, and review output types and missing counts. The Agent job interface uses read_messy_panel(); it does not apply an extra standalone inference pass.

Stable categorical labels in headerless numeric tables and paired free/total measurements are retained. For numeric text reports, provide the observation section as a plain table or data frame rather than including report metadata. Commented delimited headers retain their labels, including whitespace after the comment marker and quoted separators in field names or records. Explicit TAB records remain tab-delimited when labels contain spaces. Explicit two-column field,value tables retain every uniquely named record, including blank trailing values and long text. Hash characters inside delimited fields are treated as data rather than silently truncating the field. Explicit identifier suffixes remain distinct from financial measurement labels. Fields named exactly site preserve literal identifiers, including leading zeros. Fields named ident or ending in _ident retain text codes without filling missing positions. A qualified iso_country column retains literal NA only with at least two distinct other uppercase two-letter code observations. Coded reference catalogs require a separately prepared named observation table. Sectioned observation exports must be supplied as separately prepared named tables. Prefaced measurement exports require a separately prepared named observation table. Prefaced monthly series exports require a separately prepared named observation table. Qualified numeric total-uncertainty measurements are retained, while ordinary subtotal columns remain removable. Related measurement fields such as site_count remain eligible for numeric cleaning. Sampled averages retain their values when time and integer observation-count fields establish a measurement context. Sources with delimited metadata preambles must be supplied as plain observation tables. Headerless whitespace records with quoted labels must be converted to CSV or supplied as an explicitly named data frame; unsupported mixed separators fail clearly rather than silently combining fields. Numeric-context NaN placeholders become missing values without changing pure text labels. For scientific exports with a separate unit descriptor row, remove that row before supplying a named observation table. Customize na_strings when literal codes overlap with default missing markers. Numeric missing markers can also overlap with genuine measurements. Choose na_strings from the source's definitions; retain numeric markers that the source uses as observations by excluding them from that list. Hyphenated English month dates are converted only when every observed date is valid, independently of the time locale. Explicit date identifier fields remain text. Qualified population income measures retain their labels. Sparse country-year records remain present even when their measurements are missing. Explicitly qualified total-energy measurements are not treated as subtotal columns. Missing values in explicitly labelled calendar fields are not filled from neighboring records. Lowercase na is recognized in numeric contexts while pure text labels and recognized identifier fields remain literal. Exclude it from a custom na_strings when the marker has a different meaning in your table.

Explicit mixed-case ID suffixes remain identifier text and use _id in cleaned names. Sparse digit identifiers are not treated as prose footnotes. Type inference retains literal Trace observations unless the caller explicitly includes them in na_strings; it does not assign an invented exact measurement. Clock-bearing values and date suffixes remain text during type inference; fractional serial days are not reduced to dates. Explicit flag fields retain their literal tokens. Unit-qualified total rain, snow, and precipitation fields and their corresponding flags are retained as measurements. Slash-dated series exports require a separately prepared named observation table; unsupported metadata layouts fail clearly instead of merging labels. In both the reader and standalone inference, compact mixed digit/letter fields with leading-zero evidence stay text rather than losing letters or padding. This lexical protection does not validate code meanings or arbitrary punctuation. Existing compact-month calendar handling keeps precedence over this code guard. Scientific notation and formatted decimal amounts retain numeric conversion. Unsupported name-first reference exports require a named table with an identifier column first. Sectioned control-record reports require a separately prepared named observation table; the simplified public reader does not combine their sections.

Large-table subtotal checks retain the existing row-selection rules while avoiding repeated per-record matching.

Row detection classifies bounded chunks rather than retaining several full-table logical masks. Auxiliary header counters are also limited to the current row chunk. This reduces temporary storage without changing supported cleaning rules. The source table and cleaned output still need to fit in memory; this is not an out-of-core reader.