The AI assistant manual that ships with this package

tflspec is too new to be in any chat model's training data: asked for a table specification or the program made from one, an assistant invents sheet columns and function names and flags neither as a guess. The fix is to give it the facts first, and this file is those facts – the specification formats (ARD, table, report, listing, figure design), the functions that turn them into R code, what that code looks like, the former names that are refused, and the complete export list – sized to sit in one chat session.

Usage

tflspec_ai_manual(
  dir = NULL,
  overwrite = FALSE,
  define = NULL,
  adam_spec = NULL,
  datasets = NULL
)

Arguments

dir: Optional folder. When given, the kit is written there (the folder is made if need be) and the paths of its files are returned invisibly; the files, their sizes and what to attach are printed.
overwrite: Write over a kit already in dir. Default FALSE.
define, adam_spec, datasets: With dir: the study's ADaM metadata, from its define.xml, its ADaM spec workbook or the datasets' column information (see tfl_adam_metadata()); written as 07_adam_metadata.csv.

Value

The path to the manual when dir is NULL; otherwise the kit's files, invisibly.

Details

tflspec_ai_manual() returns the path to the copy installed with this package, so the manual you attach always describes the version you actually have. For the rendering side attach rtfreporter's own manual instead (rtfreporter::rtfreporter_ai_manual()).

With dir, it writes everything a chat needs to draft a study's specs from the company's TOC and shells into one flat folder – seven files (eight with the ADaM metadata), under the ten files most chats take:

00_START_HERE.md – the request to paste, and what to attach;
01_tflspec-ai-user-manual.md – this manual;
02_tflspec-ai-toc-shell-guide.md – how to read a TOC and shells into the spec files, the checklist the assistant runs, the assumptions it must list;
03_spec_schema.yml – the installed inst/spec/schema.yml: every function a spec item calls, its arguments and their types (tfl_schema()), and the spec sheets' columns (tfl_spec_columns());
04_blank_spec.md – every file of a study's YAML spec, blank, its keys said in comments, in the form of the answer;
05_example_toc_shell.xlsx, 06_example_spec.md – a made-up study end to end: its TOC and four shells (a demographics table, an AE table by SOC and PT, a listing of serious AEs, a figure of mean blood pressure), and the spec files written from them, with the spec/_assumptions.yml the guide asks for. The example is written now through the one-time conversion (tfl_yaml_from_sheets()), so it is this version's.

The assistant answers with ONE text holding every file, each a heading with its path and a yaml block; tfl_ai_unpack() writes it into a study folder and tfl_ai_review() checks the folder.

With define, adam_spec or datasets it also writes 07_adam_metadata.csv, the study's ADaM metadata (tfl_adam_metadata(): one row a variable, its label, kind and levels; never the data), and the request tells the assistant to take every dataset, variable and level from it. Give one; with several, the first of define, adam_spec, datasets is used.

Examples