One user turn for the same chat: the problems of an answer as a table, the schema of the parts they are in (only those: the chat already has the first prompt), and the request to return the whole block again, corrected. Paste it as the next message of the chat; in API mode it is the next user message of the conversation.
Arguments
- answer
A
tfl_ai_answer(tfl_ai_parse()).- problems
The problems to send: by default those of
tfl_ai_check()withcontext.- context
The
tfl_ai_context()of the prompt, for the checks.
Examples
answer <- c("```yaml",
"tflspec_ai: {task: toc, version: 1}",
"toc:",
"- {output_id: T-14-1-1, title: Demographics, colour: red}",
"```")
tfl_ai_repair_prompt(tfl_ai_parse(answer, "toc"))
#> Your answer had these problems. An error must be fixed; a warning is a name or value worth checking again.
#>
#> | severity | where | problem |
#> |---|---|---|
#> | error | toc row 1 field colour | is not a column of toc; its columns are output_id, type, title, population, footnote, program, file, note, section, datasets, label |
#> | warning | toc row 1 field type | blank is not table, figure or listing; read as table |
#>
#> The schema of toc:
#>
#> `toc` is a list of rows, one a report. Its columns:
#>
#> - `output_id`: The report's number as the documents give it, unique in the TOC (required). Example: `T-14-1-1`
#> - `type`: The kind of report: table, figure or listing. Example: `table`
#> - `title`: The title lines in order, without the report number and the analysis set; ` | ` between lines. Example: `Summary of Demographic and Baseline Characteristics`
#> - `population`: The analysis set the report is on, as the documents name it; it becomes the last title line. Example: `Safety Population`
#> - `footnote`: The footnote lines in order; ` | ` between lines. Example: `Percentages are based on the number of subjects in the analysis set.`
#> - `program`: The name of the program that makes the report. Example: `t_14_1_1.R`
#> - `file`: The name of the report's output file. Example: `t_14_1_1.rtf`
#> - `note`: A remark for people; never read by tflspec.
#> - `section`: The heading the report is listed under in the documents. Example: `14.1 Demographic and Baseline Data`
#> - `datasets`: The ADaM datasets the report reads; ` | ` between them. Example: `ADSL`
#> - `label`: The report's number as printed in its title. Example: `Table 14.1.1`
#>
#> Return the whole block again, corrected, with the same header. Keep everything that was right and change only what the problems name. When a problem asks for a name the context does not list, leave the item out and say so under `assumptions`.