Continue Long Medical History Terms Between Words

1library(yamaa)23yamaa_spec("spec.yaml")$run()$output

1import yamaa23yamaa.spec("spec.yaml").run().output

1# Reference solution for the yamaa benchmark sdtm-suppmh-long-terms (R track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67suppressPackageStartupMessages({8  library(dplyr)9  library(readr)10})1112mh <- read_csv("/app/input/mh_raw.csv", col_types = cols(.default = col_character()),13               na = character(), trim_ws = FALSE)1415split_term <- function(x, width = 200) {16  pieces <- character()17  rest <- x18  while (nchar(rest) > width) {19    head <- substr(rest, 1, width + 1)20    sp <- gregexpr(" ", head, fixed = TRUE)[[1]]21    sp <- sp[sp > 1]22    if (length(sp) == 0) {23      # no break point: a single word longer than width24      pieces <- c(pieces, substr(rest, 1, width))25      rest <- substr(rest, width + 1, nchar(rest))26    } else {27      pos <- max(sp)28      pieces <- c(pieces, substr(rest, 1, pos - 1))29      rest <- substr(rest, pos + 1, nchar(rest))30    }31  }32  c(pieces, rest)33}3435labels <- c(MHTERM1 = "Reported Term for the Medical History 1",36            MHTERM2 = "Reported Term for the Medical History 2")3738rows <- lapply(seq_len(nrow(mh)), function(i) {39  p <- split_term(mh$MHTERM[i])40  if (length(p) < 2) return(NULL)41  extra <- p[-1]42  qnam <- paste0("MHTERM", seq_along(extra))43  tibble(44    STUDYID = mh$STUDYID[i],45    RDOMAIN = "MH",46    USUBJID = mh$USUBJID[i],47    IDVAR = "MHSEQ",48    IDVARVAL = as.character(as.integer(mh$MHSEQ[i])),49    QNAM = qnam,50    QLABEL = unname(labels[qnam]),51    QVAL = extra,52    QORIG = "Collected",53    QEVAL = NA_character_54  )55})5657suppmh <- bind_rows(rows)58if (nrow(suppmh) == 0) {59  suppmh <- tibble(STUDYID = character(), RDOMAIN = character(), USUBJID = character(),60                   IDVAR = character(), IDVARVAL = character(), QNAM = character(),61                   QLABEL = character(), QVAL = character(), QORIG = character(),62                   QEVAL = character())63}64suppmh <- suppmh %>%65  arrange(STUDYID, USUBJID, as.integer(IDVARVAL), QNAM) %>%66  select(STUDYID, RDOMAIN, USUBJID, IDVAR, IDVARVAL, QNAM, QLABEL, QVAL, QORIG, QEVAL)6768write_csv(suppmh, "/app/output/suppmh.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-suppmh-long-terms (Python track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67"""Build SUPPMH from MH_RAW: continuation pieces of MHTERM beyond 200 characters."""8import os910import pandas as pd1112INPUT = "/app/input"13OUTPUT = "/app/output"14LIMIT = 20015COLUMNS = ["STUDYID", "RDOMAIN", "USUBJID", "IDVAR", "IDVARVAL",16           "QNAM", "QLABEL", "QVAL", "QORIG", "QEVAL"]171819def split_term(text, limit=LIMIT):20    """Split text into pieces of at most `limit` characters at word boundaries.2122    The space at each break belongs to neither piece; a word ending exactly23    at the limit fits.24    """25    pieces = []26    rest = text27    while len(rest) > limit:28        idx = rest.rfind(" ", 0, limit + 1)29        if idx <= 0:30            # No whole word fits: fall back to a hard break.31            pieces.append(rest[:limit])32            rest = rest[limit:]33            continue34        pieces.append(rest[:idx])35        rest = rest[idx + 1:]36    pieces.append(rest)37    return pieces383940def main():41    mh = pd.read_csv(os.path.join(INPUT, "mh_raw.csv"), dtype=str,42                     keep_default_na=False)43    rows = []44    for rec in mh.itertuples(index=False):45        pieces = split_term(rec.MHTERM)46        seq = str(int(float(rec.MHSEQ))) if rec.MHSEQ.strip() else ""47        for n, piece in enumerate(pieces[1:], start=1):48            rows.append({49                "STUDYID": rec.STUDYID,50                "RDOMAIN": "MH",51                "USUBJID": rec.USUBJID,52                "IDVAR": "MHSEQ",53                "IDVARVAL": seq,54                "QNAM": f"MHTERM{n}",55                "QLABEL": f"Reported Term for the Medical History {n}",56                "QVAL": piece,57                "QORIG": "Collected",58                "QEVAL": "",59            })60    out = pd.DataFrame(rows, columns=COLUMNS)61    os.makedirs(OUTPUT, exist_ok=True)62    out.to_csv(os.path.join(OUTPUT, "suppmh.csv"), index=False)636465if __name__ == "__main__":66    main()

1/* opensas reference solution for the yamaa benchmark sdtm-suppmh-long-terms.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45/* A quoted term can hold commas; DSD reads it as one field. */6data mh_raw;7  length STUDYID USUBJID MHTERM $1024;8  infile "/app/input/mh_raw.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;9  input STUDYID : $char1024. USUBJID : $char1024. MHSEQ MHTERM : $char1024.;10run;1112/* The term is cut into pieces of at most 200 characters, each ending after13   the last word that ends within its first 200 characters (a word ends14   where a non-blank is followed by a blank); the blanks at the break belong15   to neither piece. The first piece stays on the MH record, and the second16   and third each give one SUPPMH record. */17data result;18  set mh_raw;19  length RDOMAIN IDVAR IDVARVAL QNAM QLABEL QVAL QORIG QEVAL REST PIECE $1024;20  RDOMAIN = 'MH';21  IDVAR = 'MHSEQ';22  IDVARVAL = strip(put(MHSEQ, best32.));23  QORIG = 'Collected';24  QEVAL = '';25  REST = MHTERM;26  do P = 1 to 3;27    LEN = lengthn(REST);28    PIECE = '';29    if LEN > 0 then do;30      if LEN <= 200 then do;31        PIECE = REST;32        REST = '';33      end;34      else do;35        BRK = 0;36        do K = 1 to 200;37          if substr(REST, K, 1) ne ' ' and substr(REST, K + 1, 1) = ' ' then BRK = K;38        end;39        if BRK = 0 then do;40          PIECE = substr(REST, 1, 200);41          NXT = 201;42        end;43        else do;44          PIECE = substr(REST, 1, BRK);45          NXT = BRK + 1;46          SKIP = 1;47          do K = BRK + 1 to LEN;48            if SKIP = 1 and substr(REST, K, 1) = ' ' then NXT = K + 1;49            else SKIP = 0;50          end;51        end;52        if NXT > LEN then REST = '';53        else REST = substr(REST, NXT);54      end;55    end;56    if P = 2 and LEN > 0 then do;57      QNAM = 'MHTERM1';58      QLABEL = 'Reported Term for the Medical History 1';59      QVAL = PIECE;60      output;61    end;62    if P = 3 and LEN > 0 then do;63      QNAM = 'MHTERM2';64      QLABEL = 'Reported Term for the Medical History 2';65      QVAL = PIECE;66      output;67    end;68  end;69run;7071proc sql;72  create table final as73  select STUDYID, RDOMAIN, USUBJID, IDVAR, IDVARVAL, QNAM, QLABEL, QVAL, QORIG, QEVAL74  from result;75quit;76proc export data=final outfile="/app/output/suppmh.csv" dbms=csv replace; run;