Arm Assignment

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

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

1# Reference solution for the yamaa benchmark sdtm-dm-arms (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 1.67library(dplyr, warn.conflicts = FALSE)8library(readr)9library(stringr)1011rand <- read_csv(12  "/app/input/rand.csv",13  col_types = cols(.default = col_character())14)15ex <- read_csv(16  "/app/input/ex.csv",17  col_types = cols(.default = col_character())18)1920# The planned arm follows randomization; the actual arm follows exposure.21# PLACEBO was the PBO arm and VITAMIN D3 the TRT arm.22exposure <- ex |>23  filter(!is.na(EXTRT) & EXTRT != "") |>24  distinct(STUDYID, USUBJID, .keep_all = TRUE) |>25  select(STUDYID, USUBJID, EXTRT)2627dm <- rand |>28  select(STUDYID, USUBJID, RANDCD, RAND, SCRNFL) |>29  full_join(exposure, by = c("STUDYID", "USUBJID")) |>30  mutate(31    RANDCD = na_if(str_trim(RANDCD), ""),32    RAND = na_if(str_trim(RAND), ""),33    EXTRT = na_if(str_trim(EXTRT), ""),34    SCRNFL = coalesce(str_trim(SCRNFL), ""),35    # Never randomized leaves the planned arm empty; never treated or a36    # treatment matching no planned arm leaves the actual arm empty.37    ARMCD = if_else(is.na(RANDCD), NA_character_, RANDCD),38    ARM = if_else(is.na(RANDCD), NA_character_, RAND),39    ACTARMCD = case_when(40      EXTRT == "PLACEBO" ~ "PBO",41      EXTRT == "VITAMIN D3" ~ "TRT",42      .default = NA_character_43    ),44    ACTARM = case_when(45      EXTRT == "PLACEBO" ~ "Placebo",46      EXTRT == "VITAMIN D3" ~ "Vitamin D3",47      .default = NA_character_48    ),49    # A treated subject carries the other planned arm when that is what50    # the exposure says; the reason for a blank arm depends on why both51    # are blank, or on never being treated.52    ARMNRS = case_when(53      is.na(ARM) & is.na(ACTARM) & SCRNFL == "Y" ~ "SCREEN FAILURE",54      is.na(ARM) & is.na(ACTARM) ~ "NOT ASSIGNED",55      !is.na(ARM) & is.na(ACTARM) & is.na(EXTRT) ~ "NOT TREATED",56      .default = NA_character_57    ),58    ACTARMUD = if_else(59      !is.na(EXTRT) & is.na(ACTARM), EXTRT, NA_character_, missing = NA_character_60    ),61    DOMAIN = "DM"62  ) |>63  arrange(USUBJID) |>64  select(65    DOMAIN, STUDYID, USUBJID, ARMCD, ARM,66    ACTARMCD, ACTARM, ARMNRS, ACTARMUD67  )6869dir.create("/app/output", showWarnings = FALSE)70write_csv(dm, "/app/output/dm.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-dm-arms (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 1.67import csv8import os910INPUT_DIR = "/app/input"11OUTPUT_DIR = "/app/output"121314def read_rows(path):15    with open(path, newline="", encoding="utf-8") as f:16        return list(csv.DictReader(f))171819def clean(value):20    return (value or "").strip()212223def main():24    rand_rows = read_rows(os.path.join(INPUT_DIR, "rand.csv"))25    ex_rows = read_rows(os.path.join(INPUT_DIR, "ex.csv"))2627    # Planned arms as they appear in the randomization records, keyed by28    # upper-cased description and by upper-cased code.29    arms_by_desc = {}30    arms_by_code = {}31    for row in rand_rows:32        code = clean(row.get("RANDCD"))33        desc = clean(row.get("RAND"))34        if code and desc:35            arms_by_desc.setdefault(desc.upper(), (code, desc))36            arms_by_code.setdefault(code.upper(), (code, desc))3738    # One record per subject: randomization order first, then any39    # exposure-only subjects in exposure order.40    subjects = []41    seen = set()42    rand_by_subject = {}43    for row in rand_rows:44        key = (clean(row.get("STUDYID")), clean(row.get("USUBJID")))45        if key not in seen:46            seen.add(key)47            subjects.append(key)48        rand_by_subject.setdefault(key, row)4950    # Treatment received per subject (first exposure record).51    treatment = {}52    for row in ex_rows:53        key = (clean(row.get("STUDYID")), clean(row.get("USUBJID")))54        if key not in seen:55            seen.add(key)56            subjects.append(key)57        if key not in treatment:58            treatment[key] = clean(row.get("EXTRT"))5960    out_rows = []61    for studyid, usubjid in subjects:62        key = (studyid, usubjid)63        rand = rand_by_subject.get(key, {})64        armcd = clean(rand.get("RANDCD"))65        arm = clean(rand.get("RAND")) if armcd else ""66        scrnfl = clean(rand.get("SCRNFL"))6768        trt = treatment.get(key, "")69        actarmcd = ""70        actarm = ""71        actarmud = ""72        if trt:73            match = arms_by_desc.get(trt.upper()) or arms_by_code.get(74                trt.upper()75            )76            if match:77                actarmcd, actarm = match78            else:79                actarmud = trt8081        armnrs = ""82        if not armcd and not actarmcd and not trt:83            armnrs = "SCREEN FAILURE" if scrnfl == "Y" else "NOT ASSIGNED"84        elif armcd and not trt:85            armnrs = "NOT TREATED"8687        out_rows.append(88            {89                "DOMAIN": "DM",90                "STUDYID": studyid,91                "USUBJID": usubjid,92                "ARMCD": armcd,93                "ARM": arm,94                "ACTARMCD": actarmcd,95                "ACTARM": actarm,96                "ARMNRS": armnrs,97                "ACTARMUD": actarmud,98            }99        )100101    os.makedirs(OUTPUT_DIR, exist_ok=True)102    fieldnames = [103        "DOMAIN",104        "STUDYID",105        "USUBJID",106        "ARMCD",107        "ARM",108        "ACTARMCD",109        "ACTARM",110        "ARMNRS",111        "ACTARMUD",112    ]113    with open(114        os.path.join(OUTPUT_DIR, "dm.csv"), "w", newline="", encoding="utf-8"115    ) as f:116        writer = csv.DictWriter(f, fieldnames=fieldnames)117        writer.writeheader()118        writer.writerows(out_rows)119120121if __name__ == "__main__":122    main()

1/* opensas reference solution for the yamaa benchmark sdtm-dm-arms.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data ex;6  length STUDYID USUBJID EXTRT $1024;7  infile "/app/input/ex.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. EXTRT : $char1024.;9run;1011data rand;12  length STUDYID USUBJID RANDCD RAND SCRNFL $1024;13  infile "/app/input/rand.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. RANDCD : $char1024. RAND : $char1024. SCRNFL : $char1024.;15run;1617/* Keep the first nonmissing collected exposure per subject. */18data nonmissing_ex; set ex; if not missing(EXTRT); EXORDER=_N_; run;19proc sort data=nonmissing_ex; by STUDYID USUBJID EXORDER; run;20data exposure; set nonmissing_ex; by STUDYID USUBJID; if first.USUBJID; run;21proc sql; create table joined as select a.*,b.EXTRT from rand a left join exposure b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID; quit;22data result; set joined; length DOMAIN $2 ARMCD ARM ACTARMCD ACTARM ARMNRS ACTARMUD $1024; DOMAIN='DM'; ARMCD=RANDCD; ARM=RAND;23if EXTRT='PLACEBO' then do; ACTARMCD='PBO'; ACTARM='Placebo'; end;24else if EXTRT='VITAMIN D3' then do; ACTARMCD='TRT'; ACTARM='Vitamin D3'; end;25else if not missing(EXTRT) then ACTARMUD=EXTRT;26if missing(EXTRT) then do; if not missing(ARMCD) then ARMNRS='NOT TREATED'; else if SCRNFL='Y' then ARMNRS='SCREEN FAILURE'; else ARMNRS='NOT ASSIGNED'; end; run;27proc sql;28  create table final as select DOMAIN, STUDYID, USUBJID, ARMCD, ARM, ACTARMCD, ACTARM, ARMNRS, ACTARMUD from result;29quit;30proc export data=final outfile="/app/output/dm.csv" dbms=csv replace; run;