Subscale Scoring

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

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

1# Reference solution for the yamaa benchmark adam-adqs-subscale-score (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)910qs <- read_csv(11  "/app/input/qs.csv",12  col_types = cols(13    QSSEQ = col_integer(),14    QSSTRESN = col_double(),15    .default = col_character()16  )17)1819# Each physical functioning item response, on its zero to four scale.20items <- qs |>21  filter(QSTESTCD %in% c("PF01", "PF02", "PF03", "PF04")) |>22  mutate(23    PARAMCD = QSTESTCD, PARAM = QSTEST, AVAL = QSSTRESN, AREASND = QSREASND24  ) |>25  select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL, AREASND)2627# One score record per visit with a PF01 item record: the mean of the28# answered items on a zero to one hundred scale, with no value when fewer29# than three of the four items were answered. A score without a value takes30# the items' reason when none was answered, and reads NOT CALCULABLE31# otherwise.32visits <- items |>33  filter(PARAMCD %in% "PF01") |>34  distinct(STUDYID, USUBJID, VISIT)3536stats <- items |>37  group_by(STUDYID, USUBJID, VISIT) |>38  summarise(39    n_answered = sum(!is.na(AVAL)),40    mean_answered = mean(AVAL, na.rm = TRUE),41    n_reasons = n_distinct(AREASND, na.rm = TRUE),42    reason = first(na.omit(AREASND)),43    .groups = "drop"44  ) |>45  mutate(46    mean_answered = if_else(47      n_answered < 3 | is.nan(mean_answered),48      NA_real_,49      mean_answered * 2550    ),51    reason = case_when(52      n_answered >= 3 ~ NA_character_,53      n_answered == 0 & n_reasons == 1 ~ reason,54      .default = "NOT CALCULABLE"55    )56  )5758scores <- visits |>59  left_join(stats, by = c("STUDYID", "USUBJID", "VISIT")) |>60  mutate(61    PARAMCD = "PFSCORE",62    PARAM = "Physical Functioning Subscale Score",63    AVAL = mean_answered,64    AREASND = reason65  ) |>66  select(STUDYID, USUBJID, VISIT, PARAMCD, PARAM, AVAL, AREASND)6768adqs <- bind_rows(items, scores) |>69  arrange(STUDYID, USUBJID, VISIT, PARAMCD)7071dir.create("/app/output", showWarnings = FALSE)72write_csv(adqs, "/app/output/adqs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adqs-subscale-score (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 csv8from collections import OrderedDict910INPUT_QS = "/app/input/qs.csv"11OUTPUT_ADQS = "/app/output/adqs.csv"1213ITEM_CODES = ("PF01", "PF02", "PF03", "PF04")14SCORE_PARAMCD = "PFSCORE"15SCORE_PARAM = "Physical Functioning Subscale Score"161718def fmt(value):19    if value is None:20        return ""21    if value == int(value):22        return str(int(value))23    return repr(value)242526def main():27    with open(INPUT_QS, newline="") as f:28        rows = list(csv.DictReader(f))2930    # Item records: the physical functioning records (PF01..PF04).31    items = [r for r in rows if r["QSTESTCD"] in ITEM_CODES]3233    # Group item records by subject and visit, preserving input order.34    groups = OrderedDict()35    for r in items:36        key = (r["STUDYID"], r["USUBJID"], r["VISIT"])37        groups.setdefault(key, []).append(r)3839    out = []40    for (studyid, usubjid, visit), recs in groups.items():41        answered = []42        reasons = set()43        for r in recs:44            raw = (r["QSSTRESN"] or "").strip()45            aval = float(raw) if raw != "" else None46            if aval is not None:47                answered.append(aval)48            reason = (r["QSREASND"] or "").strip()49            if reason:50                reasons.add(reason)51            out.append(52                [studyid, usubjid, visit, r["QSTESTCD"], r["QSTEST"], fmt(aval), reason]53            )54        # Every visit with a PF01 item record also has a score record. A score55        # without a value takes the items' reason when none was answered, and56        # reads NOT CALCULABLE otherwise.57        if any(r["QSTESTCD"] == "PF01" for r in recs):58            if len(answered) >= 3:59                score = sum(answered) / len(answered) * 2560                areasnd = ""61            elif not answered and len(reasons) == 1:62                score = None63                areasnd = next(iter(reasons))64            else:65                score = None66                areasnd = "NOT CALCULABLE"67            out.append(68                [69                    studyid,70                    usubjid,71                    visit,72                    SCORE_PARAMCD,73                    SCORE_PARAM,74                    fmt(score),75                    areasnd,76                ]77            )7879    with open(OUTPUT_ADQS, "w", newline="") as f:80        writer = csv.writer(f)81        writer.writerow(82            ["STUDYID", "USUBJID", "VISIT", "PARAMCD", "PARAM", "AVAL", "AREASND"]83        )84        writer.writerows(out)858687if __name__ == "__main__":88    main()

1/* opensas reference solution for the yamaa benchmark adam-adqs-subscale-score.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data qs;6  length STUDYID USUBJID VISIT QSCAT QSTESTCD QSTEST QSSTAT QSREASND $1024;7  infile "/app/input/qs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. VISIT : $char1024. QSSEQ QSCAT : $char1024. QSTESTCD : $char1024. QSTEST : $char1024. QSSTRESN QSSTAT : $char1024. QSREASND : $char1024.;9run;1011proc sql;12  create table collected as select STUDYID,USUBJID,VISIT,QSTESTCD as PARAMCD,QSTEST as PARAM,QSSTRESN as AVAL,QSREASND as AREASND from qs where QSTESTCD in ('PF01','PF02','PF03','PF04');13  create table anchors as select distinct STUDYID,USUBJID,VISIT from collected where PARAMCD='PF01';14  create table scores as select a.STUDYID,a.USUBJID,a.VISIT,'PFSCORE' as PARAMCD,'Physical Functioning Subscale Score' as PARAM,15    case when count(b.AVAL)>=3 then 25*mean(b.AVAL) else . end as AVAL,16    case when count(b.AVAL)>=3 then ' ' when count(b.AVAL)=0 and count(distinct b.AREASND)=1 then max(b.AREASND) else 'NOT CALCULABLE' end as AREASND length=102417    from anchors as a left join collected as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.VISIT=b.VISIT group by a.STUDYID,a.USUBJID,a.VISIT;18quit;19data result; set collected scores; run;2021proc export data=result(keep=STUDYID USUBJID VISIT PARAMCD PARAM AVAL AREASND) outfile="/app/output/adqs.csv" dbms=csv replace; run;