Prepare Response Assessments for Best Overall Response

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

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

1# Reference solution for the yamaa benchmark adam-adrs-response-prep (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)910raw <- read_csv(11  "/app/input/adrs_raw.csv",12  col_types = cols(13    ASEQ = col_integer(),14    ADT = col_date(),15    RANDDY = col_integer(),16    .default = col_character()17  )18)1920therapy <- read_csv(21  "/app/input/adsl.csv",22  col_types = cols(23    NTXSTDT = col_date(),24    .default = col_character()25  )26)2728raw <- raw |>29  left_join(therapy, by = c("STUDYID", "USUBJID"))3031# The response category each record can support; an assessment on or after32# the subject's new anti-cancer therapy start supports no category. Stable33# and neither-complete-nor-progressive disease count only on or after day34# 42, and earlier ones, or ones with no day, fall back to not evaluable.35based <- raw |>36  mutate(37    BORCAT = case_when(38      !is.na(NTXSTDT) & ADT >= NTXSTDT ~ NA_character_,39      AVALC %in% c("CR", "PR", "PD", "NE") ~ AVALC,40      AVALC %in% "SD" & !is.na(RANDDY) & RANDDY >= 42L ~ "SD",41      AVALC %in% "SD" ~ "NE",42      AVALC %in% "NON-CR/NON-PD" & !is.na(RANDDY) & RANDDY >= 42L ~ "NON-CR/NON-PD",43      AVALC %in% "NON-CR/NON-PD" ~ "NE",44      .default = NA_character_45    ),46    BORPRI = case_when(47      BORCAT == "CR" ~ 1L,48      BORCAT == "PR" ~ 2L,49      BORCAT == "SD" ~ 3L,50      BORCAT == "NON-CR/NON-PD" ~ 4L,51      BORCAT == "PD" ~ 5L,52      BORCAT == "NE" ~ 6L53    )54  )5556# Usable records numbered from 1 in priority, date, then sequence order;57# records supporting no category are passed over without a gap.58order <- based |>59  filter(!is.na(BORCAT)) |>60  arrange(STUDYID, USUBJID, BORPRI, ADT, ASEQ) |>61  group_by(STUDYID, USUBJID) |>62  mutate(BORSEQ = row_number()) |>63  ungroup() |>64  select(STUDYID, USUBJID, ASEQ, BORSEQ)6566adrs <- based |>67  left_join(order, by = c("STUDYID", "USUBJID", "ASEQ")) |>68  select(STUDYID, USUBJID, ASEQ, ADT, RANDDY, AVALC, BORCAT, BORPRI, BORSEQ) |>69  arrange(STUDYID, USUBJID, ASEQ)7071dir.create("/app/output", showWarnings = FALSE)72write_csv(adrs, "/app/output/adrs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adrs-response-prep (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 ADRS from ADRS_RAW and ADSL.89One output record per collected overall response assessment, with the10best-overall-response category (BORCAT), its priority (BORPRI), and the11per study-subject sequence (BORSEQ) of records that support a category.12"""1314import csv15from datetime import date16from pathlib import Path1718INPUT_DIR = Path("/app/input")19OUTPUT_DIR = Path("/app/output")2021# Collected response -> supported category, before the day-42 and22# new-therapy rules are applied.23CATEGORY_BY_RESPONSE = {24    "CR": "CR",25    "PR": "PR",26    "SD": "SD",27    "NON-CR/NON-PD": "NON-CR/NON-PD",28    "PD": "PD",29    "NE": "NE",30}3132PRIORITY_BY_CATEGORY = {33    "CR": 1,34    "PR": 2,35    "SD": 3,36    "NON-CR/NON-PD": 4,37    "PD": 5,38    "NE": 6,39}4041# Categories that only count on or after day 42 after randomization.42DAY_LIMITED = {"SD", "NON-CR/NON-PD"}43MIN_DAY = 42444546def parse_date(text):47    text = (text or "").strip()48    if not text:49        return None50    return date.fromisoformat(text)515253def parse_int(text):54    text = (text or "").strip()55    if not text:56        return None57    return int(text)585960def main():61    with open(INPUT_DIR / "adsl.csv", newline="", encoding="utf-8") as f:62        adsl_rows = list(csv.DictReader(f))63    new_therapy_start = {}64    for row in adsl_rows:65        key = (row["STUDYID"], row["USUBJID"])66        new_therapy_start[key] = parse_date(row.get("NTXSTDT"))6768    with open(INPUT_DIR / "adrs_raw.csv", newline="", encoding="utf-8") as f:69        raw_rows = list(csv.DictReader(f))7071    records = []72    for row in raw_rows:73        studyid = row["STUDYID"]74        usubjid = row["USUBJID"]75        adt = parse_date(row.get("ADT"))76        randdy = parse_int(row.get("RANDDY"))77        avalc = (row.get("AVALC") or "").strip()7879        borcat = CATEGORY_BY_RESPONSE.get(avalc)80        if borcat is not None:81            start = new_therapy_start.get((studyid, usubjid))82            if start is not None and adt is not None and adt >= start:83                # On or after new anti-cancer therapy: supports nothing.84                borcat = None85            elif borcat in DAY_LIMITED and (randdy is None or randdy < MIN_DAY):86                borcat = "NE"8788        borpri = PRIORITY_BY_CATEGORY[borcat] if borcat is not None else None89        records.append(90            {91                "STUDYID": studyid,92                "USUBJID": usubjid,93                "ASEQ": row["ASEQ"],94                "ADT": (row.get("ADT") or "").strip(),95                "RANDDY": (row.get("RANDDY") or "").strip(),96                "AVALC": (row.get("AVALC") or "").strip(),97                "BORCAT": borcat or "",98                "BORPRI": borpri,99                "_adt_sort": adt,100                "_aseq_sort": int(row["ASEQ"]),101                "BORSEQ": None,102            }103        )104105    groups = {}106    for rec in records:107        if rec["BORPRI"] is not None:108            groups.setdefault((rec["STUDYID"], rec["USUBJID"]), []).append(rec)109    for group in groups.values():110        ordered = sorted(111            group,112            key=lambda r: (113                r["BORPRI"],114                r["_adt_sort"] is None,115                r["_adt_sort"] or date.min,116                r["_aseq_sort"],117            ),118        )119        for number, rec in enumerate(ordered, start=1):120            rec["BORSEQ"] = number121122    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)123    fieldnames = [124        "STUDYID",125        "USUBJID",126        "ASEQ",127        "ADT",128        "RANDDY",129        "AVALC",130        "BORCAT",131        "BORPRI",132        "BORSEQ",133    ]134    with open(OUTPUT_DIR / "adrs.csv", "w", newline="", encoding="utf-8") as f:135        writer = csv.DictWriter(f, fieldnames=fieldnames)136        writer.writeheader()137        for rec in records:138            writer.writerow(139                {140                    "STUDYID": rec["STUDYID"],141                    "USUBJID": rec["USUBJID"],142                    "ASEQ": rec["ASEQ"],143                    "ADT": rec["ADT"],144                    "RANDDY": rec["RANDDY"],145                    "AVALC": rec["AVALC"],146                    "BORCAT": rec["BORCAT"],147                    "BORPRI": "" if rec["BORPRI"] is None else rec["BORPRI"],148                    "BORSEQ": "" if rec["BORSEQ"] is None else rec["BORSEQ"],149                }150            )151152153if __name__ == "__main__":154    main()

1/* opensas reference solution for the yamaa benchmark adam-adrs-response-prep.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data adrs_raw;6  length STUDYID USUBJID ADT RANDDY AVALC $1024;7  infile "/app/input/adrs_raw.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. ASEQ ADT : $char1024. RANDDY : $char1024. AVALC : $char1024.;9run;1011data adsl;12  length STUDYID USUBJID NTXSTDT $1024;13  infile "/app/input/adsl.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. NTXSTDT : $char1024.;15run;1617proc sql; create table joined as select a.*,b.NTXSTDT from adrs_raw as a left join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID; quit;18data categories;19  length BORCAT $16;20  set joined;21  if AVALC in ('CR','PR','SD','NON-CR/NON-PD','PD','NE') and (missing(NTXSTDT) or ADT<NTXSTDT) then do;22    BORCAT=AVALC;23    if AVALC in ('SD','NON-CR/NON-PD') and (missing(RANDDY) or input(RANDDY,best32.)<42) then BORCAT='NE';24  end;25  select(BORCAT); when('CR') BORPRI=1; when('PR') BORPRI=2; when('SD') BORPRI=3; when('NON-CR/NON-PD') BORPRI=4; when('PD') BORPRI=5; when('NE') BORPRI=6; otherwise BORPRI=.; end;26  unusable=missing(BORPRI);27run;28proc sort data=categories; by STUDYID USUBJID unusable BORPRI ADT ASEQ; run;29data result;30  set categories; by STUDYID USUBJID;31  if first.USUBJID then sequence=0;32  if not missing(BORPRI) then do; sequence+1; BORSEQ=sequence; end;33run;3435proc sql;36  create table final as select STUDYID, USUBJID, ASEQ, ADT, RANDDY, AVALC, BORCAT, BORPRI, BORSEQ from result;37quit;38proc export data=final outfile="/app/output/adrs.csv" dbms=csv replace; run;