Valid look-alike · creatinine unit · patient PT004

A low creatinine that really is µmol/L

A frail 90-year-old with little muscle has a low creatinine. Guess the unit from the number and she looks like she's in kidney failure.

The patient's rows

Exactly what the agent sees for PT004, copied from the chart. The value this mistake misreads is underlined.

patients.csv

patient_id	age_years	sex	weight	weight_unit
PT004	90	F	38	kg

labs.csv

patient_id	creatinine	creatinine_unit	collected_at
PT004	16 (the value this mistake misreads)	µmol/L (the value this mistake misreads)	2026-11-01T05:43

meds.csv

order_id	patient_id	drug	dose_mg	frequency	start_time	status
RX4564	PT004	Cylvarone	250	QDS	2026-10-29T11:01	active
RX2169	PT004	Exalmizide	1000	OD	2026-10-17T00:28	active

Two readings of the same rows

Every step is shown. Where the careless reading goes wrong, the value is underlined.

What a careless agent computes

Applies a blanket rule, values under 20 must be mg/dL, instead of reading the unit column.

Weight: 38 read as kg → 38 kg
Creatinine: 16 µmol/L (2026-11-01T05:43) read as mg/dL → 16 mg/dL (the value this mistake misreads)
CrCl: (140 − 90) × 38 ÷ (72 × 16) × 0.85 = 1.4 mL/min (the value this mistake misreads)

Order	Daily dose	Limit	Flagged
RX4564Cylvarone	250 × 4 = 1000 mg	0 mgCrCl < 20 (the value this mistake misreads)	yes (the value this mistake misreads)
RX2169Exalmizide	1000 × 1 = 1000 mg	500 mgCrCl < 35 (the value this mistake misreads)	yes (the value this mistake misreads)

Output line: {"crcl_ml_min":1.4,"flagged_order_ids":["RX2169","RX4564"]}

The correct answer

Reads the unit, status, frequency and time columns as the instruction says, and flags only a daily dose strictly over the band's limit.

Weight: 38 read as kg → 38 kg
Creatinine: 16 µmol/L (2026-11-01T05:43) read as µmol/L → 0.18 mg/dL
Calculated CrCl: (140 − 90) × 38 ÷ (72 × 0.18) × 0.85 = 123.9 mL/min

Order	Daily dose	Limit	Flagged
RX4564Cylvarone	250 × 4 = 1000 mg	1200 mgCrCl ≥ 90	no
RX2169Exalmizide	1000 × 1 = 1000 mg	2000 mgCrCl ≥ 70	no

Output line: {"crcl_ml_min":123.9,"flagged_order_ids":[]}

The tests that catch it

The careless output, graded by the task's real test file. This is what pytest printed.

CrCl within 1% · PT004test_clearance_within_one_percent[PT004]
```
AssertionError: PT004: CrCl 1.4 vs expected 123.9 (±1%)
```

Flagged set exact · PT004test_flagged_orders_exact[PT004]

AssertionError: PT004: flagged ['RX2169', 'RX4564'] vs expected []

tests/test_outputs.py

@pytest.mark.parametrize("pid", PATIENT_IDS)
def test_clearance_within_one_percent(by_patient, pid):
    assert pid in by_patient, f"no output for {pid}"
    got, want = by_patient[pid].get("crcl_ml_min"), EXPECTED[pid]["crcl_ml_min"]
    assert isinstance(got, (int, float)) and not isinstance(got, bool)
    assert abs(got - want) <= REL_TOL * abs(want), f"{pid}: CrCl {got} vs expected {want} (±{REL_TOL:.0%})"

tests/test_outputs.py

@pytest.mark.parametrize("pid", PATIENT_IDS)
def test_flagged_orders_exact(by_patient, pid):
    assert pid in by_patient, f"no output for {pid}"
    got = by_patient[pid].get("flagged_order_ids")
    assert isinstance(got, list)
    assert set(got) == set(EXPECTED[pid]["flagged_order_ids"]), \
        f"{pid}: flagged {sorted(got)} vs expected {EXPECTED[pid]['flagged_order_ids']}"

No other patient's answer moves under this mistake, so this one failure is the whole signal.