Datasets:

---

bbkbz

/

eduBench

like 1

What this changes

One value in DSE_chem_2025_paper1(en)/question_rubric.json, question 2(c):

 "rationales": {
   "rationale_1": {
     "Mentioned any TWO of substances": 1,
-    "Mentioned all the THREE underlined substances": 2
+    "Mentioned all the THREE underlined substances (additional mark)": 1
   }
 }

Why

2(c) has "max_score": 2, but its two rationale entries currently sum to 3.

Every other leaf question in this dataset — and in DSE_chem_2025_paper1,
DSE_phy_2025_paper1 and DSE_phy_2025_paper1(en) — has rationale values that sum
exactly to the question's max_score, so downstream consumers read the entries as
additive per-criterion maxima. 2(c) is the only exception across all four datasets.

The current values are a faithful transcription of the printed English marking
scheme, which awards by a ladder rather than by addition:

[Mentioned all the THREE underlined substances, award 2 marks;
Mentioned any TWO of substances, only award 1 mark.]

The rationales field has no way to express a ladder, so transcribing the tier
totals (1 and 2) makes the question look like it is worth 3 marks. A grader reading
the rubric additively can award 3 on a 2-mark question.

This patch keeps the ladder intact by recording the second tier as the increment it
adds rather than as its cumulative total:

• mentions any two underlined substances → 1
• mentions all three → 1 + 1 = 2

Both attainable totals are unchanged, and the entries now sum to max_score. The
(additional mark) suffix is there so the entry is not misread as "all three is worth
only 1 mark"; happy to drop it and change the number alone if you prefer a smaller diff.

Note on the Chinese edition

DSE_chem_2025_paper1/question_rubric.json already sums correctly for this question,
but that is not because it was transcribed differently — the printed Chinese
marking scheme is itself a different document, with two separate one-mark lines:

用濃氫氯酸潤濕一支清潔的鎳鉻 / 鉑絲，以鎳鉻絲沾少量水合硫酸銅(II) 晶體。 [1]
以無光焰把鎳鉻絲的末端加熱，並觀察所產生的焰色。 [1]

So this is not a case of the two editions disagreeing about the same source. I am
deliberately not proposing that the English entries be replaced with the Chinese
decomposition, since that would misstate the English marking scheme.

Verification

• Only this one line changes; the file is otherwise byte-identical.
• No student score changes: the 10 recorded scores for 2(c) are 0 (×7), 1 (×1) and
  2 (×2), all still reachable under the additive reading.
• 2(c) is the only rationale-sum mismatch in DSE_chem_2025_paper1(en),
  DSE_chem_2025_paper1, DSE_phy_2025_paper1 and DSE_phy_2025_paper1(en).