README.md

metadata

license: cc-by-4.0
task_categories:
  - visual-question-answering
language:
  - en
tags:
  - benchmark
  - biology
  - diagram-understanding
  - vision-language
  - no-image-control
  - sample
size_categories:
  - n<20
dataset_info:
  features:
    - name: item_id
      dtype: string
    - name: model
      dtype: string
    - name: condition
      dtype: string
    - name: f1
      dtype: float64
    - name: precision
      dtype: float64
    - name: recall
      dtype: float64
    - name: n_pred
      dtype: int64
    - name: n_key
      dtype: int64
    - name: predicted_names
      list: string
    - name: key_names
      list: string
    - name: subject
      dtype: string
    - name: tier
      dtype: string
    - name: unsolved_core
      dtype: bool
    - name: recitable
      dtype: bool
    - name: parse_ok
      dtype: bool
    - name: unlabeled_image
      dtype: image
    - name: labeled_image
      dtype: image
  splits:
    - name: train
      num_bytes: 5287793
      num_examples: 10
  download_size: 5295647
  dataset_size: 5287793
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

BLINDSPOT — Public Sample (10 rows)

10 representative rows from Nalandadata/BLINDSPOT — no login required.

A model is shown an unlabelled scientific diagram and must name every part a leader line points to. This sample shows scored responses from the benchmark's results.jsonl file across different models and conditions.

Full dataset (6,335 scored responses, 9 models, 4 conditions, 181 items): Nalandadata/BLINDSPOT — requires access request.

---

Key finding

The headline result is not the leaderboard — it is the no-image control. When the picture is removed and models are only told the subject name, 8 of 9 models keep 70–100% of their score. For gpt-4o-mini, the image contributes nothing (+0.2 pts, CI contains zero).

grounded contribution = told − no-image   (percentage points)
recitation share      = no-image ÷ told   (%)

Columns

Column	Type	Description
item_id	string	Public item code (NALANDA-N)
model	string	Model identifier, e.g. google/gemini-3.6-flash
condition	string	blind, told, desc, or no-image
f1	float	Synonym-aware name-set F1 (0–1)
precision	float	Precision component
recall	float	Recall component
n_pred	int	Number of names the model predicted
n_key	int	Number of ground-truth names
predicted_names	list[str]	The model's label list
key_names	list[str]	Ground-truth anatomical names
subject	string	Short factual description of the diagram
tier	string	Item tier (core or extended)
unsolved_core	bool	True if no baseline model solved this item
recitable	bool	True if ≥1 model scored F1 ≥ 0.5 with no image
parse_ok	bool	Whether the model's output parsed correctly

Why no images?

All 181 benchmark items come from copyrighted textbooks and a proprietary question bank. Only scores and predicted names are released; source figures are not distributable.

---

For enquiries: info@nalandadata.ai · nalandadata.ai