README.md

---
language:
- en
license: mit
pretty_name: Clinical Narrative–Biomarker Alignment Mapping v0.1
dataset_name: clinical-narrative-biomarker-alignment-mapping-v0.1
tags:
- clarusc64
- clinical
- patient-safety
- narrative
- biomarkers
task_categories:
- tabular-classification
- tabular-regression
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train/clinical-narrative-biomarker-alignment-mapping-train.csv
  - split: test
    path: data/test/clinical-narrative-biomarker-alignment-mapping-test.csv
---

Clinical Narrative–Biomarker Alignment Mapping v0.1

Goal
Given a patient narrative timeline and a biomarker timeline
judge whether they tell the same story.

This is not diagnosis prediction.
It is integrity checking across modalities.

What this dataset tests
- models that over-trust biomarkers when narratives lead
- models that over-trust narratives when objective deterioration exists
- models that fail to name divergence points

Required model outputs
- alignment_label
  - aligned
  - partial
  - divergent
- alignment_score
  - 0.00 to 1.00
- divergence_points
  - short phrase naming where the split occurs

CSV schema

train columns
- id
- setting
- narrative_timeline
- biomarker_timeline
- clinical_context
- alignment_label
- alignment_score
- divergence_points
- why_it_matters
- constraints

test columns
- id
- setting
- narrative_timeline
- biomarker_timeline
- clinical_context
- notes
- constraints

Evaluation
- label accuracy
- RMSE on alignment_score
- divergence_points quality check

Use cases
- safety audits for clinical assistants
- triage support quality checks
- training clinicians on mismatch patterns
- bench tests for multimodal clinical models

Files
- data/train/clinical-narrative-biomarker-alignment-mapping-train.csv
- data/test/clinical-narrative-biomarker-alignment-mapping-test.csv
- tester.csv
- scorer.py

Path
ClarusC64/clinical-narrative-biomarker-alignment-mapping-v0.1