NOMI Dashboard
NOMI Framework ← Back to Samposhyam.org

Sundarbans Benchmark Dataset

Interactive exploration of the NOMI framework validation data

DEMO MODE: Anonymized Data
317
Codes
19
Themes
473
Relationships
4,533
Evidence
45%
Cross-Pillar
Overview
Pillars
Relationships
CoE Scores
AMRP

Pillar Distribution

Near-perfect pillar balance (Shannon evenness: 0.99). 317 unique codes distributed across 4 pillars from 44 respondents in Gosaba and Patharpratima blocks, Indian Sundarbans.

Key Metrics

MetricValueInterpretation
Code Density7.20317 codes / 44 respondents
Relationship Density1.49473 relationships / 317 codes
Evidence Depth9.584,533 evidence records / 473 relationships
Modularity (Q)0.770Strong community structure (Louvain)
Cross-Pillar Rate45%213 of 473 relationships cross pillars
Shannon Evenness0.99Near-perfect pillar balance across 4 pillars

Nutrition as a Socioecological System (NSS)

95 codes encompassing ecological, agricultural, and environmental determinants of nutritional adequacy.

ThemeCodesRelationships
T01: Salinity-Nutrition Nexus1832
T02: Water Security1528
T03: Dietary Diversity1424
T04: Climate-Food Pathways1221
T05: Agricultural Stress1119

Observational Resilience (OR)

75 codes capturing community-level adaptive capacities, coping strategies, and livelihood diversification.

Multi-Lens Equity (MLE)

60 codes addressing structural inequities in food access across gender, age, caste, and economic position.

Inclusive Knowledge (IK)

87 codes encompassing traditional ecological knowledge, intergenerational food practices, and indigenous preservation techniques.

Cross-Pillar Relationships (213 of 473, 6 types)

Relationships classified across 6 types: causal, correlative, moderating, mediating, feedback, and compound.

FromToCountDirection
NSSOR58Bidirectional
NSSMLE42Bidirectional
NSSIK31Bidirectional
ORMLE37Bidirectional
ORIK24Bidirectional
MLEIK21Bidirectional

Convergence of Evidence (CoE) Distribution

25
Strong
448
Moderate
0
Weak

CoE computed using Shi et al. (2025) methodology adapted for qualitative CQT framework. 4,533 evidence records from peer-reviewed literature validate 473 relationships.

AMRP (Adaptive Metric Recalibration Protocol) compares a new site's NOMI analysis against the Sundarbans benchmark to identify methodological strengths and gaps.

Benchmark Profile

DimensionSundarbans ValueInterpretation
Code Density7.20High saturation
Cross-Pillar Rate0.45Strong integration
Pillar Balance0.99Near-perfect evenness
Relationship Density1.49Rich connectivity
Evidence Depth9.58Thorough triangulation
Modularity (Q)0.770Strong community structure
CoE Strong Rate0.053Top-tier evidence

How to Use AMRP

Run nomi-apply amrp-diagnose on your project to compare your site's metrics against this benchmark. The diagnostic radar chart highlights dimensions where your analysis may need strengthening.

Apply NOMI to Your Data

Transform raw qualitative data into a structured NOMI framework analysis for program design and policy translation

Launch Full Pipeline on HuggingFace
🔬

For Researchers

Analyze qualitative field data through the four NOMI pillars. Discover thematic communities, map cross-pillar relationships, and validate findings with literature.

🎯

For Program Designers

Turn NOMI analysis outputs into actionable program recommendations. Use AMRP benchmarking to identify gaps and the One Health pathway for policy translation.

What you need to start: Interview transcripts (plain text, .txt/.csv/.xlsx), a coding framework (or use NOMI's 317-code codebook), and access to the NOMI Dashboard on HuggingFace Spaces.

7-Step Pipeline: Raw Data to NOMI Output

1

Upload Transcripts

Upload interview transcripts in .txt, .csv, or .xlsx format. Bengali text is auto-detected. Supports 3 formats: one-per-file, tabular, or multi-column.

Input: Raw interview transcripts from your field site
Output: Parsed, cleaned text segments ready for coding
If this fails: Check file encoding (use UTF-8). For Bengali text, ensure the file is not saved as ANSI. If .xlsx fails, try exporting as .csv first.
2

Four-Pillar Coding

Auto-code text segments against the NOMI 317-code codebook using TF-IDF similarity. Each segment is assigned to NSS, OR, MLE, or IK pillars.

Input: Parsed text segments from Step 1
Output: Coded segments with pillar assignments and confidence scores
If this fails: Low confidence scores (<0.3) mean your data may not align with the existing codebook. Try adding custom codes specific to your study context, or lower the threshold to capture broader matches.
3

Community Detection

Run Louvain/Leiden algorithm on the code co-occurrence network to discover thematic communities. Target: Q > 0.4 for meaningful structure.

Input: Coded segments with co-occurrence matrix
Output: Thematic communities (themes) with modularity score
If this fails: Low modularity (Q < 0.3) suggests insufficient data. Collect more interviews (aim for 30+ respondents) or refine coding to increase code diversity. If Q is very high (>0.9), your pillars may be too siloed; check for cross-pillar codes.
4

Map Cross-Pillar Relationships

Identify directional relationships between codes using 6 types: causal, correlative, moderating, mediating, feedback, and compound.

Input: Coded data with thematic communities
Output: Relationship network with cross-pillar rate (target: >30%)
If this fails: Cross-pillar rate below 20% suggests your interview protocol may not capture inter-pillar dynamics. Add probing questions that bridge pillars (e.g., "How does water quality affect your market access?"). Review the 6 relationship type definitions.
5

CQT Literature Validation

Validate relationships against published literature using the Convergence-Qualification-Triangulation protocol (Shi et al. 2025 weights).

Input: Relationship network from Step 4
Output: CoE scores (Strong/Moderate/Weak) for each relationship
If this fails: Many "Weak" scores are normal for novel contexts. Focus on relationships with Moderate+ scores for program design. For weak relationships, search for grey literature or regional studies that may provide supporting evidence.
6

AMRP Benchmarking

Compare your site's 7 AMRP dimensions against the Sundarbans benchmark to identify where your analysis is strong and where it needs attention.

Input: Your site's metrics (code density, cross-pillar rate, modularity, etc.)
Output: Radar chart comparing your site vs. benchmark across 7 dimensions
If this fails: Dimensions below 50% of benchmark suggest methodological gaps. Low Code Density: collect more data or refine coding. Low Evidence Depth: expand literature search. Low Pillar Balance: ensure interview protocol covers all 4 NOMI pillars equally.
7

One Health Policy Translation

Generate program design recommendations through the AMRP pathway, connecting human-animal-environment dimensions for actionable policy outputs.

Input: Validated NOMI analysis with AMRP benchmarking
Output: Policy briefs, program design recommendations, publication-ready figures
If this fails: Weak One Health connections may mean your study context is more sector-specific. Focus on the strongest cross-pillar pathways for program design. Consider which NOMI pillars are most relevant to your local policy context.

Ready to Apply NOMI?

The full interactive pipeline runs on HuggingFace Spaces with step-by-step guidance, file upload, and real-time results.

Launch NOMI Pipeline