Science writing

Causal Inference Research Challenge:
Topic: Identification strategies in observational epidemiology
Technical focus:
- IV validity in Mendelian randomization studies with pleiotropy  
- Regression discontinuity design for policy evaluation in health systems
- Sensitivity analysis for unmeasured confounding using E-values
- Causal mediation analysis with time-varying mediators
Output: **Insight Note** comparing identification assumptions across methods, then **Contradictor** analysis of when each approach fails in practice.

Meta-Analysis Innovation:
Research target: Network meta-analysis for drug effectiveness
Methodological challenges:
- Handling inconsistency in treatment effect networks
- Ranking treatments under uncertainty using SUCRA scores
- Individual participant data vs. aggregate data approaches  
- Bias assessment in mixed treatment comparisons
Generate **Context Map** linking study characteristics to statistical heterogeneity patterns, with focus on transitivity assumptions.

Single-Cell Analysis Deep Dive:
Research focus: Pseudotime inference accuracy across trajectory topologies
Technical investigations:
- Benchmarking Monocle3, PAGA, and Slingshot on simulated branching processes
- Batch effect correction in trajectory space using Harmony vs. scVI approaches
- Integration of RNA velocity with pseudotime to validate trajectory direction
- Differential expression testing along pseudotime with tradeoffs between sensitivity and specificity
Create **Insight Note** on method selection criteria based on experimental design, followed by **Audio Brief** on interpreting trajectory confidence intervals.

Structural Biology Computation:
Target: AlphaFold confidence scores and experimental validation  
Research vectors:
- Correlation between pLDDT scores and crystallographic B-factors
- Domain-specific accuracy patterns in membrane proteins vs. soluble proteins
- Structure-based drug design using predicted vs. experimental structures
- Conformational sampling limitations in static structure predictions
Generate **Context Map** connecting confidence metrics to downstream application success rates.

NLP Interpretability Research:
Topic: Attention mechanism analysis in large language models
Technical focus:
- Head-specific functionality across transformer layers
- Attention pattern stability across prompt variations
- Probing tasks for syntactic vs. semantic representations
- Causal intervention experiments to test attention importance
Output: **Insight Note** synthesizing attention visualization techniques with mechanistic interpretability findings, then **Contradictor** analysis of alternative explanation frameworks.

Revolutionary Network Analysis:
Research focus: Mapping ideological connections across historical revolutions
Methodological approach:
- Network reconstruction from correspondence archives and pamphlet distribution
- Ideological similarity measurement using natural language processing
- Geographic diffusion modeling of revolutionary ideas
- Temporal correlation analysis between revolution outbreak timing and ideological transmission
Output: **Context Map** visualizing revolutionary idea networks across 18th-19th century Europe and Americas, then **Contradictor** analysis challenging traditional theories of revolutionary causation.

Historical Methodology Innovation:
Target: Automated bias detection in historical source materials
Technical challenges:
- Language model training on period-specific texts for anachronism detection
- Source reliability scoring based on contemporary cross-references
- Perspective bias quantification using sentiment analysis
- Historical fact verification through cross-source correlation analysis
Create **Insight Note** on computational approaches to historical source criticism, followed by **Audio Brief** on how AI can enhance rather than replace historian expertise.