Literature

• Reading a paper critically: claims, ablations and what the appendix hides
• Reproducing a published result and diagnosing why it does not reproduce
• Tracking a research area without drowning in preprints

Method

• Baselines, ablations and controls that isolate the contribution
• Seed variance, multiple runs and reporting distributions not single numbers
• Statistical significance and effect size in benchmark comparisons
• Compute-fair comparison when budgets differ between methods

Execution

• Experiment tracking, configuration management and reproducible environments
• Scaling experiments across GPUs without losing determinism
• Negative results: recognising them early and reporting them honestly

Output

• Structuring a paper so the contribution is unmistakable
• Writing a rebuttal that engages rather than deflects
• Reviewing for a venue and the obligations that carries
• Research ethics, dual use and responsible disclosure