csen2025.org https://validator.w3.org/feed/docs/rss2.html A Function That Needs a Section Header Wants to Be Two Functions Normalising Between Layers Made Deep Training Practical Technical Feasibility Is Not Ethical Permission Published Code Gets Checked, Which Is the Point If It Only Runs on Your Laptop It Is Not a Result A P Value Does Not Tell You the Probability You Are Right The Abstract Is Read a Hundred Times More Than the Paper The Optimiser Tells You Exactly What It Is About to Do Deploying Often Makes Each Deploy Boring A Broken Build Is Everyone’s Problem Immediately Long Lived Branches Are Merge Conflicts With a Calendar Errors Are Values, Not Interruptions Deprecate Loudly and Remove Slowly Breaking Changes Are a Cost You Pass to Everyone Else You Write Code Once and Read It Fifty Times Accuracy Is Useless When One Class Is Rare A Tree of Clusters You Cut Wherever You Like K Means Needs You to Know K Before You Start No Training at All, and a Very Expensive Prediction Finding the Widest Possible Gap Between Two Classes Every Split Asks One Question and Commits to It High Cardinality Breaks One Hot Encoding Quietly Making the Model Worse on Purpose So It Generalises A Model That Memorises Looks Perfect Until It Meets Reality Touching the Test Set Once Is Once Too Often Stop Calling the Service That Is Already Down Every Address Your Program Uses Is a Lie the Kernel Tells Turning a Two Dimensional Problem Into an Ordered List Every Substring of a Text, Indexed Once Skipping Ahead Instead of Comparing Every Character Research Slips Because the Unknown Cannot Be Scheduled A Filter That Is Sometimes Wrong on Purpose Wrapping a Set of Points in the Tightest Possible Rubber Band Nearly Constant Time by Rewriting History Greedy Works Until Somebody Constructs the Counterexample Recursion With a Notebook Is Dynamic Programming Big O Describes Growth, Not Speed You Do Not Choose Two, You Choose During a Partition Deciding When to Forget Is the Hard Half Eventual Consistency Means Eventually, Not Soon One Loop, Four Hundred Round Trips Provenance Matters More Than Volume Test Twenty Hypotheses and One Will Look Significant Statistically Detectable and Practically Useless Coexist Easily A Weak Baseline Makes Any Method Look Strong Every Flag Is a Branch in Production You Must Test Validating the Same Input Five Times Hides Where It Should Happen Both Work, Mixing Them Does Not Low Coupling Is What Makes Change Cheap Deep Networks Fail to Train Before They Fail to Generalise