Writing
Essays, analysis, and notes on applied AI, digital assets, investigations, and the systems around them.
- Sep 15, 2026 · Investigation The Lead Is Not the Case I keep unverified observations separate from case work. An interesting address is not a case until evidence justifies promoting it for deliberate investigation.
- Sep 11, 2026 · Investigation Why My Numbers Are Floors, Not Totals A trace total is the minimum the record shows, not all that moved. Pagination, coverage windows, and source limits make a clean figure a claim with boundaries.
- Sep 7, 2026 · Investigation I Name the Crime Last, Not First The instinct is to label a matter first and gather data that fits. I do the opposite: the record arrives first, and any classification has to earn itself.
- Sep 3, 2026 · Investigation Falsify First: The Question I Ask Before I Believe My Finding Before reporting a finding, I ask what else could produce the same on-chain record. A conclusion is only as strong as the alternatives it has survived.
- Aug 30, 2026 · Investigation Evidence Tiers: Why a Good Report Says Consistent With, Not Proven A sound forensic report separates what the chain proves, what evidence supports, what describes infrastructure not identity, and what stays unresolved.
- Aug 15, 2026 · Thoughts The Expertise Paradox: AI Is Eating the Junior-to-Senior Career Ladder The real AI threat is not mass job loss alone. It is the collapse of the training ground. Automate the easy work and you may erase how expertise gets built.
- Aug 11, 2026 · Investigation How a Standing USDT Approval Can Expose Funds That Arrive Later An anonymised Tron USDT drain: a standing approval can outlive an empty balance and expose funds that arrive later. Recovery needs off-chain evidence.
- Aug 6, 2026 · Thoughts First Build the Machine. Then Add the AI Engine. Most companies add AI before they have a workflow worth amplifying. AI amplifies: structured processes scale, broken ones create faster chaos.
- Aug 4, 2026 · AI If You Still Babysit Every Output, You Do Not Have an Agent Most launches labelled AI agent are chatbots in costume. A real agent is architecture: goals, tools, permissions, memory. Map the workflow, not the label.
- Aug 1, 2026 · Thoughts AI Should Enhance the System, Not Subsidize Unclear Thinking A powerful model cannot fix unclear thinking. It makes it faster and more confident. The advantage is directing the workflow: context, constraints, checks.