I Turned My Newsletters Into a Podcast. Here's What Broke.
A month of listening to a self-hosted newsletter-to-podcast pipeline, and what it revealed about TTS for journalism.
Read more →Recent Writing
All posts → 
Taiwan Solo, April 2026
A practical two-week clockwise loop through Taipei, Taitung, Tainan, Alishan, and back to Taipei.
Read more →
Things Just Work, and a Bit of Chaos
Notes from two weeks in Taiwan: civic order, street-level chaos, and the infrastructure that quietly carries the load.
Read more →
London in Three Dimensions
Two 3D maps from Threshold: one showing where London is most expensive by the square foot, the other showing where buyers pay more or less than distance from the centre would predict.
Read more →What a London Property Model Has Learned
Four maps from Threshold, a hedonic LightGBM model for London property prices: where space is expensive, what an extra bedroom is worth, and where the model knows it's guessing.
Read more →
Where to Start
On reading This Is Vegan Propaganda, the gap it left, and building a small tool to fill it.
Read more →
The Honest Price of a Flight
If airline offsets do not work, the next-best move is to treat the ticket price as artificially cheap and budget for climate advocacy on top.
Read more →
The Problem With Carbon Offsets
Why airline offsets fail on additionality, permanence, and scale, and why one long-haul trip home is not a rounding error.
Read more →
The Monks Were Tracking Climate Change All Along
Kyoto's cherry blossom record turns a thousand years of hanami diaries into a climate dataset, and the modern bloom dates now move with warming March temperatures.
Read more →
When the AI Narrator Gets It Wrong: A Stress Test of TTS on Serious Journalism
A stress test of OpenAI TTS and ElevenLabs on journalism-specific edge cases: context-sensitive words, units, acronyms, and publication-style proper nouns.
Read more →
The Eggs You Can't See
On caged hens, the restaurant visibility gap, and why I wrote to my MP about eggs.
Read more →Language Models Are Not Uncertain in One Way
GPT-3.5-turbo on 120 questions, six difficulty tiers, and what the model's own signals do and do not tell you about when to trust it.
Read more →Does Try Location Matter? I Looked at the Data.
I assumed touchline conversions were nearly hopeless and central kicks were automatic. The data says I was wrong at both ends.
Read more →Why I Used to Think None of This Mattered
On rationalisation, Peter Singer, factory farming, and the slow process of aligning what you believe with how you actually live.
Read more →