Blog

Engineering notes, leadership insights, and hobby experiments.

• Heikin-Ashi: Anatomy of an Edge That Dissolves Under Controls Jun 15, 2026

  A discovery follow-up that surfaces a clean, mechanistic-looking Heikin-Ashi signal — the bullish colour reversal — and then watches it dissolve under the right controls: a raw (non-HA) reversal does as well or better, the edge is concentrated in survivors and vanishes on delisted names, and it never beats buy-and-hold. A case study in how a carefully-validated signal can still evaporate.

• Heikin-Ashi – An Empirical Study with Ablation, Across Equities, FX and Crypto Jun 14, 2026

  An empirical evaluation of Heikin-Ashi trend-following — with an ablation that isolates the HA signal from its EMA/stop scaffolding — across 25 equities, 12 FX pairs and 8 crypto assets, plus a 20-year daily cross-section (2006–2026).

• Heikin Ashi – Part 7: From Failures to Checklists May 25, 2026

  Every AI-assisted project session generates failures worth keeping. Almost none of them get kept. The final discipline in the series: extracting what went wrong into artifacts the next session can actually use.

• Heikin Ashi – Part 6: When AI Confidence Lies May 22, 2026

  AI confidence fails in at least five distinct ways, each with its own recognition signal, its own typical cost, and its own counter-move. A field guide to premise drift, false architectural confidence, speculation as diagnosis, over-generalization from precedent, and confident invention of identifiers.

• Heikin Ashi – Part 5: The Operator's Manual May 20, 2026

  Debugging a running serverless system is not design chat. A third mode — operational chat — has its own rules: one command per turn, hypotheses labelled as hypotheses, file paths verified not guessed. A taxonomy of where my own deployment sessions broke those rules and why it mattered.

• Heikin Ashi – Part 4: From Spec to Production May 17, 2026

  Eight hours debugging production-only failures on a system that took thirty hours to design and build. A taxonomy of the boundary defects that CLAUDE.md can state but is structurally unable to prove.

• A watchlist sentinel that emails me only when a stock does something interesting May 16, 2026

  A serverless daily watchlist watcher that stays silent until a stock forms a meaningful Heikin Ashi pattern — then sends one email with a chart, the numbers, and AI-generated context. The inbox is the interface.

• Heikin Ashi – Part 3: The Two Claudes Pattern May 13, 2026

  A workflow pattern using two Claude products as distinct roles — chat for architecture and decisions, Claude Code for implementation — with CLAUDE.md as shared memory and a context bundle as the regenerable bridge.

• Heikin Ashi – Part 2: Designing by Chat May 11, 2026

  A 50-turn design conversation reconstructed as a dialogue with an imaginary critic — on premise drift, false architectural confidence, and why CLAUDE.md is not documentation but executable belief.

• Heikin Ashi – Part 1: Designing the Monitoring Service May 10, 2026

  A serverless monitoring service for equities — computes Heikin Ashi candles, detects patterns, enriches alerts with AI-generated fundamental context, and sends them by email. Designed entirely in a long chat with Claude.

• Terracotta Army – Part 3: Ergonomics and Protocol Apr 26, 2026

  Before the first drop of paint, we address the ergonomic gap and define the protocol — four trials pitting 78 specialists against Raphaël 8404 and Winsor & Newton Series 7 controls.

• Terracotta Army – Part 2: Mechanics and Behavior Apr 25, 2026

  A dive into hair micro-mechanics — how snapback, stroke memory, and capillary flow explain why a 20-cent specialist brush might beat a $30 Kolinsky at its own game.

• Terracotta Army – Part 1: The Strategy Apr 24, 2026

  A $56 wager on 78 specialist Chinese brushes — testing whether extreme functional decomposition of quality can outperform a few expensive all-rounders in miniature painting.

• Web Services & REST

  A study guide on REST principles, HTTP semantics, and API design — including four quizzes to test understanding of the core concepts.

• Contract Testing

  A testing approach that verifies each side of a service integration meets the agreed contract, catching integration bugs without requiring both services to be running together.

• Smoke Testing

  Notes on smoke testing in regulated environments — fast sanity checks run after a deployment to confirm critical paths still work before deeper verification begins.

• System Architecture

  Study notes covering system design fundamentals — scalability, reliability, trade-offs, and the vocabulary used in architecture discussions and technical interviews.

• Circuit Breakers

  A pattern that prevents cascading failures in distributed systems by short-circuiting calls to a failing service, letting it recover before traffic is restored.