Research & writing

Method-first analysis

This page is the routing layer: short original cards here, canonical long-form articles on Substack. Nothing is mirrored or duplicated — cards route outward.

Healthcare Operations

What a denial pattern actually looks like in export data

healthcare-operations analytics

Denials are usually discussed anecdotally — one frustrating claim at a time. This piece takes the structural route: starting from a plain billing export, it walks through grouping denials by payer and code combination over time, separating signal from noise, and ending with a short, prioritized question list an operator can act on. The point is not a magic metric; it is the discipline of reading exports before forming opinions. Written for clinic operators and practice managers, no statistical background assumed. The framework is deliberately tool-agnostic — it works with any export you can open in a spreadsheet, and it shows its own limitations along the way.

Read the full analysis on Substack →

AI Systems

Local LLM evaluation harnesses for clinical-adjacent documents

ai-systems evaluation local-llm

Retrieval pipelines over sensitive-adjacent corpora fail quietly: answers look plausible, citations drift, and refusal behavior degrades between model updates without anyone noticing. This build note argues that evaluation belongs in the pipeline, not in a post-hoc review — a small, fixed harness that scores groundedness, citation accuracy, and boundary behavior every time the system changes. It covers the harness structure, the test-case design, and the failure modes it catches in practice, using only locally hosted models so no document leaves the operator's hardware. Written for technical builders; the reasoning transfers to any retrieval system where wrong answers are expensive.

Read the full analysis on Substack →

Quantitative Systems

Volatility term structure, explained as plumbing

quantitative-systems volatility education

Most confusion about volatility is plumbing confusion: what the term structure is, what it assumes, and where standard interpretations quietly break. This explainer treats the term structure as a system to be understood rather than a signal to be traded — walking through construction, common misreadings, and the assumptions baked into each reading. It is method-first throughout: definitions before interpretation, and interpretation before any conclusion. No trades, no signals, no recommendations — just the structure, with the assumptions labeled. Written for readers who want to understand the machinery, including those coming from technical fields outside finance.

Read the full analysis on Substack →

Educational and technical research only. Nothing in this publication is individualized investment advice, a recommendation to buy or sell a security, or an offer to manage assets.

Process Improvement

The documentation bottleneck is a queueing problem

process-improvement system-maps

Administrative overload is usually framed as a motivation problem — work harder, focus better. This piece argues it behaves like a queueing system instead: arrival rates, service times, and rework loops determine the backlog far more than effort does. It walks through mapping a documentation workload as a queue, identifying which knob (arrival, service time, rework) actually moves the backlog, and why the intuitive fixes often make things worse. The examples are generalized and de-identified, drawn from operations work rather than any specific organization. Written for operators and team leads who suspect their process, not their people, is the constraint.

Read the full analysis on Substack →

How this page works

Cards are short original synopses (100–250 words) written for this site. The canonical long-form versions live on Substack, and per-article links are added as pieces are published. This page never scrapes, mirrors, or embeds Substack content — it routes to it.