Real systems · real decisions · reusable models

I make complex systems easier to reason about.

I test ideas through engineering, products and real decisions, then document the models and tools that survive contact with reality.

Start here · flagship case

Automate the job, not the typing

One engineer pushed software automation on a live product with more than 500 users without delegating judgment or production authority.

More cases02

Release safety

Situation: More than 1,700 tests and a staging walk passed, but the production sequence could still serve and cache an empty result.
What changed: The release gate changed from checking one push to proving that every inherited data obligation was covered in commit ancestry.
Useful takeaway: Tests approve examples. Independent reviewers search for harmful sequences.

4 min read

Data and product claims

Situation: The product displayed 534 while the source displayed 975. Every local calculation was correct, but the product used the wrong aggregation grain.
What changed: The investigation normalized the visible claim, traced one value back to source, and preserved the disagreement as executable semantics.
Useful takeaway: Correctness belongs to the complete path from source grain to product language.

4 min read

Deployment ordering

Situation: A later copy-only commit could deploy code that inherited an earlier commit's unfinished data dependency while a file-diff gate saw nothing relevant.
What changed: The dependency became a graph invariant that requires a successful workflow commit between the required change and the release.
Useful takeaway: A release obligation belongs to ancestry, not to the event that happens to trigger deployment.

3 min read