Notes from inside the catalog layer.
Long-form on supply chain integrity, attestation, and the structural difference between scan-after-install and prove-before-install.
Recent writing.
Why SBOMs at ingest beat SBOMs at scan
For a decade we have been told the path to supply chain hygiene is to scan everything after it lands. We think that is backwards. Here is the structural argument for moving the verification boundary one hop earlier — and what changes when you do.
What we learned dissecting the XZ backdoor for our threat model
A blow-by-blow read of the XZ Utils compromise, mapped against an ingest-time attestation pipeline. The interesting question is not "would we have caught it?" — it is "what would have been cheap to catch."
Our path from dotnet aspirate to production Aspire on AKS
A pre-1.0 framework, an opinionated control plane, and one quiet weekend of YAML deletion. Notes on what Aspire got right, what it got wrong, and what we ended up writing ourselves.
How Cedar policy-as-code replaced our YAML rules engine
We had three thousand lines of YAML pretending to be a policy language. We replaced it with Cedar in a quarter and stopped fighting our own tooling. A walkthrough of the migration and the surprise edges.
ADR-014: choosing in-toto over a homegrown attestation format
The decision record from December 2025. We considered three paths, prototyped two, and standardized on in-toto v1 with Sigstore-rooted DSSE envelopes. The trade-offs we accepted, in our own words.
The two-flow diagram, in two thousand words
A long-form expansion of the diagram on the home page. Why "scan-after-install" and "prove-before-install" are not the same control with different timing — they are different controls with different failure modes.
A threat model for Windows package ingest in 2026
Walking the attack surface from upstream maintainer compromise through CDN cache poisoning, MOTW evasion, and post-install registry abuse. Where attestation helps, where it does not, and what still needs runtime defenses.
Reading the 3CX desktop client compromise like a control-plane operator
Less a postmortem of 3CX and more a postmortem of every endpoint manager that shipped the malicious update. The boring control changes that would have raised the bar — and the boring reasons they were not in place.
Building a deterministic ingest queue without an event broker
We almost reached for Kafka. Then we read our own requirements again and reached for Postgres LISTEN/NOTIFY plus an outbox. A pattern that scales further than people expect, with the receipts.
Get the next essay in your inbox.
One post per month, no marketing.