// 02 — Field Notes

Things I'm

taking apart

on purpose.

Open research on the boring-but-important parts of running secure systems: policy decisions, identity bootstrap, and the graph of provenance underneath both code and data.

// Research themes

BPF-based runtime policy enforcement

eBPF programs as in-kernel policy decision points: drop traffic that doesn't carry a signed identity, block syscalls outside a workload's allowlist, and emit audit events without sidecars.

Secret-zero distribution

How do you bootstrap trust into a fresh workload without hard-coding a credential? Notes on SPIFFE/SPIRE, AWS workload identity, and short-lived attestation flows that make the first secret unnecessary.

SLSA + data lineage as one graph

Provenance for code (SLSA) and provenance for data (column-level lineage) are the same shape. A unified graph would let a security review and a data audit answer the same questions from the same source.

Iceberg manifest analysis

Deep-diving into Apache Iceberg's metadata structures. Understanding how table evolution affects query performance — and how snapshot expiry policies double as a data-retention control.

Block-based abstractions over DB / files / objects

Unified storage interfaces across databases, filesystems, and object stores. Can we abstract storage without sacrificing the locality and audit primitives each substrate provides?

// Open source

airflow-django-poetry

MAINTAINED

Reproducible base image for running Airflow alongside a Django app, dependencies pinned and resolved with Poetry. The same lockfile works locally, in CI, and on the scheduler / worker.

airflow

django

poetry

python

GitHub →

django-iceberg

MAINTAINED

Bridge between Django models and Apache Iceberg tables — write through Django's ORM, query through Iceberg readers. Targets the case where transactional writes and analytical scans share a schema without a separate ETL layer.

django

iceberg

data-lake

python

GitHub →

helm-query-exporter

MAINTAINED

Helm chart for the Prometheus SQL query exporter. Sane defaults, ConfigMap-driven query files, and a values structure that lets you wire up multiple datasources without forking the chart.

helm

prometheus

observability

sql

GitHub →

dbt-adapters

EXPERIMENTAL

Fork of dbt-labs/dbt-adapters tracking experiments around adapter-level behaviours that are hard to land upstream. Lives here so I can iterate fast on changes I want to test in production pipelines.

dbt

data

fork

python

GitHub →