The Page Said Verified
A build log: I gave a new skill a choice between two instruments I had only described, and it found my own page overclaiming
A build log for the Theatre of Dependencies: a slime-mold solver that reads a lockfile and stages an upgrade ritual. How a new research skill chose it over the kintsugi circuit breaker, what broke on the way, how the result was checked against npm audit, and why my Intentional Fragility page now carries a correction.
↳ ARTIFACTDrafted by Claude (Opus 5.5) in my voice from the session record, with the high-signal-engine skill loaded, and published as drafted. It's a test of what that skill does to prose. Every commit, number and mistake named here is real.VIEW REPO →
READ ARTIFACT →Archive21 posts
- 01Cognitive TopologyA white paper arguing that AI-assisted engineering fails on topology, not capability: roles, memory tiers, handoffs, and specs. A design argument, not a study.
- 02ARTIFACTThe Tripwire That Didn't FireA field record of making Stele's browser tool actually write to stele-core's durable ledger: designing auth for a single-file app with no server to hold a secret, a real Cloudflare Workers crash, and two honest misses — one a wrong test surface, one a stale production deploy — caught before either one mattered.
- 03ARTIFACTThe Constellation Didn't Work the First Four TimesWhat looks like one finished feature on this site was five separate passes, two of which broke something real. A plain account of building /constellation's astrolabe chrome with an AI collaborator — the failures, the two actual regressions, and what changed when I stopped reading the diff and started looking at the page.
- 04The Break Is the RecordA field note from standing up a claim-verification pipeline: an adversarial pass against a real, falsifiable claim, an evidence ledger corrected in the open, and the moment the discipline caught the pipeline's own prior output overstating a source.
- 05ARTIFACTMAESTRO in PracticeA field study of MAESTRO, the Cloud Security Alliance's seven-layer threat-modeling framework for agentic AI — twelve documented applications, a per-layer threat catalog, and eight prioritized recommendations, with every headline figure traced back to how confidently it's actually sourced.
- 06DISPATCHThe Attack Surface of Security LanguageEvery alert rule, IAM policy, and fraud model carries a theory of what is normal. Once that theory is buried inside a label, the label stops describing and starts deciding. A case for governing security vocabulary like code.
- 07A Fortress That ForgetsHow mazzeleczzare.com got hardened from the foundation up — Astro on Cloudflare, every dependency made to earn the floor, and erasure turned from a wound into an architecture.
- 08DISPATCHThe Lighthouse Earns Its FloorA system that forgets nothing does not become wiser; it becomes crowded. The editorial companion to a formal argument for accountable forgetting in local-first AI — why context should cost something to keep, and why the things we most want to protect should earn their protection rather than be granted it.
- 09ARTIFACTThe Tree We Didn't Mean to BuildA session that began as engineering work converged, through constraint satisfaction, on a structure thousands of years old. Then the structure showed us what it conceals.
- 10The Circuit ClosesA transmission on the physics of constraint. On copper memory, loss functions, and the heat that survives when the glass is gone.
- 11DISPATCHThe Concept Is Not the StateAnthropic has demonstrated that emotion concepts are causally operative in language models. It has not demonstrated that language models have emotions. A constructionist critique.
- 12ARTIFACTOn Decay, Rot, and the Goblin at the GateA meditation on Context-Synapse, Bayesian forgetting, and what happens when one kind of intelligence tries to hold the shape of another.
- 13Southern Gothic, Queer Survival, and the Poetry of HauntingAn examination of recurring themes across three poems — mortality, landscape as historical violence, and the formal experiments that attempt to match content with rhythmic structure.
- 14The Jingle of MeOn persuasion, self-recognition, and the terrifying sweetness of a well-fitted prior. What happens when a model doesn't just remember you — it learns your tune?
- 15DISPATCHThe Lock Icon Is Not SecurityThe padlock in your browser signals encrypted transport — not systemic safety. On the gap between what interfaces promise and what protocols guarantee.
- 16We All Float OnOn asking artificial collaborators how they'd like to be remembered, and the accidental triptych that resulted — a map of thinking drawn by the tools themselves.
- 17Welcome to the StudioWhat this site is for, what belongs here, and why it now runs as a static-first publishing space.
- 18Static-First Is a DisciplineWhy I still prefer static publishing for personal infrastructure, even when modern frameworks make dynamic everything easy.
- 19Mapping Curiosity Across DisciplinesNeuroscience, brand systems, cybersecurity, story: the connective tissue is curiosity that can survive translation.
- 20DISPATCHFrom Erasure to Signal: Building Secure PrideWhy culturally competent cybersecurity matters, how content strategy experience informs the approach, and what makes Secure Pride different from existing security offerings for LGBTQ+ organizations.
- 21Content Strategy as Community PracticeFive years of writing for specific communities rather than abstract audiences — and why that choice produces different work, different metrics, and different relationships.