blog.sateda.dev
Source-level analysis of cloud economics, lakehouse infrastructure, AI agent systems, security forensics, and artificial-life simulation. Every claim traced to code, price sheets, or a running probe.
GCP compute and data-warehouse economics — benchmarks, price modeling, and where the crossover points actually sit.
Performance characteristics of Cloud Run under load — the full benchmark writeup.
The formal paper version, viewable in-browser and downloadable.
Companion whitepaper distilling the benchmark findings.
Monthly cost, latency, and price-performance across seven workload classes in europe-west3 — with the break-even points.
The Apache Iceberg ecosystem, inspected at source level — client libraries, catalogs, streaming tiering, and data-product standards.
All six real Iceberg clients audited against docs and spec — including 31 places where the documentation is wrong.
Does the "sub-second real-time lakehouse" survive the object-storage commit floor? Right claim, adjacent noun.
A security-first Iceberg REST catalog that mostly earns the adjective — with the safety switch off by default.
A competent solo spec marketed as an industry standard: ~$44 lifetime funding, pay-to-vote governance, LLC hosting sold as "Linux Foundation family."
Agent harness design, MCP infrastructure, workflow methodology, and what agent sandboxes look like from the inside.
Verdict-gated speech: letting a language model answer immediately and audibly repair itself mid-utterance.
The governance-lane MCP gateway whose code mostly keeps its word — except for one unsafe default in the encrypted vault.
claude.ai chat, Claude Cowork, and Kimi probed from the inside — isolation, egress intent, and product posture.
The consolidated methodology: the philosophy, the settled six-phase spec, the proposed amendments, and the threads still open.
chat, Cowork, Kimi, and Google AI Studio — the three-way probe extended, with AI Studio shown off-axis as a preview host, not an execution sandbox.
Google AI Studio's environment audited: a Cloud Run app-preview host scored as if it were a code-execution sandbox — and why that's a category error.
Four layers, one brand: rigorous routing research whose statistical guarantee can't transfer to the open-ended-generation product being sold on its credibility.
Binary forensics and the behaviour of the machines that judge software.
EvoLab — an artificial-life simulator on conservative, biologically-faithful physics, aiming at emergent trophic webs and predation-driven multicellularity.
The full design-spec suite (7 documents) lives at sim.blog.sateda.dev →
World, physics, sensing & effectors, step loop, genome, constants, and grouping — the locked design documents in one place.
The field lattice, diffusing resource fields, POM particles, light, and the benthic floor.
The step-loop physics — reserve/structure dynamics, Michaelis-Menten uptake, low-Re diffusion.
The recurrent gene-regulatory network — node cap, wiring, mutation and duplication.