07 / BLOG
Writing from
the laboratory
Notes on AI systems, reasoning architecture, performance engineering, and the ideas behind RED MAGIC.
FEATURED
Why This Website Is a Living System
AI is the leverage. The will behind it is what makes the software behave like an organism instead of a brochure.
Most websites behave like printed pages that occasionally blink. This one is built as an organism — and in 2026 that choice is the demonstration, not the decoration.
September 10, 2026 · 5 min read · Parsa Tak
Read articleengineering4 min read
FreeIran Engineering Notes
FreeIran is a local-first configuration manager built around a shared Go engine: three protocol cores behind one abstraction, checksummed chunked storage with crash recovery, a deterministic fake-core test harness, and no telemetry at all. These are the engineering notes.
Parsa Tak
research2 inbound references4 min read
Measuring Machine Intelligence
UHIT is the measurement arm of this laboratory, and its public form today is two canonical specifications — AIST-2026.09 and ASI-100-Elite-2026.09 — built on one doctrine: capability is multiplicative, verification is a load-bearing factor, and an ungameable benchmark is the only benchmark worth scoring.
Parsa Tak
research4 min read
Red Theory and the Living Web
RED THEORY is this laboratory's model of living systems: five dynamics that turn static structure into behaviour. The website is the current test bench — its organism demonstrates every dynamic in code you can read, and the boundary between the model and the demonstration is stated honestly.
Parsa Tak
engineering2 inbound references4 min read
SHEYTAN: The Local-First Engineering Laboratory
SHEYTAN Local Agent is a local-first desktop AI engineering environment: a managed llama.cpp engine, a real agent loop, seventeen governed tools, an isolated Coding Lab, and objective verification gates — because the model is never the authority on whether an engineering task succeeded.
Parsa Tak
research4 inbound references9 min read
AI Instructions: A Constitutional Operating Framework for AI Systems
AI Instructions is the master constitutional layer of a three-framework architecture: one document that governs how an AI system reasons, verifies, uses tools, and escalates. This is what it is, how it is built, and why the boundary between prompts and infrastructure matters.
Parsa Tak
systems5 inbound references5 min read
Building Under Constraints
GitHub Pages cannot run a backend, resize an image, or render on demand. Those limits are not obstacles — they are the design, and in 2026 they are also the discipline that makes AI-assisted building safe.
Parsa Tak
research7 inbound references6 min read
Reasoning Is a System Property
Three connected frameworks — AI Instructions, REP, USEF — built on one claim: intelligence lives in the system around the model, and the model has never been the smallest part of the problem.
Sep 11, 2026Parsa Tak
engineering5 inbound references5 min read
The Anatomy of a Fast Static Site
A tour of the time architecture behind this website: loading priorities, frame budgets, refresh-rate-aware quality, and the rule that banned fake progress bars before it banned fake frame-rate claims.
Parsa Tak
engineering4 inbound references5 min read
Why This Website Is a Living System
Most websites behave like printed pages that occasionally blink. This one is built as an organism — and in 2026 that choice is the demonstration, not the decoration.
Sep 11, 2026Parsa Tak
