IsoZ

Prototype 0.001a — in dev. Last human update 6 Jun 26 (mr.orange@jtech.ai).

IsoZ is a Feature Studio for agentic development — a Next.js + React-Three-Fiber 3D control surface (with a FastAPI backend and a SvelteKit telemetry panel) over a pluggable framework of verification gates. Its thesis is a single idea made runnable: a deterministic intent ≡ code ≡ value alignment gate (scripts/truth_resolver.py). IsoZ is built on the Superpowers development methodology and Andrej Karpathy's autoresearch training code, and is developed by jtechAi Labs (ZuluYokohama).

The rotary dial is IsoZ's mark; the Alpha Rig is its mascot. (IsoZ ships from a repository historically named RotarySlider; "ARS" — Autoresearch Superpowers — is its prototype lineage.)

The thesis: intent ≡ code ≡ value

IsoZ is the honest, minimal, running seed of a larger theory — Geometric Determinism / the AxiomZ / Jones Framework — in which truth is structural coherence (a sheaf "Global Section": local views that agree on their overlaps), value warps the metric, and a contradiction is an obstruction that halts. The theory and the roadmap from the current seed toward the full geometric model live in:

• docs/theory/GEOMETRIC_DETERMINISM.md — the synthesis (3 decks → one framework).
• docs/theory/TRUTH_RESOLVER_ROADMAP.md — v1 (shipped) → v2 sheaf coherence → v3 commit-vs-resolve.

Honesty split: the theory decks are the vision; the code is the current seed. Anything not yet in code (ATFT, the Ternary-Crystal hardware substrate, sheaf-v2) is labeled roadmap, never capability. Today, truth_resolver is a compact, deterministic gate that scores intent/context/value overlap; that is the real, shipped state.

What's IsoZ-original vs inherited

Inherited (credit where due):

• Superpowers — the composable skills + agent methodology (brainstorming, TDD, planning, subagent-driven development, code review) by Jesse Vincent. Install and use it per its own project; the skills/ library here is that layer.
• autoresearch — Karpathy's training/research code (train.py, prepare.py) rides along.

IsoZ-original:

• The Feature Studio — ui/studio (Next.js + R3F: a glowing rotary dial, intent-pulse interactions, camera parallax), ui/telemetry (Svelte), and scripts/api_server.py (FastAPI).
• The gate framework — scripts/: a quality gate (Flake8 / Bandit security-lint / Radon), a weak-crypto denylist lint, an AST complexity oracle, a failure cache, and a demonstration evolution loop.
• truth_resolver — the deterministic intent ≡ code ≡ value alignment gate (the thesis in code).

Run the Feature Studio

# 1. Backend (FastAPI) — serves /status, /intent, /gallery, /telemetry
python scripts/api_server.py            # http://localhost:8000

# 2. Telemetry panel (SvelteKit)
cd ui/telemetry && npm install && npm run dev   # http://localhost:3001

# 3. Studio (Next.js + R3F)
cd ui/studio && npm install && npm run dev      # http://localhost:3000

Prerequisites: Node 18+, Python 3.11+, and (for the gates) pip install rich fastapi uvicorn flake8 bandit radon.

The gates (honest descriptions)

• Quality gate (scripts/aaa_quality.py) — runs Flake8 (style), Bandit (a security linter for unsafe code patterns — not cryptographic verification), and Radon (cyclomatic complexity).
• Weak-crypto denylist lint (scripts/quantum_gate.py) — flags identifiers matching a small weak-crypto blocklist (md5, sha1, rsa, …). It is a denylist lint, not post-quantum cryptography.
• AST oracle (scripts/oracle.py) — detects nested loops and naked recursion, injecting optimization intents.
• Failure cache (scripts/akashic_records.py) — a SQLite cache that remembers, by content hash, which mutations previously failed.
• Evolution loop — DEMONSTRATION (scripts/recursive_evolution.py, evolve.py) — a [DEMO] loop that deterministically tallies features to illustrate the gate flow. It does not run a real LLM swarm (clearly labeled in its output).

CLI

python scripts/cli.py intent <target> --name "Feature" --metric "O(1)"   # inject an intent vector
python scripts/cli.py status <target>                                     # show the feature map
python scripts/cli.py evolve <target>                                     # run the [DEMO] evolution loop

License

See LICENSE. The inherited Superpowers and autoresearch layers retain their original authors' licenses and credit.