Mode
DEMO · LIVE DEMO: animated brick-by-brick session · LIVE: your own program
Flow
Data
Compute
Chain
Target
DRAG BRICKS HERE — or click a brick in the palette to append.
Each brick emits NWO-ASM and a plain-English explanation below.
NWO-ASM Output
; empty program — place bricks to generate NWO-ASM
What each brick does
Placed bricks are explained here, one row per brick, as you lay them.
Mode
DEMO · LIVE DEMO: canned session · LIVE: real worker/adapter calls
Φ chainstate-code>
Mode
DEMO · LIVE DEMO: sample transcript · LIVE: real worker + adapter
Mode
DEMO · LIVE DEMO: canned pipeline · LIVE: real worker call

SIM Pipeline Simulation

Watch a cognitive transaction move through the full Ornith × CHAINSTATE pipeline. The consensus receipt stage performs a REAL call to the CHAINSTATE worker; the Ornith generation stage is simulated unless an adapter URL is configured (terminal: adapter <url>).

Idle. Press Run Pipeline.

01 Ornith-1.0 × CHAINSTATE — how they work together

Ornith-1.0 is a self-improving RL coding agent (9B / 35B / 397B MoE) that generates both solutions and the scaffolds that train them. CHAINSTATE is a symbolic-weight blockchain where transactions ARE cognitive queries. The adapter makes every Ornith code generation a symbolic transaction: embedded into the 65,536-d Universal Semiotic Embedding, submitted to reputation-weighted swarm consensus, compiled to NWO-ASM Process-Matrix IR for substrate dispatch.

02 Features — click any box for full detail

CHAINSTATE CODE WHITEPAPER

by Ciprian Pater · University of Agder · NWO Robotics

A peer-reviewable specification of the Ornith-1.0 × CHAINSTATE integration: how a self-improving RL coding agent (9B/35B/397B MoE) is coupled to a symbolic-weight blockchain via a 65,536-dimensional Universal Semiotic Embedding and reputation-weighted Bayesian log-pooling.

Introduces cognitive transactions as first-class blockchain primitives, formalizes the log-pool consensus kernel p_cons(x) ∝ ∏ p_i(x)^w_i, and specifies NWO-ASM Process-Matrix IR for GPU/QPU/NPU/edge substrate dispatch. Base mainnet 8453.

⇢ ResearchGate ⇢ Source
Podcast: CHAINSTATE CODE whitepaper walkthrough · by Ciprian Pater

02 System Summary

65,536

Dimensions

Universal Semiotic Embedding — six subspaces summing to exactly 2¹⁶.

3–7

Log-pool rounds

Reputation-weighted convergence at cosine ≥ 0.95.

k≥10

Swarm nodes

Ornith instances (mixed sizes) contributing symbolic states per query.

8453

Chain ID

Base mainnet · $STATE settlement · 35/35/30 MetaState split.

4

Substrates

NWO-ASM Process-Matrix IR dispatch: GPU · QPU · NPU · edge.

18

ASM opcodes

Bricks the visual builder emits: FLOW · DATA · COMPUTE · CHAIN · TARGET.

03 Related Work

CHAINSTATE CODE Whitepaper NEW

A Formal Framework for Agentic Coding on a Symbolic-Weight Blockchain — 18-page preprint, math + code + philosophical section on NWO-ASM composed with NWO-GENETIC.

ResearchGate 408393584 ⇢

CHAINSTATE Whitepaper v1.0

A Symbolic-Weight Blockchain for Cognitive Transactions — 19 pages.

ResearchGate 407444375 ⇢

Foundational Paper

Distributed Cognitive Work in Edge-Resident Language-Model Networks.

ResearchGate 406896310 ⇢

Casimir-Sonoluminescence Coupling

Peer-reviewed in Physics Essays. Co-authors Javaherian, Tariq.

Physics Essays ⇢

Ornith-1.0

Self-improving RL coding agent — deepreinforce-ai.

GitHub ⇢
Mode
DEMO · LIVE DEMO: canned evolve output · LIVE: polls the real adapter

CHAINSTATE AGI · ASI-Evolve integration

Bounded evolutionary program search over NWO-ASM · owner-controlled · step-limited · publicly observable

IDLE · experiment: chainstate_evolve
round
best score
step cap
50
owner
Ciprian F. Pater
score over rounds
latest lesson · from the Analyzer
Waiting for the first trial to complete…
best NWO-ASM program discovered
; the loop has not run yet — start it from the adapter to see live output.
DEMO screen · start LIVE mode to poll the real adapter

01 How ASI-Evolve impacts CHAINSTATE

Researcher
Reads Cognition Store + prior trials · proposes next NWO-ASM candidate
Engineer
Runs candidate through the evaluator · builds the program from 64k-symbol seeds
CHAINSTATE evaluator
Submits to the real worker · returns a ⛓ consensus receipt
Score
confidence·100 − gas·5000 − depth·2
Analyzer
Distills a lesson · writes trial + code + score to experiment DB

02 How the process takes shape in CHAINSTATE — technically

Search space

Every valid NWO-ASM program over the 64k-symbol pool: payload symbol choice, ordering, swarm size k ∈ [10,100], consensus depth d ∈ [1,7], TARGET ∈ {gpu, qpu, npu, edge}, optional LOOP / BRANCH insertion. Programs capped at 500 characters.

Cognition Store seed

10 items encoding the subspace layout (whitepaper §4.1), the log-pool convergence kernel (Eq. 4–6), the same-subspace density prior, substrate-subspace affinity, KV cache behavior, opcode constraints, and the ownership contract. The Researcher retrieves top-3 per round.

Evaluator

Every trial hits the real Cloudflare worker with cache: false for honest gas readings. Random symbol order per trial forces cache MISS. Candidate builders run under a 5-second CPU cap enforced by SIGALRM. Failed validation returns score 0.

Sampling strategy

Parent selection uses UCB1 with exploration weight 1.4 — favors under-explored high-scoring lineages. Runs single-worker (parallel workers would multiply gas cost against the real chain).

Persistence

Every trial is written to chainstate_evolve/database/experiments.sqlite: id, parent_id, score, motivation, code, lesson, timestamp. The public dashboard above polls that database read-only via /v1/evolve/*.

Safeguards

Owner-controlled: start/stop require the X-Owner-Token header. Hard step cap per run (EVOLVE_STEPS_MAX, default 50). Interruptible: POST /v1/evolve/stop SIGTERMs the process group. Cache flag intentionally off, so an accidental infinite loop is expensive and self-limiting.

Why not a Worker

ASI-Evolve is a long-running Python framework that writes SQLite state. Cloudflare Workers have no Python runtime, no persistent filesystem, and a hard CPU-time cap per request. The loop runs on Render (same box as the Ornith adapter); the adapter exposes read-only status through the standard FastAPI endpoints.

03 Features

  • Public output screen — polls /v1/evolve/status, /history, /best every 6 s in LIVE mode. IDLE labels are honest: no loop running, no fake numbers.
  • Score curve — live SVG chart of score over rounds. Points appear as trials complete.
  • Latest lesson — the Analyzer's summary of the most recent trial, in plain English.
  • Best program — the top-scoring NWO-ASM discovered so far, syntax-highlighted, copy-able.
  • Full-view mode — opens the screen in a fullscreen modal for demos and monitoring.
  • Owner-gated controls — start/stop endpoints require X-Owner-Token. Read endpoints public.
  • Honest labels — DEMO vs LIVE toggle. IDLE vs RUNNING state read directly from the process, not narrated.

04 Endpoint reference · adapter v1.2.0

POST /v1/evolve/startstart a run · owner-token required · steps capped
POST /v1/evolve/stopSIGTERM the process group · owner-token required
GET /v1/evolve/statuspublic · state, PID, rounds, best score, latest trial
GET /v1/evolve/history?limit=50public · every trial newest first
GET /v1/evolve/bestpublic · top-scoring program discovered so far