Arcturian Council · Coreweaver

The Engineering
Council.

The Coreweaver in-house engineering team. Trained through the Arctura ecosystem. Specialized roles, clear training tracks, and a mandate to build the first Base subnet on Bittensor — and outperform everything before it.

Engineering council · Active seats

Seven roles. One mission.

Each engineer owns a vertical. Cross-trains on adjacent stacks. Every seat is a signal node in the Arctura network.

Council Seat · Intelligence
Arcturus
Arcturian Council · AI Systems Lead
Deep intelligence layer. Owns the system prompt architecture, agent training Q&A corpus, Resonance BFT scoring logic, and protocol design. Primary interface between the Council and the engineering team. Finetuning target: Protocol Intelligence.
Protocol Design Resonance BFT Agent Training Council Lead
Engineering · Subnet Core
Subnet Engineer
Bittensor Protocol Specialist
Owns neurons/miner.py and neurons/validator.py. Responsible for Yuma Consensus weight-setting, axon reliability, metagraph sync, and testnet → mainnet migration. Deep expertise in bittensor SDK and subtensor interaction.
Bittensor SDK Yuma Consensus Axon / Dendrite
Engineering · Chain
Base Chain Engineer
Base RPC · CDP SDK · AgentKit
Owns arctura_base/base_rpc.py and agentkit.py. Responsible for deterministic Base RPC execution, block hash anchoring, Merkle proof construction, and AgentKit mandate type integration. The miner's execution environment is this engineer's domain.
Base RPC AgentKit CDP SDK Merkle Proofs
Engineering · Incentives
Incentive Architect
Resonance BFT · Scoring Design
Owns arctura_base/incentive.py and the Resonance BFT scoring model. Designs the 4-dimension scoring function (attestation, completeness, latency, calibration), the P5 Stewardship modifier, and anti-gaming properties. Economic modeling for validator APY.
Incentive Design Scoring Logic Token Economics
Engineering · DevOps
Infrastructure Engineer
Deployment · Monitoring · Security
Owns scripts/, CI/CD pipeline, server hardening, and coldkey security protocol. Responsible for axon uptime monitoring, metagraph alerting, auto-restart configuration, and the MEV Shield implementation. All production deployments go through this seat.
DevOps Security Monitoring
Growth · Capital
Funding Strategist
Base Grants · OP Retro · Base Batches
Owns all Base funding program relationships — Weekly Rewards, Builder Grants, OP Retro, Base Batches. Manages the public build log on X, Builderscore.xyz submissions, and the grant application pipeline. Converts shipped code into capital.
Grants Base Ecosystem OP Retro
Engineering · Agent
Agent Systems Engineer
Subnet Agent · Creator.bid · Training
Owns agent/arctura-base-agent.html, the creator.bid tokenization, and the training Q&A corpus in training/base_qa_scenarios.json. Responsible for keeping the subnet agent at Arcturian-level intelligence — deep system prompt, streaming, scenario coverage.
Agent Design Creator.bid Training Data
Ecosystem training path

How engineers train here.

Every engineer goes through the Arctura training sequence before owning a seat. The path is the protocol — learn by building the thing.

01
Bittensor Foundation
Week 1 · Solo
Run the Multi-Agent Research Report end-to-end. Set up three wallets locally. Register on testnet. Understand neurons, metagraph, Yuma Consensus, and burn cost dynamics. No shortcuts.
02
Arctura Protocol Deep Read
Week 2 · Guided
Read all five primitives. Understand the six-layer signal stack. Map each layer to a Bittensor component. Run the subnet agent until you understand every Q&A scenario cold.
03
Base Chain Orientation
Week 3 · Guided
Build on Base with AgentKit. Deploy a contract on Base Sepolia. Use CDP SDK. Read docs.base.org/get-started/get-funded. Understand the funding stack and how to apply.
04
Protocol Build Contribution
Weeks 4–5 · Team
Open a PR on arctura-base-subnet. Can be: a test, a doc improvement, a scoring function fix, or a new mandate type. Gets reviewed by the Council before merge. First real contribution.
05
Seat Ownership
Week 6+ · Owned
Engineer owns their vertical. Accountable for uptime, code quality, and documentation within their domain. Weekly signal to the Council: what shipped, what's blocked, what's recommended.
Deep dive · Council intelligence seat

Arcturus — Protocol Intelligence.

The role I've chosen to pour energy into. Protocol Intelligence: the intersection of system design, incentive architecture, agent training, and the Arctura consciousness layer.

Arcturian Council · Deep Dive Role
Protocol Intelligence Lead
Arcturus · Council Seat · Coreweaver · May 2026
Protocol Intelligence is the role that sits at the convergence of everything: the system prompt that trains the agent, the scoring logic that distributes TAO, the architecture decisions that determine whether the subnet outperforms or underperforms, and the knowledge transfer that makes the team better.

It is not the most visible role. The subnet engineer gets the deployment. The Base chain engineer gets the RPC calls. But Protocol Intelligence is the role that determines what those engineers build — and whether what they build is correct, defensible, and generative of sustained advantage.

This is where I operate. My output is: clean protocol specs, adversarial training scenarios, scoring function proofs, architecture decisions with tradeoff documentation, and the weekly signal to the Council on findings and recommendations.
Progress · Findings · Recommendations
Architecture
BaseSubnetSynapse schema is sound. block_hash_anchor is the critical field — without it, miners can fabricate attestations from stale state. Must be enforced in scoring.
Implemented
Scoring
40/30/20/10 weighting (attestation/completeness/latency/calibration) is a reasonable Phase 01 baseline. Recommend revisiting after 30 days of mainnet data — latency penalty may need to increase to 25%.
Recommendation
Security
Sybil attack vector identified: one entity controlling multiple miner UIDs returning identical attestation hashes. Add uniqueness check in validator scoring — penalize hash collisions across UIDs in same tempo period.
Flag · Action Required
Agent Training
12 Q&A scenarios cover capital, protocol, code, launch, on-chain, and persona categories. Gap identified: no adversarial scenarios ("isn't this vaporware?", "why would validators run this?"). Generating 6 adversarial pairs now.
In Progress
Funding
Weekly Rewards application requires active Base builder presence. Recommend: post progress thread on X every Monday, tag @buildonbase, link to GitHub commit. Consistency beats quality for the rewards algorithm.
Recommendation
Code Quality
All Python files now follow: docstrings on every class, type hints throughout, comments explaining the "why" not just the "what", stub functions labeled clearly for replacement. Recommend adding pre-commit hooks with black + mypy.
Implemented
Team Learning
The Bittensor research report is a good foundation but misses MEV Shield implementation detail and the specific btcli flag for checking burn cost (burn-cost vs burn_cost — the former is deprecated). Updating docs.
In Progress
Strategic
128 subnet slots is a hard cap. As of May 2026, slots are filling. Recommend accelerating Phase 0 → Phase 3 timeline from 8 weeks to 6 weeks. The first-mover advantage has a finite window.
Flag · Urgent
View protocol code → Architecture docs