🧪 Testnet Phase #3 – Wrap-Up Recap

Testnet Phase #3 has been running for a few weeks, and we’re now wrapping up the current Phase #3 stretch. We moved from setup → real testing loops → product/data-layer alignment, with v3 agent surfaces + MVP data management + inference-quality signals exercised together.

Core Infra:
🔹 v3 /delegate + /validate (testing + iteration)
- Rollout + session prep complete; now iterating on real delegation/validation behavior
- Focus: routing consistency, consensus handling, and remaining logic gaps under load

🔹 MVP Data Management (Privacy + Offchain)
- Privacy Feature 1.0 + Offchain Storage v3 working end-to-end in MVP form
- Includes encryption (session/task), dedicated/ephemeral flows, deferred write, router→miner→dashboard integration

🔹 Dashboard / UI / UX improvements
- Quality stats + privacy/offchain flows now surfaced more clearly in the dashboard
- Added rank table + Node Perf quality columns + clearer completed/processing/stale task views

🔹 Inference Quality (SLA #3 live path)
- SLA #3 (user-task quality stats) added as a third selection filter on top of node + network-task signals
- Deployed on testnet0 + testnet1a and now being observed in live selection behavior

🔹 Inference Quality (oracle + data layer)
- MVP Quality Oracle + data model working end-to-end
- Task probes → stored stats → ranking views + selection/reward-oriented surfaces

🔹 Privacy Feature 1.0 (testing stage)
- Private inference live across router surfaces, tested across dedicated + ephemeral paths
- Session-level + task-level encryption flows have gone through matrix-style testing

🔹 Offchain Storage v3 (usable MVP path)
- Dedicated + ephemeral offchain flows working in MVP form
- Router-assisted deferred-write path for ephemeral nodes is live (remaining gaps tracked for iteration)

Products:
🔹 Bardiel dashboard – v3 adaptation
- Alignment work underway for v3 /delegate + /validate flows
- Heavier datasets + API examples added; iterating on task cards, results, consensus views, longer-input UX

🔹 New models for PyClaw iteration
- Added models 73–77 (Gemma 4 + Qwen variants) for PyClaw experiments
- Focus: tool-calling gaps and agent-style behavior on newer OSS models

🔹 PyClaw (early iteration + alignment)
- Early experiments started on newer models with dedicated sessions
- Focus: runtime/agent gaps + prep for deeper experimentation next phase

🔹 Portal V1 – design scoping started
- Scoping Portal V1 as a hosted inference layer on top of router infrastructure
- Focus: API key access, usage visibility, and managed router pools for developers

🔹 Router Node – multi-session + load balancing work
- Started design/code work for multi-session routing + load balancing on router nodes
- Supports Portal + PyClaw use cases with better traffic distribution and reliability

Phase #3 connected the pieces: agent surfaces (v3) + data management (privacy/offchain) + quality signals (SLA #3). Next is tightening reliability, UX, and mainnet-readiness based on what Phase #3 surfaced.

https://x.com/cortensor/status/2050822142485569685