# Testnet Phase #2

Phase #2 advances Cortensor from “system-wide regression” into **trust-layer hardening** — tightening validator logic, expanding redundancy, and beginning **formal gas + runtime optimization**.\
Building on Testnet Phase #1’s end-to-end stability baselines, this phase focuses on making validation **more accurate, more resilient, and cheaper to operate** at scale across both **Testnet-0 (Arbitrum Sepolia)** and **Testnet-1 (L3 COR Rollup)**.

***

### Phase 2 — Validation Refinement & Infrastructure Hardening

**Timeline:** Early Q1 → Mid Q1 2026\
**Goal:** Strengthen validator logic, expand redundancy, and begin formal gas optimization.\
**Focus:** Accuracy, trust stability, and infrastructure robustness.

***

### Contest Rules

#### Stake Requirement

* All participants must stake **10,000 $COR per node (Pool 3)** to qualify for Testnet Phase #2.
* Nodes must remain staked and active throughout the testing window to receive rewards.

#### Node Uptime & Performance

Operators must:

* Maintain **continuous uptime** and responsiveness to assigned sessions.
* Support **redundant validation flows** (more multi-run / multi-miner validation paths).
* Handle **higher validator throughput**, including scoring + reporting without bottlenecks.
* Ensure correct registration + telemetry sync across both **Testnet-0** and **Testnet-1** when participating in both.

#### Participation & Reporting

Participants are expected to:

* Report bugs, validator anomalies, scoring drift, or reliability issues.
* Help validate improvements to **Validator v2/v3**, redundancy policies, and router reliability.
* Share structured logs + findings through Discord reporting channels.

#### Disqualification Criteria

Cortensor reserves the right to disqualify any operator for:

* Misreporting uptime or falsifying telemetry data.
* Exploiting reward, routing, or validation mechanisms.
* Any manipulation that undermines fairness, accuracy, or network stability.

***

### Rewards

**Base Reward**

* Base rewards is $20 per node.
* Qualification Requirements:
  * Active uptime during the entire test window
  * **Precommit Score ≥ 60 %**
  * **Node Level Success Rate**

**Role-Based Bonus**

* **+1 % per verified Discord role** (capped at 10 %).

**Node Level Bonus**

* Level-based bonus scales linearly (**+$0.80 per level**, 0–10):
  * Level 0 = $0 → Level 10 = $5 extra per node per month.

**Performance Prize**

* Additional $5 bonus for nodes meeting both:
  * **Total Score ≥ 50 %**
  * **Precommit Score ≥ 60 %**
  * Leaderboard → [View Testnet #0 Dashboard](https://dashboard-testnet0.cortensor.network/)
  * Leaderboard -> [View Testnet #1 Dashboard](https://dashboard-testnet1.cortensor.network/)

**Multi-Node Bonus (New Ops Only)**

* 3 + nodes → +10 % bonus
* 5 + nodes → +20 % bonus

***

### Testing Focus

#### 1) Validator v2/v3 Refinement (PoI + PoUW)

Phase #2 is centered on tightening trust calibration and reducing scoring drift:

* **Deep PoI (quantitative)** validation refinement: improve numerical consistency, dispersion bounds, and convergence under redundancy.
* **Deep PoUW (qualitative)** validation refinement: strengthen rubric adherence (accuracy, relevance, completeness, clarity, safety) and reduce judge variance.
* Improve validator determinism under real load: fewer “flaky” outcomes, better repeatability, clearer failure modes.

#### 2) Validator Engine Experiments: Ollama Models + Embedding Variants

We will explicitly experiment with **Ollama-based model stacks on validator nodes** to improve hardware compatibility and verification quality:

* Validate **Ollama model variants** as LLM-verifier options (judge models / verifier models), comparing stability, latency, and scoring consistency.
* Test **multiple embedding model variants** to address known issues where certain validator GPU setups produce inconsistent/unsupported embedding behavior with the current engine.
* Target outcome: **better hardware adaptation on validator nodes**, fewer environment-specific failures, and more predictable validator throughput.

#### 3) QuantitativeStats / QualitativeStats Expansion

To tune thresholds and reduce false positives/negatives:

* Expand datasets for calibration of **PoI thresholds** and **PoUW scoring**.
* Improve LLM-verifier scoring alignment across task domains (general QA, code, reasoning, extraction, etc.).
* Track distributions over time to detect drift, regressions, and “boundary-case” instability.

#### 4) Redundancy Expansion & Trust Stability

Strengthen reliability through broader redundancy policies:

* Increase multi-run / multi-miner redundancy where needed.
* Validate aggregation logic (median/majority/weighted) and ensure it behaves correctly under partial failures.
* Measure trust stability: score variance, disagreement rate, and disagreement handling behavior.

#### 5) Router Surface Experiments: `/delegate` + `/validate`

Phase #2 expands structured testing of the router’s agent-facing surfaces:

* `/delegate` experimentation:
  * Policy-driven delegation (cost/risk hints), routing behavior, retries, and fallbacks
  * Multi-run delegation and “same task, different miners/models” variance testing
  * Response structure stability for downstream agent consumption
* `/validate` experimentation:
  * Validator tier selection, scoring stability under different verifier policies
  * Regression testing across domains and edge cases (hallucination, partial answers, ambiguous prompts)
  * Throughput/latency profiling for validation-heavy workloads

#### 6) Oracle & Internal Service Reliability

Harden the supporting pipeline that feeds validation and observability:

* Router APIs and internal services reliability under sustained load.
* Indexer consistency + freshness, including recovery from missed blocks/events.
* Dashboard feed integrity (no stale/partial views, accurate merging across networks).

#### 7) Infrastructure Redundancy: RPC, Failover, Backups

Make “survives failures” a first-class requirement:

* Multiple RPC endpoints per network + automatic failover behavior.
* Validator backups / fallback paths to prevent single points of failure.
* Validate ops playbooks: restart recovery, resync behavior, and degraded-mode handling.

#### 8) Gas Optimization & Runtime Tuning (Formal Start)

Begin structured profiling and cost reduction:

* Profile **Cortensord** runtime parameters and on-chain execution paths.
* Optimize key L2 contracts/modules:
  * **SessionQueueValidation**
  * **SessionPayment**
  * **Validator** storage layout
* Reduce unnecessary log events and nested calls to minimize gas and execution overhead.

#### 9) Dashboard Expansion: Trust Observability

Expand instrumentation so Phase #2 improvements are measurable:

* Real-time **PoI / PoUW analytics**
* Validator health views (availability, queue latency, drift indicators)
* Endpoint health indicators (RPC/indexer/router feed status) and alert-style visibility

#### 10) L3 Rollup Comparison: Self-Managed vs RaaS (Apple-to-Apple)

A dedicated Phase #2 focus to validate tradeoffs before scaling Testnet-1:

* Benchmark **Testnet-1a (self-managed L3)** vs **Testnet-1 (RaaS L3)** under the same workload and validation requirements.
* Measure:
  * Indexer/validator freshness under load
  * RPC reliability + failover behavior
  * Throughput/latency + telemetry completeness
  * Operational overhead and where cost/runtime tuning matters most
* Goal: a data-backed decision on which L3 approach best supports Testnet → Mainnet progression.

#### 11) Light MCP Integration: Component Assignment Experiments (Nice To Have)

Phase #2 includes early “agent-ready wiring” work via light MCP integration:

* Experiment with **MCP component assignment** (mapping tasks to tools/components cleanly)
* Validate minimal MCP compatibility for task execution + validation loops (without over-scoping full protocol rollout)
* Goal: prove the router can begin acting as an **agent execution surface** where tools/components are assigned deterministically and observed end-to-end.

***

### Key Outcomes

By the end of Phase #2, Cortensor should achieve:

* **Higher-accuracy validation** with more stable PoI/PoUW scoring.
* **Expanded redundancy** with measurable trust improvement and lower variance.
* **Infrastructure robustness** via failover, backups, and reduced service fragility.
* **Early gas + runtime optimizations** reducing cost and execution overhead on L2.
* A **fully instrumented trust layer**, observable in real time through dashboard analytics.

***

### Outcome Statement

Testnet Phase #2 is about making Cortensor’s trust layer **redundant, cost-efficient, and fully observable** — a foundation where validation isn’t just “working,” but **stable under load, resilient under failure, and optimized for scale**.

**Join Cortensor Discord:** <https://discord.gg/cortensor>