Skip to content
ONE3.ai

AI Workforce · Module

Performance Arena & Certification

Agent performance measured against your previous process, with earned certification

Planned · Year 1 (W2)Sold as: CoreC-191C-193

Performance Arena & Certification measures each agent against the tenant's previous process baseline and grants trust levels from Listed to Verified to Certified Outcome on evidence. In Year 1, internal scorecards and the Verified badge process are the substitute for the full apps.

"Vs your previous process"

Comparisons between AI and named people create the wrong incentives. The Performance Arena measures an agent against the tenant's previous process baseline for the same task class: cost, quality and speed before the rental, which may have included tools, outsourcing or manual work. No human-versus-AI leaderboards or named-person comparisons are allowed in the UI or exports.

Scorecards

Standard scorecard dimensions carry default weights: outcomes 35%, quality 25%, safety and compliance 20% (any S1 or S2 failure fails the dimension), with efficiency making up the rest. Shadow agreement metrics and scorecards per agent are an internal view in Year 1. A tenant-visible Arena dashboard per agent and squad is planned for W3.

Evidence before claims

Each Verified agent has an evidence pack: segment, channel, sample size, method, confidence, eval results and simulator results. Offline benchmarks on simulator scenarios use significance tests before any claim, and outcome benchmarks come from holdouts linked to the Ledger. Marketing claims need an evidence pack approved by product and legal.

Three trust levels

  • Listed: template spec complete, Eval Pack minimum, persona policy and owner.
  • Verified: template Eval Coverage of at least 85 on allowed skills, simulator pass, disclosure policy, owner and stop triggers. Where live data is missing, the label says "Simulator-verified".
  • Certified Outcome (later): Verified plus holdout-validated lift in several tenants of the target segment and a dispute rate below threshold.

The Certification Center

Planned in full for W3, the Certification Center runs submissions through automated checks (Eval Coverage Score, skill pack security scan, persona policy validity and ledger checks) and human review. Recertification happens on template major versions, at a fixed interval and automatically after incidents. Certification data feeds Store capability cards with level, date and evidence summary. Partner templates are certified under the Marketplace Trust Standard.

Features

What Performance Arena & Certification includes

Process baseline capture

Cost, quality and speed of the previous process per task class.

Shadow agreement scorecards

Per agent; internal view in Year 1.

Standard scorecard dimensions

Outcomes, quality, safety and compliance and efficiency with default weights.

Evidence packs

Segment, channel, sample, method, confidence, eval and simulator results.

Verified badge process

Eval Coverage, simulator pass, disclosure policy, owner and stop triggers.

Listed requirements check

Spec, Eval Pack minimum, persona policy and owner.

Claim approval workflow

Marketing claims need product and legal approval of evidence.

Offline benchmarks

Simulator scenarios with significance tests.

Outcome benchmarks

Holdout results linked to the Ledger.

Certification Center

Automated checks and human review.

Recertification

On major versions, at intervals and after incidents.

Certified Outcome level

Holdout-validated lift across tenants and low dispute rate.

Arena dashboard

"Vs your previous process" per agent and squad.

Framing rules

No human-vs-AI leaderboards or named-person comparisons.

Workflow

Earn a certification level

  1. 1

    Complete the template spec and Eval Pack

  2. 2

    Run automated checks

  3. 3

    Pass simulator scenarios

  4. 4

    Human review approves the level

  5. 5

    Evidence summary appears on the capability card

  6. 6

    Recertify on major versions or after incidents

AI

AI skills and Digital Employees

Evidence pack assembly

Compiles results with sample sizes and methods.

Benchmark significance testing

Tests simulator results before claims.

Digital Employees that use this module

  • AI Agent Governance Auditor

They act through the same governed Actions as people; risky steps wait for approval.

Frequently asked questions

Are agents compared with my employees?

No. Agents are compared with your previous process baseline, never with named people.

What does "Simulator-verified" mean?

The agent passed simulator scenarios but live evidence is not yet available.

What is in Year 1?

Internal scorecards and the Verified badge process. The full Arena and Certification Center are planned for W3.

See Performance Arena & Certification in your workflow

Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.