Questions

No. Weights are fixed constants in a versioned model, never entered by a user, and identical for every customer. Customers cannot override scores, and neither can we. Recompute re-runs the algorithm; there is no manual score input anywhere in the product.

Operational metrics for the AI agents you run: response time, override rate, throughput against benchmark, customer satisfaction, uptime, tenure, data volume, and supported KPI events tied to those agents. Connected sources are read-only. Uploaded source records are retained, so remove unnecessary personal data before sharing.

First scores depend on a ready connection and enough valid events. We aim to make initial scores available within 48 hours once those requirements are met. Scores below 60 percent confidence are marked provisional, and agents with fewer than 50 recorded events show as collecting data rather than an unreliable score.

Operational performance and the organization running it move differently. AgentScore, 0 to 1000 per agent, measures the agent. ImplementationScore, 0 to 100 per company, measures the deployment context: workflow integration, governance, data foundation. The two measures distinguish agent performance from the conditions supporting it.

Sales, field services, and manufacturing ship with bespoke operational benchmarks. The other seven verticals score against the broader operations corpus through the same algorithm, with confidence flagged accordingly.