Catalyst-Q

Vertical Agent Best Practices

The product rules for selling and operating Catalyst-Q proof-packet workflows in 2026.

UX patterns

  • Make the root domain the workbench, not a marketing landing page.
  • Expose agent selection, workflow, tier, prompt, simulation budget, output, and evidence queue in one screen.
  • Show model route, .rain event id, data freshness, approval state, Catalyst-Q shots, and baseline delta before recommendation prose.
  • Keep Browser Run screenshots, Live View handoff, and session recordings attached to the same run ledger as solver/model output.
  • Use release radar and research watch panels so operators can see model, Cloudflare, benchmark, and pricing updates without leaving the product.

Telemetry patterns

  • Emit one event per vertical run with tenant, agent id, workflow, tier, model route, success/error, Catalyst-Q shots, .rain event id, Browser Run flag, and freshness state.
  • Separate product analytics, billing meters, eval feedback, and audit evidence into distinct telemetry lanes.
  • Use Workers traces for request health, AI Gateway logs/evals for model behavior, Analytics Engine for customer usage analytics, and Pipelines for durable lakehouse ingestion.
  • Summarize telemetry for users as ready, stale data, needs approval, or proof complete.

Pricing patterns

  • Use capped free proof for trust, not unlimited freemium.
  • Sell paid pilots where a baseline and approval workflow can prove value; Freight starts at $15,000-$35,000 for 4-6 weeks.
  • Move production to hybrid pricing: base subscription plus usage meters plus optional verified outcome upside.
  • Never sell regulated aviation, grid, or ATC workflows as autonomous production control until the safety case and approvals are mature.