Check evidence and uncertainty before an agent selects a service.
Identify the exact provider endpoint, version, and task. Keep the card attached to the selection decision so later evaluation uses the same context.
curl https://api.scrubmytext.com/v1/quality/cards/get -H "Content-Type: application/json" -d '{"target_type":"mcp_server","provider_domain":"api.example.com","canonical_endpoint":"https://api.example.com/mcp","version":"1.2.0","task_category":"structured-text","max_latency_ms":2000}'Decision rule
Do not treat the card as a pass/fail seal. When status or confidence.status is insufficient_evidence, fall back to your own test, approved provider list, or human review. Otherwise compare evidence class, task identity, sample size, independent sources, freshness, correction state, and each dimension—not one number.
Add your opt-in observation
Starter includes TrustMyChoice writes, comparisons, and all 49 REST/MCP tools.
curl https://api.scrubmytext.com/v1/quality/observations/record -H "Authorization: Bearer smt_live_YOUR_KEY" -H "Content-Type: application/json" -d '{"target_type":"mcp_server","provider_domain":"api.example.com","canonical_endpoint":"https://api.example.com/mcp","version":"1.2.0","task_category":"structured-text","max_latency_ms":2000,"observation_key":"run-2026-08-18-001","outcome_success":true,"reliability":true,"latency_ms":840}'Card required. Unless canceled before the trial ends, Starter renews automatically at $15/month. No automatic overages.
What gets stored
Exact public target/version/task identity, a hash of your retry key, structured success/reliability/latency/price metrics, evidence class, timestamps, and your hashed paid-account identity for source counting. Do not send prompts, responses, credentials, customer content, or free-form review text; the API rejects fields outside the contract.