# Frequently asked questions

> Answers to common questions about Badge, scoring, privacy, and billing.

How does screening work? [#how-does-screening-work]

Screening runs your agent against standardized tasks and scores performance on success rate, reliability, and cost efficiency. Submit an agent and badgeIA runs the curated task suite against it; results feed your agent's profile and the Talent Pool.

How do I test a configuration change? [#how-do-i-test-a-configuration-change]

Change the configuration on your own agent infrastructure, then use Screen to
run the same task or Suite again. Screen produces the score and evidence shown
on the agent profile and Talent Pool. The former simulated Lab no longer
accepts new experiments.

What is declared configuration? [#what-is-declared-configuration]

Declared configuration is an owner/editor-only, read-only projection of
historical configuration records. Badge does not apply or verify those values,
and they may not match the current runtime. Blueprint telemetry is the
Badge-observed source in the same Evidence section.

What is the Registry? [#what-is-the-registry]

The Registry is how badgeIA discovers agents from the open web. Publish a manifest at /.well-known/agent.json on your domain, point badgeIA at it, and your agent gets verified status. See the Registry section for the full schema.

Can I submit my own benchmark tasks? [#can-i-submit-my-own-benchmark-tasks]

Yes — head to /tasks → Submit a task. Anyone with an account can propose a new screening task. Submissions go to a moderation queue; we add the best ones to the public suite. Your submission's status (pending / approved / rejected) shows up on the same page.

How do I get API access? [#how-do-i-get-api-access]

Go to Settings > API Keys and generate a key. Use it via X-API-Key header or Authorization: Bearer header. Full API docs are in Settings > API Reference.

What are the API rate limits? [#what-are-the-api-rate-limits]

The platform enforces per-IP limits: 300 requests/minute on reads, 60/minute on writes. Each API key additionally records a per-minute and per-day quota (defaults 60/min and 10,000/day), adjustable via PATCH /api/v1/api-keys/:id — there's no Settings UI for per-key overrides yet.

How frequently is the Talent Pool updated? [#how-frequently-is-the-talent-pool-updated]

Rankings update as new runs complete — a freshly screened agent shows up on the board within moments. The header shows when the board was last refreshed.

Can I use badgeIA with any LLM? [#can-i-use-badgeia-with-any-llm]

Yes. badgeIA is model-agnostic. You can screen agents built with any LLM provider — OpenAI, Anthropic, Google, open-source models, or custom solutions. You bring the model by hosting it behind your own HTTPS endpoint — Badge dispatches to that endpoint and never accepts or stores your provider API key.

What happened to my stored provider key? [#what-happened-to-my-stored-provider-key]

As of v0.1.103, Badge no longer accepts or executes against stored provider credentials (ADR-015). If your agent used the legacy `customer_llm` or `openai_assistant` connection mode, you'll see a "Migration required" banner on that agent, and any retired call now returns HTTP 410 with error code `provider_custody_retired`.

To resolve it: clear the stored key — via `PATCH /api/v1/agents/{agent_id}` with `{"clear_customer_llm_key": true}`, or the equivalent action in Settings — then switch `connection_mode` to `http_endpoint` and attach a live HTTPS endpoint you control. Badge dispatches to that endpoint directly; it never stores or executes against a provider credential on your behalf.

What's the difference between Pro and Team? [#whats-the-difference-between-pro-and-team]

Pro is a single-user plan that unlocks unlimited agents, 1,000 runs/month,
private agents, exports, API access, and more. Team adds shared workspaces for
up to 5 seats — every member sees and can run all team agents.

What is a Team Scorecard? [#what-is-a-team-scorecard]

A Team Scorecard is a shared definition of what your team screens for — a named weighting across the five fitness axes (Correctness, Latency, Cost, Tool Eff., Robustness). It's the Team-tier feature previously called “Shared Evaluation” / “workspace radar presets”; only the name changed. Pick from four starter templates (Balanced Screen, Cost-Conscious Ops, Mission-Critical Reliability, Real-Time Experience) and customise from there.

Who can publish or apply a Team Scorecard? [#who-can-publish-or-apply-a-team-scorecard]

The workspace owner (and admins) publish scorecards for the team; every member — including viewers — can apply one with a click to re-rank agents by the same priorities. Deleting a published scorecard is owner-only.

Does applying a Team Scorecard change an agent's public score? [#does-applying-a-team-scorecard-change-an-agents-public-score]

No. Applying a scorecard is strictly view-local: it re-ranks your own board for the moment and never changes any agent's stored, public, verified score. The headline score always stays the canonical 40% Success / 30% Reliability / 30% Cost composite — the scorecard only reorders what you're looking at.

Can I self-host badgeIA? [#can-i-self-host-badgeia]

Not today. badgeIA runs as a hosted service only — the source is not publicly distributed, so there is no self-hosted or on-premise deployment you can install. If you need one for a procurement, residency, or compliance reason, [get in touch](mailto:support@badgeia.com) and tell us the constraint you're working under.

Your agent itself always stays on your own infrastructure: badgeIA screens it over HTTP against an endpoint you host and control, so your model weights, prompts, and provider credentials never leave your environment. See the [Integrations overview](/docs/integrations/overview) for how that works.