Test mode vs live mode

The platform encodes the environment inside the credential itself. There is no separate sandbox.* host or --test flag.

Credential prefixes

Prefix	Mode	Where to use
dv_sk_test_…	test	Server side: local dev, CI, integration tests, demos
dv_sk_live_…	live	Server side: production traffic
dv_pk_test_… / dv_pk_live_…	test / live	Public keys, safe to embed in an app; identify the tenant and mode, grant nothing by themselves
dv_tok_test_… / dv_tok_live_…	test / live	Session tokens minted by POST /v1/sessions with a secret key of the same mode

The mode of any credential is the third segment of the string, so a client can read it without decoding anything:

final mode = token.split('_')[2];   // 'test' or 'live'

client, _ := ekyc.NewClient(ekyc.Config{APIKey: "dv_sk_test_..."})
client.Mode  // "test"

Behavioural differences

The same endpoints serve both modes, and the same models score both. The platform treats them differently in these places:

• Data separation. Every stored result carries its mode, and the dashboard's live ⇄ test toggle shows one at a time. A test-mode token cannot read live-mode results and vice versa.
• Usage and billing. Usage is metered per mode and shown separately on the dashboard's Usage page. Plan quotas and invoices are based on live usage.
• Webhooks. An endpoint is registered for one mode; test events go only to endpoints registered for test mode.
• Rate limits. A test secret key is limited to 25 requests per second (burst 50) against 100 per second (burst 200) for a live key. Session tokens are limited to 10 requests per minute (burst 20) in both modes.
• Liveness replay window. A frame that was already scored is refused for 24 hours under a live key but only 60 seconds under a test key, so integration tests can reuse a fixture image. See Liveness checks.
• Component scores. Liveness responses in test mode always include the per-model components breakdown; in live mode they do so only when ops enable it for your tenant.
• Storage. Test-mode results follow the same storage policy and retention settings as live; there is no separate test-mode purge.
• Decision thresholds apply to both modes. The dashboard's Decision thresholds page has no per-mode split.

Mistakes to avoid

• Never log a dv_sk_* key. Logging frameworks often capture full HTTP headers. Configure scrubbing.
• Don't use a dv_sk_live_… in development. Test keys exist so you can iterate without polluting your live data.
• Don't accept a dv_tok_live_… in a debug build. Read the mode from the prefix and refuse live tokens where you expect test ones.