Test mode & magic inputs

Every organization gets test mode from day one, before activation. Requests made with a test key (opis_test_…) run the complete workflow — statuses, confirmation gate, expiry timers, webhooks, usage — but stages are executed by a simulator that returns deterministic fixtures. Nothing reaches the live engine or any AI provider.

	Test mode	Live mode
Key	`opis_test_…`	`opis_live_…`
Results	Deterministic fixtures	The production pipeline
Stage time	2–6 seconds each (simulated)	Real
Billing, quotas	None (separate free meters)	Yes
Rate limits	Yes	Yes
Webhooks	To test endpoints only	To live endpoints only
Confirmation `expires_after`	`PT10S` to `P7D`	`PT5M` to `P7D`

Test and live data are fully separate: a test key gets 404 for live resources and vice versa.

Magic inputs

Put a magic value in the meal analysis's description (or external_id) to choose a scenario. Any other value behaves like test:succeeded.

Value	Simulated behaviour
`test:succeeded`	Default. A normal identification and nutrition result
`test:no_food`	No food detected (`no_food_detected`). Automatic mode succeeds with an empty result; manual mode waits for confirmation with an empty list
`test:unresolved_item`	One item cannot be matched: `match.source: "none"`, null nutrients, `item_unresolved` flag
`test:failed_identify`	The identification stage fails → `failed`, `error.stage: "identification"`
`test:failed_nutrition`	The nutrition stage fails → `failed`, `error.stage: "nutrition"`
`test:slow`	Each stage takes about 40 seconds — exercise your timeouts and `Prefer: wait` fallbacks
`test:invalid_image`	The image is rejected → `failed` with `error.code: "invalid_image"`
`test:low_confidence`	Items come back with low confidence and the `low_identification_confidence` flag

Shell

curl https://platform.opis.health/v1/meal-analyses \
  -H "Authorization: Bearer $OPIS_TEST_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
        "images": [{ "file_id": "file_01j9…" }],
        "description": "test:failed_nutrition",
        "confirmation": { "mode": "manual", "expires_after": "PT10S", "on_expiry": "auto_confirm" }
      }'

Test mode still validates your requests exactly like live mode — a bad body is still a 422, an image still has to be a real JPEG, PNG or WebP — so a green test suite means your requests are well formed.

Testing confirmation and expiry

Manual flow. Create with "confirmation": { "mode": "manual" }, wait for requires_confirmation, then confirm with accept, edit or reject — exactly as in production.
Expiry in CI. Use "expires_after": "PT10S" with on_expiry set to cancel or auto_confirm. Ten seconds later the analysis is canceled (confirmation_expired) or moves on with method: "auto_on_expiry".
From the console. The Analyses view lists test analyses waiting for confirmation with their deadlines, and offers Accept / Edit / Reject buttons so you can drive the flow by hand.

Going live

Create a live key in the same project once your organization is activated. Nothing else changes: endpoints, payloads, webhooks and error codes are identical. Remember to register live webhook endpoints too.