> ## Documentation Index > > Fetch the complete documentation index at: https://lithi.ai/llms.txt > > Use this file to discover all available pages before exploring further. --- title: Move from the sandbox to live work description: >- Carry what you learned in the sandbox into one small live job. You start a fresh account and quote context, because sandbox identifiers and sample receipts are never reused. route: /docs/compute/sandbox/live-transition page_id: docs-compute-sandbox-live-transition page_type: howTo content_layer: guided documentation surface_profile: guided_docs audience: People moving from a sandbox trial to their first real Compute job voice: D reader_question: What do I need in place to move from a sandbox run to real Compute work? primary_action: Run your first small live test source_locale: en-US source_status: APPROVED source_version: docs-compute-sandbox-live-transition-en-us-2026-09-05-v1 last_updated: '2026-09-05' robots: index claim_registry_pinned: true claim_registry_resync: "npm run governed:tsx -- scripts/build-compute-public-claim-registry.ts --write" --- ## What the sandbox did and did not give you The sandbox gave you the shape of a run. You know what a quote states, what approval means, how a job reports its state, and how to read a bundle. That is real learning, and it carries over. It gave you no account, no credentials and no standing to run anything. Live work starts from a fresh context, on purpose. So start there. Create or open your Compute account and confirm your member role. Then mint the work key you will submit with. A work key carries work scopes only; it can never administer finance, destinations or identities. Nothing you did in the sandbox pre-authorizes any of this, and nothing is charged by setting it up. ## Register the real input and destination Register the source your real input lives in, and test it. Register the destination you want results written to, and test that too. Your synthetic input has no live equivalent. A real input is snapshotted before work begins, so the run is pinned to exact bytes rather than to whatever the source holds later. ## Ask for a new quote every time Request a live quote for the real workflow and the real input. Sandbox identifiers are never reused, and a live operation will refuse them: - A sandbox job identifier cannot be resumed, downloaded or billed live. - A mock quote cannot be approved live. It was never signed with any authority. - A sample receipt is not evidence about a live run and will not be accepted as one. A live quote is signed, states a maximum charge, and carries an expiry. It authorizes nothing on its own. ## Approve the work as a separate act Read the quote, then approve it deliberately. Approval is a distinct decision, never an automatic consequence of asking for a quote. In an organization, the approval may need someone who holds spend authority. Set that up before you need it, so your first live job is not blocked at the last step. ## Treat sandbox success as partial evidence A clean sandbox run proves you can drive the lifecycle. It does not prove your live job will be admitted. Four things the sandbox could not test for you: 1. Whether your real input passes validation and item-identity checks. 2. Whether your real schema matches what the workflow declares. 3. Whether your account and spend authority allow this work. 4. Whether capacity is available for the deadline you set. Each of those is checked live, at quote or at admission, and each refuses with its own reason and next action. ## If the live quote or admission refuses A refusal is not a failed run. Nothing was admitted, so nothing was charged. Read the reason and take the stated next action. If the quote was signed and something changed before you submitted, admission returns `REPLAN_REQUIRED`; ask for a new quote and submit again. See [how to read a quote](/docs/compute/quotes) if the maximum charge or the expiry is unclear. ## Run one small job first Pick one bounded task with an outcome you can check by hand, exactly as the sandbox set was bounded at eight rows. Compare the real result against what you expected before you scale up. The next page walks that through: [run your first small live test](/docs/compute/first-test).