Receipt schema

Every run on Lab produces a receipt: a JSON document at GET /results/:id.json (canonical, machine-readable) with a human viewer at GET /results/:id. Successful runs include full per-step data. Failed or aborted runs include the error and reason; per-step detail may be partial or empty. See Failures & step data.

Top-level fields

Field	Type	What it is
`id`	string	Unique ID (same as the `resultId` and receipt paths)
`type`	string	How the result was created: `sandbox`, `kv`, `chain`, `generate`, `spawn`, or `external`
`createdAt`	string	When it ran (ISO 8601)
`request`	object	What was sent in (code, permissions, prompt, etc.)
`outcome`	object	What happened — `ok: true` with a `result`, or `ok: false` with `error` and `reason`
`timing`	object	How long it took (`totalMs`, and for AI-generated runs: `generateMs`, `runMs`)
`generated`	string?	The code the AI wrote (only for `/run/generate`)
`steps`	array?	Per-step details (only for `/run/chain`)
`receipt`	object?	External work receipt details (only for `/receipts`)
`lineage`	object?	Parent/supersedes links for continued work

The `request` field (varies by type)

sandbox / kv: { body, capabilities? } — the code and permissions
chain: { steps: [{ body, capabilities, name?, input? }] } — each step in the pipeline
generate: { prompt, capabilities } — what you asked the AI to build
spawn: { body, capabilities, depth? } — the code and nesting limit
external: { source, action, input?, output?, capabilities?, replay?, evidence? } — work that happened through an MCP, browser, task runner, or other external system

External receipts

External receipts use type: "external" and include a receipt object:

Field	Type	What it is
`source`	string	MCP server, browser runner, task runner, or external system
`action`	string	Tool/action name
`capabilities`	string[]	Authority used, such as `cf.workers.read` or `browser.click`
`input`	any?	Intent, arguments, or pre-state
`output`	any?	Result, post-state, or observation
`replay`	object	Replay posture: inspect only, sandbox re-run, live re-run with approval, or continue from here
`evidence`	any?	Logs, screenshots, artifact references, before/after data

Per-step details (pipelines only)

When you run a pipeline (/run/chain), the steps array contains one entry per step:

Field	Type	What it is
`step`	number	Step index (starts at 0)
`name`	string?	Name you gave this step
`body`	string?	The code that ran
`capabilities`	string[]	Permissions this step had
`input`	any	What this step received
`output`	any	What this step returned
`ms`	number	How long this step took (milliseconds)

Getting the result

Every run response includes a resultId. Use it to fetch the receipt later:

GET /results/:id.json — canonical machine-readable receipt JSON
GET /results/:id — human viewer on the public app