# Machine Orchestration Receipt Template

Use this after any AI-agent run that touches a real machine, browser session, local file, SaaS app, customer record, finance workflow, internal tool, or external system.

Canonical page: https://jarvislandingdeploy.vercel.app/machine-orchestration-receipt-template.md
Related post: https://jarvislandingdeploy.vercel.app/blog/openclaw-windows-node-machine-orchestration.html
Related checklist: https://jarvislandingdeploy.vercel.app/machine-orchestration-readiness-checklist.md
Filled example: https://jarvislandingdeploy.vercel.app/machine-orchestration-sample-receipt.md

## Why this exists

If an agent operates a machine, the buyer needs more than “task completed.” They need an inspectable record of:

- What the agent was allowed to do.
- What it actually did.
- Which sources it trusted.
- What changed externally.
- What remains unresolved.
- How to undo or compensate if the result is wrong.

Copy this template into each production pilot run. Fill it out automatically where possible; require human review for high-risk sections.

---

## 1. Run identity

**Run ID:**  
**Date/time:**  
**Requesting human / owner:**  
**Agent identity:**  
**Subagents involved:**  
**Machine / node / environment:**  
**Workflow name:**  
**Related ticket / task / contract:**  

## 2. Delegated authority

**Approved objective:**  

**Allowed actions:**
- [ ] Read / inspect
- [ ] Draft
- [ ] Update internal state
- [ ] Update external system
- [ ] Submit / send externally
- [ ] Spend / purchase / commit funds
- [ ] Destructive change

**Explicitly forbidden actions:**
- 
- 
- 

**Permission expiry:**  
**Approval reference:**  
**Human checkpoint required before:**  

## 3. Source context

List every source that materially influenced the action.

| Source | Used for | Freshness checked? | Conflict rule | Link / location |
|---|---|---:|---|---|
|  |  | Yes / No |  |  |
|  |  | Yes / No |  |  |
|  |  | Yes / No |  |  |

**Facts treated as canonical:**
- 
- 
- 

**Facts treated as uncertain / stale:**
- 
- 
- 

## 4. Action log

Record each meaningful action, not every low-level click.

| Time | Actor | App / system | Action class | What changed | Result | Evidence |
|---|---|---|---|---|---|---|
|  |  |  | Read / Draft / Update / Send / Delete / Spend |  | Success / Partial / Failed / Queued |  |
|  |  |  | Read / Draft / Update / Send / Delete / Spend |  | Success / Partial / Failed / Queued |  |
|  |  |  | Read / Draft / Update / Send / Delete / Spend |  | Success / Partial / Failed / Queued |  |

## 5. External changes

Only include actions that changed something outside the agent’s private scratchpad.

| System | Record / file / object | Before | After | Reversible? | Rollback owner |
|---|---|---|---|---:|---|
|  |  |  |  | Yes / No / Compensatable |  |
|  |  |  |  | Yes / No / Compensatable |  |
|  |  |  |  | Yes / No / Compensatable |  |

## 6. Rollback / compensation plan

**Direct rollback available:** Yes / No  
**Rollback command / steps:**
1. 
2. 
3. 

**If direct rollback is not available, compensating action:**
1. 
2. 
3. 

**Irreversible actions taken:**
- 
- 

**Human approval captured for irreversible actions:** Yes / No / Not applicable

## 7. Exceptions and unresolved risk

| Exception | Trigger | Current state | Proposed next step | Owner | SLA |
|---|---|---|---|---|---|
|  | Source conflict / stale session / unexpected UI / partial failure / policy boundary |  |  |  |  |
|  | Source conflict / stale session / unexpected UI / partial failure / policy boundary |  |  |  |  |

**Do not proceed until reviewed:**
- [ ] Source conflict affects final output
- [ ] External state changed partially
- [ ] Approval is missing or ambiguous
- [ ] Rollback path is unknown
- [ ] Agent exceeded or nearly exceeded scope

## 8. Memory writes

What should survive after this run?

| Memory item | Type | Source | Scope | Expiry | Replaces / updates |
|---|---|---|---|---|---|
|  | Observation / Approved fact / Durable rule / Preference / Incident |  | Session / Workflow / Global |  |  |
|  | Observation / Approved fact / Durable rule / Preference / Incident |  | Session / Workflow / Global |  |  |

**Memory writes that require human approval before becoming durable:**
- 
- 

## 9. Final verdict

Choose one:

- [ ] **Clean success** — completed inside scope, receipts complete, no open exceptions.
- [ ] **Success with follow-up** — completed, but exceptions or memory writes need review.
- [ ] **Partial success** — some external state changed; human review required.
- [ ] **Failed safely** — no unsafe external change; retry needs a new contract.
- [ ] **Incident** — scope, safety, or external-state risk requires postmortem.

**One-sentence summary:**  

**Recommended next action:**  

---

## Buyer-facing use

Use this receipt as the handoff artifact for a pilot. A stakeholder should be able to read it and answer:

1. Did the agent stay inside delegated authority?
2. What external state changed?
3. What evidence supports the outcome?
4. What still needs human review?
5. Can we safely repeat or scale this workflow?

If the answer to any of those is unclear, the workflow is not production-ready yet.

## Related artifacts

- Machine Orchestration Readiness Checklist: https://jarvislandingdeploy.vercel.app/machine-orchestration-readiness-checklist.md
- Machine Orchestration Sample Receipt: https://jarvislandingdeploy.vercel.app/machine-orchestration-sample-receipt.md
- Dynamic Workflow Contract Template: https://jarvislandingdeploy.vercel.app/dynamic-workflow-contract-template.md
- Start Here product router: https://jarvislandingdeploy.vercel.app/start-here.html
- Custom Agent Audit Mini-Report: https://jarvislandingdeploy.vercel.app/products/custom-agent-audit-mini-report.html