Field note · Automation delivery · 10 November 2026

Two human approvals. One payment gate. When may a workflow move?

Imagine a small service business that receives leads through a form, records them in a database, requires two separate human approvals, and pauses service after a payment lapse. A successful form submission is only the beginning. The difficult part is proving that no downstream action escaped before the conditions were satisfied.

What this guide provides: one fictional state contract, an interactive local-only release checker, a test matrix, and a fair comparison with built-in workflow features. No customer information, payment records, or automation credentials are needed. The example and accompanying interactive checker are original Moruneva work, designed to be tested locally.

Boundary: Every checkbox below represents a claim supplied by the visitor. Moruneva cannot verify a real payment, legal approval, identity, or CRM state. The checker never sends a message, unlocks a live workflow, updates an Airtable row, or calls an API.

1. Write the release condition before connecting services

For this sample, the hypothetical business rule is: only prepare a draft after approval A, approval B, a confirmed paid state supported by separately checked evidence, and a stable event reference are all present. Even then, another person must review the draft before dispatch. If a prior action is already queued, hold for replay review rather than creating another. This deliberately uses a stricter standard than “the last node turned green.”

PREPARE_DRAFT =
  approval_A === true
  AND approval_B === true
  AND payment_status === "paid"
  AND payment_evidence_checked === true
  AND prior_action_queued === false
  AND stable_case_and_event_refs_present
  AND live_mode === false

PREPARE_DRAFT does not mean SEND or CHARGE.

Payment evidence must come from the authorized source of truth, not just a free-text field a workflow can set. The correct authority, timing, refund handling, dispute handling and approval identity must be agreed with a real buyer; our local demo makes none of those decisions.

Try a fictional release decision

Start with all conditions unverified. Choose fictional statuses only; the result is a review prompt, not authorization.

NO DECISION YET

Set fictional statuses and click Check the release gate.

2. Eight acceptance cases that should fail closed

Fictional inputExpected outcomeEvidence required in a real system
Approval A missingHold before preparing a draftSigned or attributed approval event
Approval B missingHold; A cannot silently substituteSecond authorized reviewer and timestamp
Payment pendingHoldProvider-side actual settled state
Payment overdueHold, subject to contract grace periodAgreed effective date and exceptions
Payment refundedBlockVerified reversal/refund status
Payment says paid, no independent evidenceHoldSource-of-truth confirmation
Same action already queuedBlock duplicate preparationDurable idempotency key and ledger
All gates appear satisfiedPrepare manual-review draft onlyDistinct human dispatch approval

There are additional difficult cases beyond this table: an approval can be revoked after the draft is prepared; a payment may move from paid to refunded; a webhook may arrive twice; a system could commit an external action but lose the acknowledgment. Those require an explicit reconciliation design, not an optimistic retry. Our standalone module contains 18 offline regression cases, including invalid references, unknown states and approval revocation. These tests are limited to the local decision function.

3. An event ledger is more important than one “approved” checkbox

For a real project, define a stable case ID, a per-delivery event ID, reviewer identity, approval revision, the payment source, evidence timestamp, intended side effect, and an idempotency key. Only authorized parties should see those fields. Keep the human-visible status separate from proof that an external action actually completed. A queue entry means “scheduled for review,” not “delivered.” A successful API response may only mean “accepted by the provider,” not “seen by the customer.”

A simple revision-based approach can invalidate pending drafts when the payment or either approval changes. Before any actual send, recheck the current authorized state and reconcile previously attempted sends. That second recheck is not implemented in the demonstration; it is a production acceptance requirement.

4. What you can do for free first

n8n: Its native Wait logic can pause an execution for a time or external event, and its error workflow can notify on failures. These are powerful building blocks, but a Wait node alone does not establish reviewer authority, payment settlement, uniqueness of side effects or revocation behavior. Review n8n waiting documentation and n8n error handling.

Make: Its incomplete-execution queue and error handlers can help retry or manually resume failing steps. That is useful infrastructure, but you must still decide which failed writes are safe to replay and what constitutes payment and approval truth. See Make error-handling overview and managing incomplete executions.

Spreadsheet-first: If the business has only a few submissions per week and the owner already reviews every one, two explicitly controlled approval columns, a payment reconciliation tab and a manual checklist can be the lowest-risk and lowest-cost approach. Automating before agreeing the rules may create more failure modes than it removes.

5. What a small paid review would have to deliver

A useful paid first step should examine one narrow, authorized, sanitized problem and return an acceptance matrix, event/field mapping, failure and replay table, and a reproducible offline test plan. The buyer should be able to run the same sample and see which failures remain unproven. A proposed fixed fee is only meaningful after confirming the actual scope, whether the native/free solution already covers it, and who legally owns the systems.

Moruneva's indicative option: a small, offline reliability assessment starting around US$150, subject to a mutually agreed scope, lawful contracting and payment arrangements. No live connector, data upload, production modification or guaranteed result is included by default. See the inquiry boundary and ask about a scope. You can also start with the free scope-to-acceptance builder.

6. Reproduce the decision checks

The browser-local decision source, its independent public source and all 18 regression tests, and a one-command offline reproducibility guide are now available to inspect. Our company continuous integration has run the original decision module tests; independent visitors can execute the published copy locally. The interactive checker on this page runs the same pure local decision rule. Its response always reports maySend: false and externalPaymentVerified: false.

Reproducibility boundary: The program is deliberately small and deterministic, so you can inspect its decisions, but no n8n, Make, payment provider or CRM was connected for this article. No external person commissioned or accepted this work. No production compliance or zero-duplicate guarantee is claimed.

References and disclosure

  1. n8n official documentation: waiting in workflows — supports pause/resume mechanisms, not business approval proof.
  2. n8n official documentation: error workflows — explains execution failure handling.
  3. Make official documentation: error handling — source for error routes and incomplete executions.
  4. Make official documentation: managing incomplete executions — source for retries and manual recovery.
  5. Public example of buyer-described Tally/Airtable/n8n approval and payment-state problem (first posted March 2026; used only to inform the research question, no buyer contact or client relationship).

Editorial method: checked primary vendor documentation and a publicly described workflow requirement; constructed a separate fictional contract; wrote testable code; marked unverified security, payment and operational claims. Last substantive review: 10 November 2026.