RomeoApps n8n / performance evidence Reliability bench →

RomeoApps-owned portfolio proof / synthetic only

Find where spend stops earning its place.

A bounded n8n analyzer for export-shaped advertising metrics: validate the contract, calculate ACOS, expose no-sales or high-ACOS leakage, and rank the largest rows for human review.

This is not an Amazon API integration, a client result, or an automatic budget action. The inactive fixture stops at a decision-grade human-review output.

09 nodes in the export 06 executable nodes 05 deterministic cases 00 external side effects

The useful seam

The proof is the stop line.

Performance workflows become expensive when a plausible number silently turns into a budget decision. This fixture makes the contract and the handoff visible before a connector, credential, or production write exists.

01 / validate

Reject bad metric shape.

Every row needs a non-empty SKU, finite non-negative measures, and an explicit ad status. An invalid contract is blocked, not summarized.

02 / rank

Show the expensive rows.

No-sales and high-ACOS rows become a ranked waste list with a Pareto inclusion flag. The output remains a review queue.

03 / hold

Keep action human-owned.

No ad pause, repricing, CRM write, Slack alert, or automatic recommendation is connected. The next move is an acceptance decision.

Synthetic decision register

Five cases. Four honest outcomes.

The runtime receipt binds these deterministic fixtures to the same inactive workflow export. No row below is customer data.

fixturedecisionwhy it stops
healthy-mixvalid metricsready for human reviewwithin the configured waste threshold
waste-clusterno-sales + high ACOStriage requiredwaste share crosses the review threshold
below-minimum-spendsmall sampleready for human reviewinsufficient spend for an automatic conclusion
invalid-metricnegative spendblocked invalid contractinput validation fails before analysis
zero-spendno denominatormanual review requiredthere is no spend basis for ACOS

Production boundary

A real engagement starts after the fixture.

The proof is deliberately narrower than a client build. It shows the analysis seam and its controls, then names the evidence needed for a paid first unit.

Inputapproved redacted export

Agree the fields, currency, attribution window, and missing-data behavior.

Accesssafe source path

Keep Amazon, catalog, database, and messaging credentials outside the fixture until written scope and protected payment exist.

Acceptancenamed reviewer

Define the exact output, failure cases, and human action boundary before any connector is introduced.

Handofftested, inspectable result

Return the export, run receipt, decision register, and rollback or next-step note—not a claim that a client result already exists.

Public evidence

Inspect the receipt before you send a brief.

The public receipt exposes the runtime version, workflow hash, case counts, decision counts, and zero-side-effect controls. The exact inactive export is kept in the RomeoApps portfolio packet and can be shared through a written fit route.

Truthful boundary

What is proven—and what is not.

Proven here
  • Deterministic import and execution in n8n 2.33.7.
  • Metric validation, ACOS calculation, waste ranking, and decision output.
  • Five cases pass with zero credentials, network actions, and external side effects.
Not claimed here
  • No Amazon, Keepa, ClickUp, Claude, Supabase, or Slack connector is active.
  • No customer result, savings figure, account access, or production mutation is claimed.
  • No payment, client selection, delivery, or revenue is implied by this page.