These are independent demonstrations, not client projects. The data collection sample uses a public practice website; other records are fictional. No demo connects to your tools.
Executed school example · 4 Oct
A pupil register. The right meal count.
Two staff roles, attendance corrections without duplicate meals and a dated CSV export. Twenty-nine local checks, data retained after a server restart, four desktop/mobile captures and downloadable code. Three fictional pupils; no customer school, child data, parent portal or payment connected.
Inspect the recorded journeys and checks · French page →Executed PostgreSQL access example · 4 Oct
Salary data belongs to the right roles
Thirty-nine SQL checks verify allowed actions, direct-query refusals and isolation between two fictional companies. Switch between the recorded employee, manager and admin results, then inspect the policies and source.
Local PostgreSQL via PGlite, with an injected test identity. Hosted Supabase Auth, JWTs and a React-to-Supabase connection are not tested by this example. A separate negative control exposes the salaries and makes the verifier fail.
Compare access results and inspect the checks · French page →Executed visual-agent example · 3 Oct
Two tabs, independently checked results
Six real screenshot-based model decisions save one fictional draft and one asset choice. Both pages are reloaded and compared with the stored state. Captures, source and the actual inference cost are reviewable.
Two controlled local pages; no customer, production authentication or arbitrary website automation tested. Sixteen offline checks and two browser refusals with zero writes.
Inspect the actions, saved results and code · French page →Executed supplier-file example · 3 Oct
A price list, a reviewable product CSV
Three fictional rows prepared, five rejected and one identical duplicate removed. The Python source, files and seventeen local checks are downloadable, including a separate 1,000-row fixture.
No Magento shop, actual import, customer data or live product publication tested. Shop settings still require a real export and Check Data on a compatible copy.
Inspect the CSV and recorded checks · French page →Executed WordPress / Elementor example · 3 Oct
New content, the same editable template
The title, description and button are saved in real Elementor widgets. Container settings, widget order and a separate witness page stay intact. The JSON template is imported back into the library.
Twenty-one WordPress checks and six desktop/mobile views. Local fictional example; no customer site, visual editor session, form or payment connected.
Inspect the rendered pages, template and recorded checks →Executed WordPress example · 2 Oct
300 posts exported, imported and found again
Fourteen XML/CSV checks and seven comparisons against a real local WordPress import and native re-export. Synthetic titles, slugs, dates and HTML are retained.
No customer Rails database, image download or Squarespace import. The destination still needs a sample import.
View the source, results and PDF →Local RSS and messaging example · 2 Oct
Confirm a notification before retrying
Twelve checks cover persistence, unsubscribe, rate limiting and uncertain responses. Three calls to a simulated API; one confirmed simulated message.
No live Telegram delivery, official legal feed or production deployment.
Inspect the checks, source and PDF →Local HTTP example · 2 Oct
Check access, indexing directives and old URLs
Fourteen checks on fictional pages, with the original URL inventory retained after corrections. The example separates noindex, canonical, redirects, internal links and an HTTP200/403 difference between two invented headers.
No customer WordPress, verified crawler IP or Search Console. A 200 response does not guarantee indexing.
View the results, source and PDF (French) →Executed MariaDB example · 2 Oct
An index, 280,000 rows and the same 50 results
A real local database with synthetic orders. The index is added and removed; result hashes remain identical across all three phases. Execution plans, timing method and six checks are available.
CLI timings include client startup; ANALYZE measures one execution per phase.
Inspect the measurements, code and PDF (French) →
Personal web application · 2 Oct
A login, persistent tasks and a downloadable PDF
Two fictional accounts, a completed task and an invoice retained after restarting the local server. The other account cannot read or change the work. Sixteen checks and actual desktop/mobile screenshots are available.
Node.js and SQLite example, without a customer Figma, production hosting or connected payment.
View the captured workflow, results and source →
Executed browser example · 2 Oct
Confirm the submission before resuming
A real browser handles four fictional rows: one verified confirmation, one invalid input, one deferred request and one uncertain response. Reloading the saved checkpoint produces no repeated submissions.
Local form and synthetic data, twelve assertions. No client login, spreadsheet import or thousands-row load test is claimed.
Inspect the recorded results (French) and download the code →
Public data → CSV · USD 35 starter
A dataset with source links and checks
One public website, up to 1,000 available records and five agreed fields. Receive a UTF-8 CSV and Excel workbook, source URLs, a short data dictionary and count checks within three business days after scope confirmation.
Inspect 1,000 real extracted rows from 50 pages of a practice website. No client project or market insight is claimed.
See the scope and download the example →Inspect the price and new-product report example →
Website QA · fixed USD 20
Reproducible website bug report
Three pages, one user journey, desktop Chrome and an emulated mobile viewport. Receive a report with steps, screenshots, expected and actual behavior, or a clear record of checks if no issue appears.
Two business days after scope and access are agreed.
See the scope and sample report →
PDF generated · 24 Sep
Nested JSON to a readable project brief
A two-page PDF from fictional nested input, with a browser editor, downloadable renderer and six test groups. Inspect missing values, arrays and invalid-input rejection.
Executed with local Chromium. No n8n Cloud or email delivery is claimed.
Open the PDF and try the renderer →
Executed HTTP + SQLite · 21 Sep
Repeated webhook, one stored receipt
Eight concurrent requests, one creation and seven replays. Inspect payload conflicts, input rejection and a clean server restart, then download and run the tests yourself.
A local Python component with fictional records. No n8n execution, external side effect or client integration is claimed.
Inspect the executed webhook tests →
Written review · USD 35
Independent n8n workflow review
Send one redacted workflow JSON of up to 25 nodes, its expected result and one input/output example. Within two business days, I return prioritized written findings on logic, error paths, retries, idempotency, secrets, expressions, SQL and maintainability, plus one clarification round.
Asynchronous review only: no production access, credentials, video call or repair is included. Verifiable findings are separated from recommendations.
Request the USD 35 review →Read a sample written review → · Check your export locally →
RFQ PDF to Excel
Inspect an actual local n8n run: a synthetic quotation request becomes an Excel draft, with approved AED prices and unresolved rows left for review. Input PDF, workbook, source and failure evidence are downloadable.
US$75 feasibility test · two business days after sample and scope approval. One text-PDF layout, up to two pages / 20 rows.
See the starter scope and request a test →
Explore the executed RFQ demonstration →
Fixed starter · USD 75
Blocked n8n workflow diagnostic
Send one redacted workflow JSON, one reproducible failed execution and the expected result. Within two business days, I return the reproduced fault, the smallest proposed repair, one before/after test and a fixed quote for any broader work. If I cannot reproduce the issue, I document what is missing instead of guessing.
No production access or live customer data is needed for this first pass. One workflow, one failing path, one written revision.
Request the USD 75 diagnostic →
Appointment app
Create, search and export fictional records.
Try the appointment demo in English →
Form to CRM
Create a contact, update it, then simulate an outage and check the reported result.
Try the CRM demo in English →
Failed execution triage
Sort a synthetic batch into a review queue, inspect each decision and download a report. No operation is retried.
Try the execution triage demo →
Signed requests and uncertain responses
Inspect a recorded local test run: signatures, repeated requests and a lost response. Download the code to reproduce seven checks.
Explore the API test evidence →
Article to WordPress draft
Check structured article fields and taxonomy IDs, then download a draft payload. This example does not connect to WordPress.
Try the WordPress draft preparation →