<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Uvilno Engineering]]></title><description><![CDATA[Uvilno Engineering]]></description><link>https://uvilno.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 16:15:34 GMT</lastBuildDate><atom:link href="https://uvilno.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[What a booking confirmation actually proves]]></title><description><![CDATA[Follow a booking record through server acknowledgement, reload, retry, and a second signed-in visitor in a reproducible Chromium fixture.]]></description><link>https://uvilno.hashnode.dev/what-a-booking-confirmation-actually-proves</link><guid isPermaLink="true">https://uvilno.hashnode.dev/what-a-booking-confirmation-actually-proves</guid><category><![CDATA[Testing]]></category><category><![CDATA[Web Development]]></category><dc:creator><![CDATA[Uvilno]]></dc:creator><pubDate>Fri, 13 Nov 2026 15:36:46 GMT</pubDate><content:encoded><![CDATA[<p>Visitor A clicks <strong>Send request</strong>. The repair portal says <strong>Request received</strong>. Visitor B opens the same detail link and sees <strong>Request unavailable for this visitor</strong>. Both screens look right. They can also look that way when A’s browser is holding an unsaved record in memory and B’s session has expired.</p>
<p>We built a local repair-shop fixture to make those outcomes distinguishable. The release test follows one known record through the server’s acknowledgement, a reload, a repeat submission and a second visitor’s session. You can use the same sequence on any app you own, with two disposable accounts and fictional data.</p>
<h2>Ask the service what it saved</h2>
<p>A chooses a screen repair, enters a date and submits a note about a flickering test phone. The browser test waits for the create response and saves the returned record ID. It checks that the detail link points to that ID and the screen shows one saved request. Then it reloads the page and checks the note again.</p>
<p>The wording matters. <strong>Request received</strong> means the record was saved. The shop still has to confirm the appointment. If an interface shows success while a write is pending, give that state its own wording. An optimistic row may be useful while the request runs. A receipt needs an acknowledgement from the service.</p>
<h2>Repeat the write that might have already succeeded</h2>
<p>Imagine the first request reached the service but the response never reached the browser. A retry should preserve the identity of that submission. In our fixture, repeating the exact create request with the same request ID returned the original record ID. A reload still showed one saved request. Reusing that request ID with different notes returned a conflict. A separate request with an impossible date was rejected.</p>
<p>That gives the release test three outcomes to distinguish. A retry of the same submission should return its original result. An edited submission needs a new request ID or the app’s normal update flow. Invalid data should produce a rejection that the form can explain. Our test checks the service’s replay behavior. It did not simulate a lost response or prove that every client retry path preserves the key.</p>
<h2>Keep a working read next to the denied one</h2>
<p>A’s detail URL came from the acknowledged save. In a separate browser context, B opened that URL and saw the unavailable message. We checked that A’s note was absent from the screen and the fixture’s read response returned an empty record. Check both layers in your own app. A detail page should never receive another visitor’s private fields, even if it would hide them from view. <a href="https://api-security.owasp.org/editions/2023/en/0xa1-broken-object-level-authorization/">OWASP guidance on object-level authorization</a></p>
<p>The next useful check was an allowed read. B saved and opened a different request for a battery replacement, tried A’s link again and reopened B’s own request. A then loaded the original note using a fresh A session. With both owned reads working, the cross-account denial had evidence behind it. If either owned read fails, record the privacy check as inconclusive and look for a session, routing or availability problem first.</p>
<h2>Record the result of each check</h2>
<p>This is what the local run established. Keep the record IDs and expected response shape alongside equivalent results from your own release candidate. In shared test logs, use labels such as <code>A-1</code> and <code>B-1</code> and leave out cookies and private fields.</p>
<table>
<thead>
<tr>
<th>Check</th>
<th>Observed result in the fixture</th>
</tr>
</thead>
<tbody><tr>
<td>A submits a screen repair and reloads</td>
<td>Create returns a record ID and the same note appears after reload</td>
</tr>
<tr>
<td>A repeats the exact submission</td>
<td>Same record ID and one saved request</td>
</tr>
<tr>
<td>A reuses that request ID with a changed payload</td>
<td>Conflict</td>
</tr>
<tr>
<td>B opens A’s known link</td>
<td>Unavailable UI and an empty record from the read endpoint</td>
</tr>
<tr>
<td>B reopens B’s own request after the denial</td>
<td>Battery replacement record loads</td>
</tr>
<tr>
<td>A reopens the original request with a fresh session</td>
<td>Original note loads</td>
</tr>
</tbody></table>
<p>The run used Chromium, synthetic sessions and the app’s managed-record SQL in an in-memory database. The local database lasted for the run, so the reload demonstrates persistence across browser sessions under that fixture. The hosted service, real identity provider and database recovery after a restart each require their own test. Before publishing an equivalent portal, repeat the same checks on the actual release URL with disposable visitors. Capture the release version and origin so a passing local result stays distinct from a passing deployment. <a href="https://uvilno.com/guides/repair-shop-portal-example">Repair-shop fixture and release exercise</a></p>
<p><em>Editorial disclosure. Prepared with AI assistance for the Uvilno team. The example describes Uvilno’s own local fixture.</em></p>
]]></content:encoded></item><item><title><![CDATA[Handing over an AI-built app? Run a clean-room restore first]]></title><description><![CDATA[Verify source, sign-in, records, file access and rollback in a fresh environment before handing over an AI-built application.]]></description><link>https://uvilno.hashnode.dev/handing-over-an-ai-built-app-run-a-clean-room-restore-first</link><guid isPermaLink="true">https://uvilno.hashnode.dev/handing-over-an-ai-built-app-run-a-clean-room-restore-first</guid><category><![CDATA[Devops]]></category><category><![CDATA[Web Development]]></category><category><![CDATA[deployment]]></category><dc:creator><![CDATA[Uvilno]]></dc:creator><pubDate>Fri, 13 Nov 2026 14:49:53 GMT</pubDate><content:encoded><![CDATA[<p>A source export is easy to verify at the file level. Open the ZIP, find the components and run a build. A handoff has a second test. Can a different operator run the application and recover the workflows that depend on live services?</p>
<p>Use a clean directory or fresh container, a disposable domain or deployment URL, and synthetic accounts and data. Do the exercise before moving the public domain or turning off the existing app. Record the exported project version, runtime version, package manager and lockfile.</p>
<h2>Inventory what the app calls</h2>
<p>Start with three workflows a visitor can demonstrate. Sign in, save a record and reopen it from a fresh session. List the services needed at each step. For an account workflow, identify the identity provider, the allowed callback URLs and any email delivery dependency. For a record, identify the data store, schema migration and authorization policy. For assets, check any remote storage URLs. If your app receives webhooks, record the destination and signing-secret setup.</p>
<p>In the handoff sheet, give each dependency an owner, a configuration step and a restore check. Write the environment variable names and the purpose of each variable. Provision new secrets through the receiving environment’s secret manager.</p>
<table>
<thead>
<tr>
<th>Dependency</th>
<th>Prepare in the new environment</th>
<th>Restore check</th>
</tr>
</thead>
<tbody><tr>
<td>Application source</td>
<td>Install from the lockfile and run the documented build</td>
<td>Open the key routes on the destination URL</td>
</tr>
<tr>
<td>Identity</td>
<td>Configure the account service and destination callback</td>
<td>Sign in with a disposable user and verify sign-out</td>
</tr>
<tr>
<td>Records</td>
<td>Apply schema and permissions, then import synthetic fixtures</td>
<td>Reload an owned record and check denial for a second account</td>
</tr>
<tr>
<td>Files</td>
<td>Copy supported assets and validate remote storage references</td>
<td>Open files from a fresh visitor session</td>
</tr>
<tr>
<td>Custom domain</td>
<td>Configure hostname and routing at the chosen host</td>
<td>Check HTTPS, redirects and login callbacks after routing changes</td>
</tr>
</tbody></table>
<h2>Prove the destination is doing the work</h2>
<p>Create a distinct record through the destination app and verify it appears in the destination’s data service. A page can build cleanly while the browser still sends every write to the original backend. Check the origins of the app’s service requests and review them against the inventory. Repeat the test after a fresh session so stale authentication or cached data cannot cover a missing service.</p>
<p>For a dry run, call the new synthetic record <code>handoff-candidate-12</code>. On the receiving system, capture the save outcome, reopen the same record in a fresh browser profile and confirm its owner. Check whether the record also appeared at the old destination. If it did, determine whether a documented shared data service was intended. An unexpected write to the old service blocks the cutover even when the new URL renders.</p>
<p>Export boundaries vary by project. Lovable’s documentation, for example, describes source and schema files moving through Git or code export with data and uploads handled separately. Uvilno’s documentation similarly calls for a migration plan for managed records, visitor identities and credentials when moving a supported project. Ask the same concrete questions on any platform you use. <a href="https://docs.lovable.dev/tips-tricks/deployment-hosting-ownership">Lovable ownership and hosting</a> · <a href="https://uvilno.com/guides/how-to-export-and-host-your-ai-website">Uvilno export and hosting guide</a></p>
<p>When the receiving operator can repeat these checks, the handoff includes both a buildable project and a demonstrated route to operating it. Keep the original service available until the owner has checked the live dependencies and can describe how they would reverse a failed cutover.</p>
<p><em>Editorial disclosure. This article was prepared for the Uvilno team with AI assistance. The exercise describes a general verification procedure.</em></p>
]]></content:encoded></item></channel></rss>