Launch walkthrough 2

Walkthrough: holding a staging deploy in CI until the blockers are fixed

We ran the example script scan-check.mjs, unchanged, against a local copy of the Launch Ready API, the way a CI job runs it after a staging deploy. With the contact form posting to localhost the script exited 1 and printed the Block Launch finding first. With only the form fixed it exited 0 and still listed the lorem ipsum as a high blocker; run again on the same deployment with FAIL_ON=high, it reused that scan and exited 1. After the copy fix it exited 0, which means no open blocker at the chosen level, not launch approval.

Last reviewed

← All launch walkthroughs

This is a demonstration on a synthetic site, not a client launch. The site, the bakery and its copy are made up for testing; no customer, client site or customer data is involved.

Scans run on on a local copy of Launch Ready (commit 017a24d), not on the hosted service. Every finding, count, score, verdict and line of output below comes from those runs, saved in the evidence file launch-walkthrough-2-2026-11.json.

The setup

The staging site is the same synthetic two-page bakery site as in walkthrough 1, with a contact form posting to localhost, lorem ipsum on the homepage and a staging noindex tag. This time nobody opened the dashboard: each deploy was followed by the example CI script scan-check.mjs, which starts a staging scan through the API, waits for it, prints the blockers and sets its exit code.

The script ran unchanged with Node.js, from the command line. It talked to a local copy of the Launch Ready API at https://127.0.0.1:3443/api/v1 and scanned https://localhost:4443/. The key was a workspace API key created in a new local account on the Free plan, the way Settings → API keys creates one.

What did not run: The workflow file was not run on GitHub Actions or any other CI service. Its launch-ready-check job runs the same command (node scan-check.mjs) with the same variables; that command is what ran here, on a local machine.

How this local run differs from a scan on the hosted service:

  • LAUNCHGUARD_DISABLE_RENDER=1: pages were read as HTTP HTML; Chromium did not render them, as it does by default. The fixture is static HTML, so the markup the rules read is the same, but the analytics rule grades by rendering: a missing analytics tag is an Improvement on an unrendered page, and Important on a rendered page in a production scan. With rendering on, the two production scans would also have reported no_analytics as Important.
  • LAUNCHGUARD_DISABLE_LIGHTHOUSE=1: performance was not measured. Free scans do not measure performance either.
  • LAUNCHGUARD_ALLOW_HOSTS=127.0.0.1,localhost: a development allowlist that lets the scanner reach loopback. Hosted Launch Ready refuses localhost and IP addresses.
  • The fixture was served at https://localhost:4443/ over HTTPS with a certificate from a local test CA, trusted by the app, the worker and the script through NODE_EXTRA_CA_CERTS.

First run: the Free workspace had not chosen its site

A Free workspace checks one site, and the script never confirms it on its own. The very first run stopped before any scan with exit code 2, which the script uses for “no usable result”:

Run 0: before the site was confirmed

Settings for this run: IDEMPOTENCY_KEY=deploy-1. Exit code: 2.

No usable scan result: Launch Ready answered HTTP 409 (binding_required). A Free workspace checks one site. Scanning this binds the workspace to localhost for good; its subdomains and preview URLs stay open. Ask the user, then repeat the call with confirmDomain: true. This Free workspace has not chosen its one site yet (proposed: localhost). Confirm it once in the app; this script never confirms it for you. Preview hosts do not count.

We confirmed the site once with the documented API request that sends confirmDomain: true. That request also started a scan, which this walkthrough does not use.

Deploy 1: the Block Launch finding stops the job

Run 1: the original site

Settings for this run: IDEMPOTENCY_KEY=deploy-1. Exit code: 1.

Scan 97060d22-ae83-49a0-a8c2-211361c162b3 started for https://localhost:4443/ (staging).
Scan 97060d22-ae83-49a0-a8c2-211361c162b3: completed.
Block Launch (critical): 1
  - Form submits to localhost on https://localhost:4443/contact.html [form_action_localhost]
High: 1
  - Lorem ipsum placeholder copy on https://localhost:4443/ [lorem_ipsum]
Readiness: critical_blocker (score 35)
Coverage: 2 pages checked of 2 discovered.
Scanned: https://localhost:4443/ (staging)
Report: https://127.0.0.1:3443/projects/168291bb-6fba-4dd6-8068-c3071933e34a/scans/97060d22-ae83-49a0-a8c2-211361c162b3
Result: fix the blockers above before asking for production approval.

The script exited 1, so a CI job running it fails. The example workflow is written so that its promotion job only runs after the check job succeeds. The Block Launch finding is printed first, then the Important one. The staging noindex tag is not in the list: in a staging scan it is an Improvement, and the script asks only for Block Launch and high findings.

Deploy 2: the form fixed, the copy not yet

The second deploy changed only the form action:

--- a/original/contact.html
+++ b/formfixed/contact.html
@@ -17,12 +17,7 @@
     <main>
       <h1>Order a celebration loaf</h1>
       <p>Tell us the date and the size you need. We reply within one working day.</p>
-      <!--
-        PLANTED MISTAKE 3: the form still posts to the developer's machine.
-        Real visitors cannot submit it.
-        Rule: form_action_localhost (critical, "Block Launch", in every environment).
-      -->
-      <form action="http://localhost:3000/api/contact" method="post">
+      <form action="/api/contact" method="post">
         <label for="name">Your name</label>
         <input id="name" name="name" type="text" required />
         <label for="date">Collection date</label>

Run 2: FAIL_ON not set, so the default, critical

Settings for this run: IDEMPOTENCY_KEY=deploy-2. Exit code: 0.

Scan a96e7c10-c087-4698-a3b7-67fab4e082f0 started for https://localhost:4443/ (staging).
Scan a96e7c10-c087-4698-a3b7-67fab4e082f0: completed.
Block Launch (critical): 0
High: 1
  - Lorem ipsum placeholder copy on https://localhost:4443/ [lorem_ipsum]
Readiness: not_ready (score 69)
Coverage: 2 pages checked of 2 discovered.
Scanned: https://localhost:4443/ (staging)
Report: https://127.0.0.1:3443/projects/168291bb-6fba-4dd6-8068-c3071933e34a/scans/a96e7c10-c087-4698-a3b7-67fab4e082f0
Result: no open Block Launch finding. This is input for the person who approves the promotion, not an approval.

With no open Block Launch finding the script exited 0, and still listed the lorem ipsum as a high blocker.

A team that wants high findings to stop the job too sets FAIL_ON=high. Run again on the same deployment, with the same idempotency key, the script reused the scan it had just read and exited 1:

Run 3: the same deployment, FAIL_ON=high

Settings for this run: IDEMPOTENCY_KEY=deploy-2, FAIL_ON=high. Exit code: 1.

Scan a96e7c10-c087-4698-a3b7-67fab4e082f0 reused for https://localhost:4443/ (staging).
Scan a96e7c10-c087-4698-a3b7-67fab4e082f0: completed.
Block Launch (critical): 0
High: 1
  - Lorem ipsum placeholder copy on https://localhost:4443/ [lorem_ipsum]
Readiness: not_ready (score 69)
Coverage: 2 pages checked of 2 discovered.
Scanned: https://localhost:4443/ (staging)
Report: https://127.0.0.1:3443/projects/168291bb-6fba-4dd6-8068-c3071933e34a/scans/a96e7c10-c087-4698-a3b7-67fab4e082f0
Result: fix the blockers above before asking for production approval.

Deploy 3: the copy fixed

--- a/formfixed/index.html
+++ b/fixed/index.html
@@ -24,11 +24,7 @@
       <h1>Fresh bread every morning</h1>
       <p>Example Bakery bakes sourdough, rye and seeded loaves in small batches before the shop opens.</p>
       <h2>Our story</h2>
-      <!--
-        PLANTED MISTAKE 2: dummy Latin copy that nobody replaced.
-        Rule: lorem_ipsum (high on the homepage).
-      -->
-      <p>Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.</p>
+      <p>Example Bakery started as a weekend market stall. The same two bakers still shape every loaf by hand.</p>
       <h2>Visit us</h2>
       <p>Open Tuesday to Saturday, 7:00 to 14:00. <a href="/contact.html">Send us a message</a> to order a celebration loaf.</p>
     </main>

Run 4: FAIL_ON=high

Settings for this run: IDEMPOTENCY_KEY=deploy-3, FAIL_ON=high. Exit code: 0.

Scan 80999b65-f28d-45f9-bb9b-c075ecaf75e9 started for https://localhost:4443/ (staging).
Scan 80999b65-f28d-45f9-bb9b-c075ecaf75e9: completed.
Block Launch (critical): 0
High: 0
Readiness: ready_with_warnings (score 91)
Coverage: 2 pages checked of 2 discovered.
Scanned: https://localhost:4443/ (staging)
Report: https://127.0.0.1:3443/projects/168291bb-6fba-4dd6-8068-c3071933e34a/scans/80999b65-f28d-45f9-bb9b-c075ecaf75e9
Result: no open Block Launch finding. This is input for the person who approves the promotion, not an approval.

The script exited 0. Its last line says what that means: no open blocker at the chosen level, which is input for the person who approves the promotion, not an approval.

The two rules that held the deploy

The report's own text for the findings the script printed:

Rule form_action_localhost

Form submits to localhost

Why it matters: The form action points at a local machine. Real visitors cannot submit it. This is a static markup check — Launch Ready did not send a test submission.

How to fix: Point the action at your production form handler or a live form service.

Rule lorem_ipsum

Lorem ipsum placeholder copy

Why it matters: Dummy Latin copy is still on the page. Clients and customers will notice immediately.

How to fix: Replace placeholder copy with the real launch content.

What the script and the scans did not check

  • The workflow file was not run on GitHub Actions, so no runner, secret store or required reviewer was involved. Only the command its check job runs was executed.
  • The contact form was read, not submitted, and no user journeys were run.
  • This is not an accessibility audit, a security test or a legal review.
  • The script reads only Block Launch and high findings. The Improvement findings in the same scans, such as the staging noindex tag, did not affect its exit code.
  • Launch Ready does not guarantee a bug-free website.
  • It does not make the launch decision. People make it, with the report in hand.

What still needs a person

  • Approve the production promotion. In the example workflow a required reviewer does this; exit code 0 does not.
  • Remove the staging noindex tag in production and scan the live address after launch.
  • Submit the live contact form and confirm the message arrives.
  • Read the coverage line: a scan stopped by the plan's page limit did not see every page.

How to reproduce it

The deployment checks guide explains the staging scan, the blocker policy and the separately approved promotion. The API reference documents the three requests the script makes: start a scan, read the report, read the blockers. Run it after a deploy to a public staging address, with a workspace API key in LAUNCH_READY_KEY. The scan-check.mjs example used here is not published as a download yet.

Background on the two mistakes that held the deploy:

For what to check by hand before launch, use the website launch checklist.

Check your next client site before handoff

Add the staging or live URL, read the fix list, and send the client a report when it is ready.

Create a Launch Ready account to run a scan.