Fictional Launch Lab 7Q9: review example
This app is fictional. This is an illustrative sample, not a client project.
All conditions and findings below are hypothetical. Evidence types describe what a reviewer would inspect, not evidence we collected.
Scope reviewed in this example
Assume an AI-built workspace app that summarizes uploaded notes and sells usage credits.
The example uses a generated web interface, Supabase Auth and Postgres, Firebase Storage attachments, and Stripe test payments.
The checks can apply to apps built with Lovable, Bolt, v0, or Replit. Tool choice alone does not establish safety.
- Journeys: sign-in, account recovery, workspace access, attachment upload, AI summary, credit purchase, and subscription updates.
- Environments: a hypothetical pinned repository version, isolated staging database, synthetic accounts, and payment test mode.
- Inputs: schema, access rules, server handlers, deployment configuration, logs, and operating instructions.
- Exclusions: production traffic, real customer data, live payments, legal review, and a full penetration test.
Decision: fix first
Do not launch this hypothetical version. Resolve the critical and high findings, then repeat the affected journeys.
- Ship: retain the interface flow after functional checks. This does not approve the full app for release.
- Fix first: correct permissions, credential exposure, payment handling, usage limits, and recovery coverage.
- Rebuild part: replace the payment event handler if local changes cannot support safe event processing.
The sample does not justify a full rebuild. Agree on the medium finding owner and release conditions before reconsidering launch.
Critical means a launch blocker involving broad access or payment trust. High means material access, financial, or recovery risk.
Medium means an operational weakness that needs an owner and an agreed release decision. These labels are sample priorities, not CVSS scores.
12 illustrative findings
F01 · Critical · Hypothetical
Project records lack row-level protection
The fictional projects table grants browser access without enabled RLS.
- Evidence type to inspect
- Database grants and policies; direct API reads using two test accounts and an anonymous request.
- Why it matters
- A user could read another workspace's project records.
- Proposed fix
- Enable RLS and add membership policies for each allowed operation.
- Acceptance check
- Account A cannot read or change account B's projects. Authorized project access still works.
F02 · Critical · Hypothetical
A privileged key reaches browser code
The fictional browser bundle contains a Supabase service_role key.
- Evidence type to inspect
- Build output, browser requests, and the key's role. Record a redacted fingerprint, never the credential.
- Why it matters
- A service_role key can bypass RLS. Public publishable keys have a different purpose and are not equivalent secrets.
- Proposed fix
- Remove the key, rotate it, and move privileged operations behind server authorization.
- Acceptance check
- The old key fails. Browser assets contain no privileged key. Server operations reject unauthorized users.
F03 · High · Hypothetical
Users can move records between workspaces
The fictional update policy checks existing ownership but permits a changed workspace identifier.
- Evidence type to inspect
- Update policies, membership joins, and a cross-workspace update test.
- Why it matters
- A user could transfer a record into a workspace they do not control.
- Proposed fix
- Check both the existing row and the proposed row against permitted membership.
- Acceptance check
- A forged workspace change fails. Valid edits retain the correct workspace.
F04 · High · Hypothetical
The admin action trusts a hidden button
The fictional member-removal endpoint checks sign-in but omits the administrator role check.
- Evidence type to inspect
- Server handler and direct requests from a member account without administrator permission.
- Why it matters
- A member could remove another member without using the interface.
- Proposed fix
- Authorize the actor, workspace, and requested action on the server.
- Acceptance check
- Members receive a denial. Administrators can remove members only within their own workspace.
F05 · High · Hypothetical
Attachment rules allow any signed-in user
The fictional Firebase Storage attachment rules require sign-in but do not check the file owner.
- Evidence type to inspect
- Storage rules and cross-account reads, uploads, and deletes against synthetic files.
- Why it matters
- Sign-in alone does not establish permission to another user's files.
- Proposed fix
- Restrict file operations to the intended owner or authorized workspace members.
- Acceptance check
- Unrelated accounts and anonymous users cannot read or modify attachments. Authorized operations still work.
F06 · Critical · Hypothetical
Payment events lack signature verification
The fictional Stripe webhook handler trusts parsed JSON before checking its signature.
- Evidence type to inspect
- Webhook handler, raw request handling, and signed versus invalid test events.
- Why it matters
- A forged request could change payment status.
- Proposed fix
- Verify the raw payload with Stripe's signature header and the correct endpoint secret before processing.
- Acceptance check
- Invalid signatures fail without changing records. Valid test events reach the expected handler.
F07 · High · Hypothetical
Repeated payment events add credits again
The fictional handler adds credits on every delivery of the same event.
- Evidence type to inspect
- Event records and a replay of one test event.
- Why it matters
- Duplicate delivery could grant credits twice.
- Proposed fix
- Record processed events and make the business update atomic and idempotent.
- Acceptance check
- Replaying the event leaves the credit total unchanged after the first successful update.
F08 · High · Hypothetical
Subscription updates depend on arrival order
The fictional handler assumes payment events arrive in creation order.
- Evidence type to inspect
- State transition code and reversed-order test deliveries.
- Why it matters
- Late events could overwrite current entitlement state.
- Proposed fix
- Reconcile relevant Stripe objects and reject stale transitions instead of relying on delivery order.
- Acceptance check
- Reordered deliveries converge on the correct current state.
F09 · High · Hypothetical
AI requests have no enforced usage limit
The fictional summary endpoint accepts unlimited repeated requests and large inputs.
- Evidence type to inspect
- Endpoint configuration, request limits, and bounded synthetic load cases.
- Why it matters
- Excessive requests could exhaust capacity or create uncontrolled provider charges.
- Proposed fix
- Enforce account quotas, request size limits, concurrency limits, and timeouts on the server.
- Acceptance check
- Excess requests stop before provider work starts and return a clear limit response.
F10 · High · Hypothetical
File recovery is absent from the backup plan
The fictional recovery document assumes a Supabase database backup contains every uploaded file.
- Evidence type to inspect
- Backup inventory and an isolated restore exercise with synthetic records and files.
- Why it matters
- Supabase database backups exclude Storage objects. This app's separate Firebase attachments also need their own recovery plan.
- Proposed fix
- Define file backups separately and rehearse restoring records with matching attachments.
- Acceptance check
- A restored test record opens its expected file. Document missing coverage and recovery owners.
F11 · High · Hypothetical
Project administration lacks MFA
The fictional Supabase project administrator relies only on a password.
- Evidence type to inspect
- Project administration settings and the authorized account access list, without collecting credentials.
- Why it matters
- A compromised administrator account could expose project infrastructure and data.
- Proposed fix
- Enable MFA for administrative accounts and document account recovery ownership.
- Acceptance check
- Administrative sign-in requires the configured second factor. The owner can follow the recovery procedure.
F12 · Medium · Hypothetical
Queue failures have no named responder
The fictional summary job retries indefinitely without an operational owner.
- Evidence type to inspect
- Retry configuration, failure logs, and a controlled provider timeout.
- Why it matters
- Repeated failures could waste resources and leave users waiting without a clear status.
- Proposed fix
- Bound retries and execution time. Record failed jobs and assign an operational owner.
- Acceptance check
- A failed test job stops at its limit and appears in the owner's review queue.
What was not reviewed
No real repository, deployment, account, or customer data was inspected. This report provides no security certification or launch approval.
A real scope must state which checks remain open. Examples include accessibility, model answer quality, peak load, and legal obligations.
Payment disputes, tax configuration, mobile clients, and production recovery also remain outside this sample.
Next steps
- Agree on the real version, journeys, environments, access, exclusions, and review timing.
- Collect redacted evidence and assign each finding an owner.
- Scope fixes separately. Repeat acceptance checks against the proposed release.
- Record remaining risks and the release decision with the product owner.
For broader unfinished software, review software project rescue. Prepare inputs with the project brief builder.