1 ยท Configure
Resolved configuration
Override any value without editing a file:
localStorage['flightbridge.sandbox.config'] = JSON.stringify({ webUrl: 'โฆ' }),
then reload. The server takes FB_* env vars.
2 ยท The URL
โ
3 ยท The frame
sandbox="allow-scripts allow-same-origin allow-forms allow-popups
allow-popups-to-escape-sandbox allow-top-navigation-by-user-activation"
โ over-tight sandboxing is a classic cause of a blank third-party frame.
Edit the attribute in index.html to bisect.
FlightBridge's hotel search paginates only when the parent posts
'scrollEvent' into the child. It is the one part of the contract that
fails silently โ the list simply stops growing โ so it gets its own button.
The harness also sends it automatically on near-bottom scroll.
The frame is empty or shows an error. The likely causes, in order:
- The token was minted on a server their callback cannot reach.
Registration itself is done: since 2026-09-02 FlightBridge calls
GET https://flightbridge-sandbox.nerd.aero/flightbridge/auth/{token}. That is the only token store their lookup sees, so a token minted by a localserve.mjswill not be found and the page answers “You are not authorized to perform this action.” — exactly whatreturns for a request with no valid token. Fix: browse the deployed harness, or setsandboxOriginso a locally edited page mints into the deployed store (README §“Editing locally against the deployed token store”). - The callback was reached but rejected the token. Check the event log for
a
FB CALLED USentry. If it says REJECTED, the token was unknown, expired, or already spent โ tokens are single-use, so a plain frame reload always fails. Mint a fresh one. - The callback was reached but we answered the wrong shape. The log records
the raw request; compare what they sent against what
serve.mjsextracted. Their shape was confirmed by email on 2026-09-02 — the token is the last path segment of aGET, not a query parameter — so a healthy record readstokenFoundVia: "path segment /{token}". Any other value means they changed something, and the raw request in the log says what. - Framing was blocked. There is no event for this: a
frame-ancestorsrefusal firessecuritypolicyviolationin the framed document's context, not in ours, and the console message is not readable from JS. Theloadevent still fires, because the browser commits an error document into the frame. So the only signature we can capture is the composite the export records asloaded-silent-no-callback: load fired, nopostMessage, and noFB CALLED USentry. If aFB CALLED USentry is present, framing was allowed and the problem is above, not here. Theirframe-ancestorsallowslocalhostandlocalhost:*, but both are written without a scheme, and a scheme-less CSP source inherits the scheme of the page that sent the policy —https://test.flightbridge.com. So those entries meanhttps://localhostonly: browse this harness athttps://localhost:4478, not over plain HTTP. (Their list spellshttp://localhost:3030/out in full, which is the same lesson.) It would also bite on a tunnel URL or ondev-my.magellanjets.com— neither is in the list, and*.ngrok.iodoes not cover an*.ngrok-free.devdomain. Since 2026-09-02flightbridge-sandbox.nerd.aerois on the list, so the deployed harness may frame them too; every other new embedding origin is still a separate request to FlightBridge. The check is against the origin of the page doing the framing, so a tunnel used only for the callback never enters into it. - Third-party cookies were blocked. FlightBridge sets
SameSite=None; Secureand ships its own “Enable Cookie Permissions” page for this, which needs a user gesture in a new tab on Safari. If you land on that page, follow it once. Credential #6.
Full detail for each: ../CREDENTIALS-NEEDED.md.
4 ยท Event log
The case id tags every entry below and rides along on the mint request, so the
server's record of FlightBridge's callback names the case too. clear above
starts the next case. Case ids are in IFRAME-TEST-PLAN.md.
Browser events interleaved with what the server saw. When FlightBridge calls the callback this shows exactly what they sent โ which is how we settle the open question of the callback's real request shape.