MyJackfruit

Help
Browse topics · Checking that it works

Checking that it works

← Checking that it works

My AI says the tests passed, but it does not work for me. What now?

Short answer

Tell your AI exactly what you did, what you expected and what happened instead; passing tests do not prove it works for you.

Deeper

This happens, and your report is how it gets fixed. Tests check only what someone thought to test, so a fault can sit in a gap no test covers. A test can also run on a stand-in for the real thing. That is why your own try counts, and why Kai-Zen, the planning agent, asks you to check each change live.

First checks:
Read the reply again. Each finished piece is marked REAL if it ran on the real thing, MOCK if it ran on a stand-in, or UNPROVEN if nothing was run. A result marked self-QA was tested by the agent that built it, not by Agent.Tester, the agent that tests what others built.
Check that you see the newest version. Before your app is live, open the address your AI gave you for the preview on your computer; if it does not open or shows the old version, ask your AI to start the preview again. A change to an app that is live shows there only after it is put live again, with your yes.
Check what was asked. The tests check what your plan asks for. If the app does what the plan says but not what you meant, the plan needs a change rather than a fix; the answer How do I change the product while a project runs? shows how.
Check who is signed in. An app on GilbPlatform reads and writes only what the signed-in person may reach, so the same app can show nothing for one person and work for another.

Then tell your AI in the same chat: what you did, step by step, what you expected to see, what you saw instead, and the exact words of any message on the screen. The answer How do I tell my AI about a fault so it can fix it? shows how. Agent.Coder, the agent that writes your app's code, reads the browser's error messages itself with its browser tools; its instructions say not to ask you to copy them from the browser, unless no tool works. It works on the fault; its written procedure has each piece of work tested before it hands it back, and on a FAIL it fixes and tests again. It stops after three different failures it cannot fix without more scope or your input, and tells you what failed and what it tried. An agent is your AI following one written procedure, its standing instructions, for one kind of work.

Once a change is live, Kai-Zen asks you to check it, fixes what you send back, puts it live again and asks again, until you approve. You need not accept something that does not work for you.

If the fault seems to be in GilbPlatform rather than in your app, see Is GilbPlatform down? Where can I see if something is wrong? If nothing helps, write to [email protected].

This answer is written for a computer screen.

Source: Standards read from GilbPlatform on 13 November 2026: Rule_Code_testing (RuleTestProbeArtefactNotDiff: the suite tests what someone thought to test, and this defect was in the gap), Rule_Code_qa-testing (RuleQatestVerdictRules: "'Suite green' is never sufficient evidence of 'works' when the fixtures mock the wire"), Kai-Zen.EvidenceRules (Tag: REAL, MOCK, UNPROVEN), Agent.Coder (AfterIncrement: self-QA; "Stop after three distinct failures that you cannot fix without more scope or the requester's input. Report each attempt"; Browser: Do not ask the requester to copy from the browser. Ask only if no tool works), Proc_v_i_c_Handback, Kai-Zen.HumanCheck (Ask, Fix), Kai-Zen.GoLive. Live help pages, sentences shared: Can my app have its own users who sign in? (Each person sees only what they may reach); How do I build an app with an AI coder? (three different failures); How do I read what an agent hands back?; Where do I get more help? ([email protected]).

Checked against GilbApp d529620.