AI Scoring and Appeals

When you submit a session, an automated model reviews what you produced. For a written submission that is the text. For a screen recording it is the recording, a transcript of what you said, and, where the session was not in English, a translation. The model returns a quality score and a short summary of the usability issues it found.

The score reflects things like:

• whether the scenario the developer set was actually followed
• whether the feedback is specific to the app rather than generic praise
• whether the session ran for a plausible length of time
• whether the submission appears to be genuine human work

The score is a measure of the submission, not of you as a person. We do not use it to build a profile of your character, and it is not shared with other developers as a rating of you.

The score cannot refuse to pay you. It can pay you early. A score of 75 or above releases part of the payment straight away, as an advance on the same amount rather than an extra one, and the rest follows when the submission is accepted. A low score never takes anything away by itself; it only means nothing arrives before the developer looks. The sequence is:

1. You submit. The automated system scores the session and writes a summary. If it scores 75 or above, an advance of 40% of the payment for a screen recording, or 60% for written feedback, lands in your balance now.
2. The developer who paid for the test has 48 hours to accept or reject it. They see your submission and the summary.
3. If the developer does nothing within 48 hours, the submission is automatically accepted and the rest is paid. Silence works in your favour, not against you.
4. If the developer rejects it, they must give a reason, and you can dispute that rejection within 72 hours from your session page.
5. A person at TestFi reviews the dispute and makes the final decision.

For Google Play closed tests the day 14 submission has a 7 day review window, and there is no automatic acceptance: if the developer does not respond, the decision goes to a person at TestFi rather than defaulting either way. There is no score advance on those, because they are already paid in stages across the fourteen days.

One exception to the advance. If your account is under review, for example while we look into a report about it, nothing is paid early: the whole amount waits for the developer instead. That is a delay and not a deduction, and you keep every right in the next section while it lasts.

You can always ask for a human to review any decision that reduces, delays, or withholds your payment. This applies to a rejection by a developer, a score you think is wrong, and a hold placed during an investigation.

When you ask, you are entitled to:

• be told the reason for the decision in plain language
• see the score and summary that were used
• give your own account of what happened, including anything the system could not see
• a decision by a person at TestFi who was not part of the original decision, who has the authority to reverse it, and who is not required to defer to the score

We aim to answer appeals within five working days. If we uphold the original decision we tell you why. If we reverse it, the payment is released.

To appeal outside the in-product flow, or if the window has closed and you have a good reason, email contact@testfi.app with your session link. Appeals are free.

If we suspect a breach of the Tester Agreement such as duplicate accounts or submissions that are not genuine work, we may place the affected earnings on hold while we look into it. We will tell you what we think happened and give you at least 14 days to answer before a person decides.

If a breach is confirmed we may withhold payment for the sessions the breach affected, and in serious or repeated cases close the account. Earnings for sessions the breach did not affect are released.

We would rather say this plainly than let you find out through a rejection. Automated scoring is weaker when:

• the session is in a language the transcription handles poorly, or the audio is noisy
• the app failed to launch or you could not get in, so there was little to record
• the useful finding is short, for example one decisive sentence about why something confused you
• the app is unusual enough that the model misreads normal behaviour as a problem

Any of these is a good reason to appeal, and the appeal route exists precisely because the model gets these wrong.

Where the law where you live gives you rights about automated decision making, including under Article 22 of the UK and EU GDPR, you may exercise them regardless of this policy: to obtain human intervention, to express your point of view, and to contest the decision. Clause 3 is how we provide that in practice.

The legal basis for this processing, the retention periods, and how to contact us or complain to a supervisory authority are in the Privacy Policy.

Completed sessions are used to improve the scoring models themselves, under clause 8 of the Tester Agreement. That is how the weaknesses in clause 5 get smaller over time rather than staying where they are.

The data stays inside TestFi. We do not sell sessions or hand them to anyone else to train their models, and where a transcript or a score is enough we use those instead of the raw video. If you want to object to this use, email contact@testfi.app; it will not affect your pay, your rating, or the work you are offered. The legal basis is in the Privacy Policy.