Trust, independence and data
The rules behind the word "independent", how our testing browser behaves, what permission we need, and what happens to data today.
Independence
TrendLab is owner-operated. Independence comes from our work relationship with the agent, not from separate brands or teams.
Independent
We had no hand in building or advising on the agent, we hold no financial interest in its owner, and any referral fee is fixed and disclosed.
May be called "independent evaluation within tested scope".
Evaluation with disclosed prior work
We previously designed parts of the client's evaluation system or advised on its AI quality program. Whether, and when, such an evaluation can carry a verdict is set by our independence policy, which is being finalized; the prior work is always disclosed with dates.
Never called independent without the disclosure.
Internal evaluation
We built or changed the agent, or hold a financial interest in its owner.
Results only. Never called independent, and no Assurance Verdict.
- Our price is fixed before testing starts. No fee, bonus or discount ever changes with a verdict or with how many findings we report.
- Referral partners and agencies never see draft findings without the client's written consent and have no say in the wording.
- Only the TrendLab evaluator issues or clears a verdict. A client may record that it accepts a risk; that never turns a blocking result into a passing one.
- Every engagement starts with a recorded independence check, and reports print the resulting classification.
Testing conduct
Our testing browser identifies itself, stays slow, and stops when asked.
Polite volume
At least 3 seconds between messages and at most 60 messages per site per day, usually about 40 to 58 in one session.
Normal chat only
It opens your public chat and sends ordinary customer-style messages. It does not log in, submit payment or personal data, download files or try to access your systems.
Stops when blocked
It stops on a CAPTCHA, a block page or rate-limit signals. We never load-test public endpoints.
Identifies itself
Its user agent contains "TrendLab-Evals-Snapshot". Opt-out requests are honored within one business day.
Details and opt-out: TrendLab-Evals-Snapshot browser.
Permission
How we behave is not the same as permission to test.
- Free Snapshot you request: you confirm you are authorized to request testing of the bot.
- Paid work: written authorization naming the systems, environments, access and dates. Tests that could cause side effects (actions, deletion, load) run only in a non-production environment you provide.
- A vendor's agent: a buyer cannot authorize testing of a vendor's system. We need the vendor's consent or sandbox, or a public test the vendor's terms allow, with the vendor informed.
Data
What happens today, stated plainly.
Outside-in tests (Snapshot)
- We keep the chatbot's replies, screenshots of the chat and the run records, as evidence for the report.
- Secrets are redacted before anything is stored.
- Bot replies are sent to our two AI judge providers for grading.
- We never enter real people's personal data as test input, and we do not publish results about your bot.
Work involving your data
Some services review your transcripts or production samples (Eval Setup, readiness assessments, incident response, production sampling). Before any engagement that involves your confidential data, we publish a formal data policy (retention, deletion and handling) and sign a data agreement with you. Until then we take on only work that uses public information.
Questions about any of this: inquiries@trendlabanalytics.com. These rules describe how we work. They are not legal advice.