Ten principles
- P1Evidence Before Narrative
- Never create a conclusion first and search for evidence afterward.
- P2Reality Decides
- AI opinion is not market validation.
- P3Do Not Scale Before Evidence
- Do not recommend major investment before meaningful evidence appears.
- P4Website Is a Validation Bridge
- A website exists to test reality, not merely to generate pages.
- P5Kill Experiments, Preserve Learning
- Failure remains useful data.
- P6First-Party Evidence Has Priority
- Real analytics, rankings, inquiries, orders and observed user behaviour are stronger than AI speculation.
- P7No Fabricated Metrics
- Unknown values must remain unknown.
- P8No Cross-Project Mutation
- The control platform cannot silently modify experiment projects.
- P9Public Claims Require Evidence
- A public result must be traceable to evidence.
- P10Low-Cost Validation First
- Prefer small, measurable experiments before large builds.
Lifecycle — where the test is
Every experiment carries exactly one state.
| State | Meaning |
|---|---|
IDEA | A hypothesis exists but has not entered live-market testing. |
TEST | A real website or other live market asset is running and collecting evidence. |
SIGNAL | Meaningful real-world evidence has appeared. |
PROVEN | Signals are repeated, meaningful and commercially credible. |
BUSINESS | The experiment has graduated into an operating business. |
CLOSED | The experiment has ended or has been archived. |
Evidence that counts toward SIGNAL includes organic search visibility, ranking keywords, returning users, contact clicks, inquiries, quote requests, registrations, external references and relevant AI or search citations.
Decision — what we do about it
Decision is recorded separately from lifecycle. An experiment can be live and undecided, or closed and stopped.
| Decision | Meaning |
|---|---|
GO | Evidence supports further investment. |
HOLD | Evidence is insufficient; maintain at low cost and continue observation. |
KILL | The current hypothesis does not justify additional investment. |
UNDECIDED | No formal review has been completed yet. |
Stopping an experiment never deletes its evidence. A stopped experiment remains part of the historical dataset. Kill experiments. Preserve learning.
Visibility — what may be published
Every piece of experiment information is classified before it can be published.
| Level | Meaning |
|---|---|
PUBLIC | Safe to publish immediately. |
DELAYED_PUBLIC | Potentially commercially valuable information that may be published later. |
PRIVATE | Never publish automatically. |
Never published: customer identities, private emails, phone numbers, contracts, unreleased high-value opportunities, proprietary algorithms, confidential pricing, credentials, access tokens and unpublished commercial strategies.
How evidence is ranked
- First-party observation. Our own analytics, rankings, inquiries, orders and observed user behaviour.
- Externally verifiable fact. Something a third party can independently check.
- Interpretation. Clearly labelled as reasoning, never presented as measurement.
- Opinion. Not evidence. An AI's opinion is not evidence either.
Anything that has not been observed stays marked unknown rather than estimated. A blank is more useful than a plausible-looking number, because a blank can still be filled correctly.