Methodology
The web optimization industry runs on rumor. A patent gets misread, a correlation study gets repeated as causation, a tactic that worked on one site in 2019 becomes gospel. Our answer is a research pipeline borrowed from engineering practice: every candidate technique is verified against primary sources, implemented here in a controlled setting, measured, and only then — if it survives — allowed to touch a production website.
Evidence requirements
No single source is sufficient. Primary documentation outranks everything: Google Search Central, web.dev, the HTML Living Standard, W3C and WHATWG specifications, Schema.org, and official crawler documentation from the AI providers. Reputable enterprise case studies count as supporting evidence; industry consensus counts as weak evidence; a YouTube thumbnail with a shocked face counts as nothing.
Confidence levels
| Confidence | Basis | Action allowed |
|---|---|---|
| 100% | Official documentation states it directly | Implement |
| 99% | Official documentation + validated in this lab | Implement |
| 95% | Official documentation + multiple enterprise case studies | Implement with monitoring |
| 90% | Multiple respected enterprise case studies | Test here first |
| 80% | Strong industry consensus | Test here first |
| < 80% | Anything weaker | Research only — never implemented |
The pipeline
- Research — primary sources first, always.
- Evidence verification — cross-check against independent sources; assign confidence.
- Controlled deployment — implemented on this site. One hypothesis, one change, one measurement. Never multiple unrelated changes at once.
- Observation — Search Console (indexing, impressions, position), Core Web Vitals field data, validation tooling, crawler behavior. Long enough to establish signal, not vibes — search data lags days and trends take weeks.
- Measurement & review — before/after recorded in a permanent experiment log: hypothesis, sources, files changed, metrics, rollback procedure, verdict.
- Graduation — proven techniques become versioned BYHS Engineering Standards.
- Production — only graduated standards ship to client websites.
Hard rules
- Nothing ships to a client site that hasn't graduated.
- Changes capable of harming search visibility — URL structures, redirects, canonicals, robots directives, page removal — are never automated. A human reviews and approves every one.
- Every experiment is reversible, and its rollback procedure is written before the change is made.
- Negative results are recorded with the same care as wins. A technique that did nothing is a finding, and it saves every future client from paying for it.
- Users first. If a technique helps an algorithm but degrades the human experience, it fails regardless of the metrics.
Why publish it
Two reasons. Accountability: public methodology is a standard we can be held to. And honesty about incentives: an agency that documents what doesn't work has a harder time selling it to you. The research section is the public record; the standards page is the current state of what has earned its place.