Research log / Evidence classes

Platform policy, peer review, preprints, observation, and hypothesis are not the same thing.

Every research note labels its evidence class, publication state, scope, conflicts, and limits. Null and negative findings remain visible.

Research register open
01

Initial programs

01

Crawler controls

Test how search, user-fetch, and training agents reach the same public evidence through Cloudflare controls and WAF policy.

02

Local AI discovery

Measure how an established local entity, real reviews, owned pages, and independent mentions appear across AI-search surfaces.

03

Multilingual retrieval

Compare source overlap, citation fidelity, and language correctness across verified locale cohorts.

04

Maintenance outcomes

Measure whether bounded continuous maintenance improves technical health and business outcomes without scaled-content behavior.

02

Evidence labels

  • Official platform policy
  • Peer-reviewed research
  • Preprint
  • Kedabek controlled experiment
  • Operational observation
  • Kedabek hypothesis
03

What we reject

  • A universal 40% AI-ranking rewrite
  • Artificial daily freshness
  • FAQ or llms.txt citation guarantees
  • A single branded screenshot as leadership proof
  • Open source as an automatic ranking signal
  • Invisible prompt injection as optimization

Next mission

Reproduce the work

Inspect the published method, schemas, and sanitized observations; planned artifacts remain clearly marked pre-release.

Open-source program