SearchScore Labs

We do our homework in public.

Every change to the SearchScore intelligence engine is a registered experiment. We publish the hypothesis and method before we have the result — then we publish the result, whether it confirms our idea, finds nothing, or proves us wrong. This is the whole registry.

public experiments

pre-registered

nulls / rejections published

33%

of finished experiments published

open hypotheses

◆Recommendation experiments

How the engine turns a question into a trustworthy recommendation.

EXP-0001 Rejected

Coverage-discount guard for homograph false positives

Discounting a recommendation’s confidence by how much of the query the matched pattern explains ("coverage") will eliminate confident homograph false positives (e.g. "budget my salary" → Crawl Budget).

EXP-0002 Complete

Founding recommendation-quality baseline (benchmark v1)

A frozen 1001-query benchmark can establish a reproducible baseline for recommendation quality and confidence calibration that every future release must beat.

EXP-0003 Pre-registered

Neural embeddings for semantic retrieval (A/B vs baseline)

Dense neural embeddings will eliminate the homograph FP class and improve MRR/NDCG without hurting in-domain recall, lowering ECE below 5%.

▤Framework experiments

Which SearchScore frameworks actually work, and how well.

EXP-0004 Pre-registered

Retiring never-surfaced frameworks does not reduce coverage

The 6 frameworks that never surface on benchmark v1 (SS-FW-005/006/007/009/018/023) can be retired or merged without reducing recommendation coverage or quality.

◈AI Visibility experiments

What makes a site visible to — and cited by — AI answer engines.

EXP-0005 Pre-registered

Does an llms.txt file predict higher AI visibility?

Sites that publish an llms.txt file achieve a materially higher SAVI (AI Visibility Index) than sites without one.

◇GEO experiments

Generative Engine Optimisation: the mechanics of ranking in AI search.

EXP-0006 Pre-registered

Does Organization schema raise the GEO entity/authority score?

Sites with valid Organization schema score higher on the GEO entity/authority sub-score than sites without it.

◷Hypothesis library

Questions we intend to answer. Each becomes a registered experiment; the outcome is published regardless of direction.

ID	Category	Question	Status
HYP-001	AI Visibility	Does publishing an llms.txt file predict higher AI visibility? 69.7% of sites have no llms.txt (benchmark)	experiment-registered → EXP-0005
HYP-002	GEO	Does Organization schema raise the GEO entity/authority score? 52.2% missing Organization schema	experiment-registered → EXP-0006
HYP-003	AI Visibility	Do sites that allow AI crawlers get cited more often by AI answer engines? 6.9% of sites block at least one major AI crawler	open
HYP-004	GEO	How much of the GEO score gap between top-3 Google rankers and the rest is explained by technical vs content factors? 43% of top-3 rankers score below 30 for AI visibility	open
HYP-005	Framework	Which framework produces the largest measured ranking uplift per completed recommendation? outcome ledger live-and-empty until canary traffic	awaiting-outcomes
HYP-006	Recommendation	Does confidence calibration (ECE) predict real-world recommendation acceptance? baseline ECE 13.8% recorded; acceptance needs event data	awaiting-outcomes
HYP-007	AI Visibility	Is the AI-Ready tier (SAVI 80+) growing or shrinking quarter over quarter? only 0.05% reach the AI-Ready tier today	open
HYP-008	GEO	Does adding first-party author/EEAT signals move the GEO authority sub-score? KOS authority & trust domain	open

Methodology and integrity standards: every figure carries a sample size and a 95% confidence interval; the frozen benchmark is the yardstick; no result is hidden. See the SearchScore research programme and intelligence charter.