#
α risk
The budgeted false positive: rejecting H0 when there is no effect. 5% means one false alarm in twenty true-null tests.
SYS/RES
Technical glossary
The vocabulary of CRO, analytics, SEO/SEA and AI engineering defined once, properly, with the pitfalls and the tool that uses it. Instant search, A–Z index, direct link to every term.
A
Agent Skills
On-demand instruction packs for AI agents: a SKILL.md, a licence, and just enough context to act without swallowing the whole library.
A
AOV
Average order value: revenue per order, not per visitor.
A
Attribution
The rule that decides which touch “owns” a conversion — last click, data-driven, or a blend nobody read.
B
Bayesian inference
Start from a prior, swallow the data, read a posterior — “chance that B beats A” is not a p-value in costume.
B
Bonferroni correction
Divide α by the number of tests to control the family-wise false-positive rate. Blunt, readable, often too conservative.
C
Context budget
The token window the agent can actually read. Anything past that is cut, summarised, or forgotten.
C
CAC
Customer acquisition cost: what you pay, on average, for a won customer — not a click, not a lead.
C
Cannibalisation
Two URLs on your site fighting the same intent — Google often picks one, and not always the right one.
C
Chi-square test
Compares observed counts to expected counts. The tool for 2×N tables, SRM, and anything that does not fit a two-proportion z.
C
Conversion
The event you decided to count as a success. Until it is in the tracking plan, it is not a metric, it is a wish.
C
Core Web Vitals
LCP, INP, CLS: three field measures of real experience, at the 75th percentile — not a lab Lighthouse score.
C
CPA
Cost per acquisition: campaign spend divided by the conversions that campaign is credited with.
C
CUPED
Controlled-experiment using pre-experiment data: strip variance with a pre-period covariate. Less noise, not a different lift.
C
CVR
Conversion rate: conversions divided by the denominator you chose (sessions, users, clicks).
C
CRO hypothesis
Because [evidence], if we [change], then [metric] will move by [size], measured by [setup]. Anything else is opinion.
C
Confidence interval
The range of effects still plausible given the data. A p-value without a CI is a decision without a magnitude.
E
E-E-A-T
Experience, Expertise, Authoritativeness, Trust: Google’s quality grid. An acronym, not a ranking factor named in the API.
L
Lift
The relative gap between B and A, (p_B − p_A) / p_A. +10% on a 2% CVR is +0.2 points, not a revolution.
L
LTV
Lifetime value: expected margin from a customer over their life, not GMV of the first basket.
M
MDE
Minimum detectable effect: the smallest effect your traffic, α and power can still see. Below that, you are flipping a coin.
N
Novelty effect
Users react to newness, not to the treatment. Monday’s lift is gone by Thursday.
P
p-value
Under H0, the probability of a gap at least as large as the one observed. It is not the probability that B is better.
P
Peeking
Looking at the test before the end and stopping when it suits you. Each look spends α you had not budgeted.
P
Progressive disclosure
Show the agent only what it needs now: description first, then the body, then files. Spare the context.
Q
Quality Score
Google Ads rating (1–10) mixing expected CTR, ad relevance and landing. A CPC lever, not SEO karma.
R
ROAS
Return on ad spend: attributed revenue / spend. 4 is not “×4 margin”. Check whether revenue is tax-in, margin, or GMV.
R
RPV
Revenue per visitor: AOV × CVR (at visitor unit). The continuous metric that absorbs both rate and basket.
S
Semantic cocoon
A graph of pages around a parent intent, explicit internal links, one role per URL. Not a keyword cloud.
S
Search intent
What the person wants to accomplish with the query: know, compare, do, go. The keyword is only the surface.
S
Statistical power
1 − β: the probability of detecting an effect if it is real, at a given MDE. 80% is not a law of nature, it is a convention.
S
SKILL.md
The single file of an Agent Skill: YAML frontmatter (name, description, licence) + Markdown body. That is the spec, not a decorative README.
S
SRM
Sample ratio mismatch: the observed split is not the promised split. Until χ² says otherwise, you do not read the uplift.
S
Sequential test
A boundary that moves with n, so intermediate looks do not explode α. SPRT is not a classic A/B looked at more often.
T
Tracking plan
The contract: which events, which properties, which deduplication, which tool receives them. Without it, analytics is folklore.
Z
z-test
The two-proportion test (or a mean, σ known): z = effect / standard error. The workhorse of binary A/B.