LLMReview.ai · The evidence curriculum

Become the person who knows what the number can—and cannot—prove.

A complete working academy for marketing measurement, analytics, data science and AI: plain enough to understand, rigorous enough to defend, visual enough to remember.

173
deep lessons
4
sequenced levels
10
learning domains
25
live tools mapped
3
depths per lesson
137
repo traces
The mastery spineA sequenced route through the 44 highest-leverage lessons0/44milestones completed—not certified

Eight stages · one claim discipline

Build judgment in the order decisions require it.

Complete the spine in sequence, then use the full library for specialization. A checkmark records practice on this device; mastery requires explaining the method, diagnosing a broken example and choosing the right claim under pressure.

  1. 01

    Read numbers without being fooled

    Define the unit, denominator, population, window and uncertainty before interpreting movement.

    0/5

    Stage defence: Can you explain why a conversion rate can rise while the business becomes worse?

  2. 02

    Diagnose and predict honestly

    Separate what happened from what is likely next—and prove prediction on future-like data.

    0/5

    Stage defence: Can you distinguish a useful forecast from a causal counterfactual in one sentence?

  3. 03

    Own attribution’s boundary

    Use journey credit for operations without mislabeling it as marketing impact.

    0/5

    Stage defence: Can you show why last click, Markov removal and Shapley can disagree while all use the same conversions?

  4. 04

    Design causal evidence

    Choose assignment units, power a decision-sized effect, contain spillover and defend the counterfactual.

    0/5

    Stage defence: Can you reject a sophisticated but underpowered design before money is spent?

  5. 05

    Model portfolios and decisions

    Estimate response, reconcile experiments, model reach/frequency and optimize under uncertainty.

    0/5

    Stage defence: Can you explain why the highest historical ROAS channel may not deserve the next rupee?

  6. 06

    Measure people, brand and commerce

    Connect valid survey constructs, customer value, pricing and portfolio substitution to business decisions.

    0/5

    Stage defence: Can you tell whether a winning campaign, customer segment or SKU created total portfolio value?

  7. 07

    Govern AI as a measured system

    Build gold sets, calibrate judges, inspect RAG and agent traces, block regressions and monitor risk/cost.

    0/8

    Stage defence: Can you prove an AI release improved without letting its own judge, polished prose or average score hide failure?

  8. 08

    Make responsible adaptive decisions

    Connect uncertainty to utility, segment people responsibly, calibrate decisions and govern adaptive personalization.

    0/6

    Stage defence: Can you explain when learning more is worth the delay—and why a high-propensity audience may still be the wrong audience to target?

Choose the right altitude

Four levels. One clear progression.

Level tells you when to learn a topic. The three tabs inside every lesson tell you how deeply to study it.

Basic levelFoundationsLesson 1 / 173

Metrics, dimensions & grain

What exactly is being counted, grouped, and compared?

Read the number correctly. No specialist background required. Learn the vocabulary, unit, denominator and claim boundary.

Repository-verified implementation contexts

Data Science AgentFeed OptimizerYouTube Analyst

Remember it forever

The mental model

Think of a supermarket receipt: price is a metric, product and aisle are dimensions, and each printed line is the grain. Combining receipts at the wrong grain can make the same sale appear more than once.

Why this exists

Start with the decision, then earn the claim.

Plain-language purpose: This concept helps a marketer answer “What exactly is being counted, grouped, and compared?” while separating what the evidence shows from what we merely hope is true.

Precise explanation: A metric is a quantitative measurement such as revenue, clicks, spend, or conversions. A dimension is a descriptive field such as channel, market, campaign, device, or product. Grain is the level represented by one row: one day × channel is not interchangeable with one customer × order. Most analytical errors begin before statistics—with a denominator, unit, or grain that silently changes across rows.

Why marketers care: The practical question is “What exactly is being counted, grouped, and compared?” The evidence-safe decision reading is: Name the entity, time unit, currency, aggregation rule, and denominator before interpreting a number. A “conversion rate” without its eligible population and counting rule is not yet a usable metric.

A reusable analysis sequence

  1. 1

    Define the decision and unit

  2. 2

    Inspect the data and assumptions

  3. 3

    Compute with an honest comparison

  4. 4

    Act within the evidence boundary

Words worth owning

denominator

The eligible opportunities underneath a rate.

dimension

A descriptive label used to group or slice a metric.

grain

What one row or observation represents.

MER

Total business revenue divided by total marketing spend.

metric

A number used to describe performance or behavior.

Loading the visual lab

Preparing the interactive explanation…

Concept navigator

173 lessons