CareerTalkLab / Grain First
2 Live · 71 In DevelopmentGrain First — Data Analytics
Practical measurement, denominator hygiene, and bottleneck analysis for data and software careers. Every unit progresses from Level 1 (Modelled) and Level 2 (Supported Practice) to live cohort cold-transfer (L3) and peer teaching (L4).
Grain, Rows & Silent Exclusions: What Does One Row Represent?
Audit a 12-row Driving Test booking log (`B01`–`B12` across candidates `C01`–`C08`), reconcile attempt-level vs. candidate-level pass rates, and earn Level 1 & Level 2 attainment. Also paired with Business English C1-EXEC-EVD-01.
Measurement Levels & Valid Summaries: What Summaries Can You Trust?
Classify Nominal, Ordinal, Interval, and Ratio columns on a 15-row branch Feedback export (RSP-01–RSP-15), catch nominal code traps, and resolve the North vs. South rating rank flip for Level 1 & Level 2 attainment.
Track 0 · Orientation & Analytical Framing
Establish workspace setup, question framing, and analytical workflow habits before touching raw tables.
The Decision Contract
Can state what question an analysis answers, who acts on it, and what changes as a result
Isolated Analysis Environments
Can build an isolated, reproducible analysis environment from nothing
Version Control for Analysis
Can version an analysis: branch, commit, diff, revert, and recover a deleted file
Zero-Question Reproducibility
Can publish a repository a stranger can clone and reproduce without asking questions
Track 1 · Evidence, Grain & Denominators
Identify table grain, distinguish event rows from entity counts, and audit silent denominator exclusions.
Unit of Analysis & Row Grain
Can state the unit of analysis and what exactly one row represents in an unfamiliar table
Measurement Levels & Valid Summaries
Can classify variables by measurement level and say which summaries are legitimate for each
Distributions, Spread & Skew
Can describe a distribution by centre, spread and shape, and say when the mean misleads
Fair Denominators & Rates
Can choose a denominator that makes a comparison fair, and convert counts to rates
Sampling Frames & Unrepresented Populations
Can identify how a sample was drawn and name the populations it does not represent
Signal vs. Routine Variation
Can distinguish variation from change, and say how much movement is unremarkable
What p-Values & Intervals Claim
Can state what a p-value and an interval estimate do and do not claim
Confounding & Simpson's Paradox
Can detect confounding and demonstrate a Simpson's-paradox reversal in real data
Survivorship, Selection & Measurement Bias
Can audit an analysis for survivorship, selection and measurement bias
Track 2 · Spreadsheets & Tabular Reconciliation
Build auditable spreadsheet checks, pivot reconciliations, and explicit grain keys without silent row loss.
Raw Sheet Profiling
Can profile a raw sheet: row count, types, missingness, duplicates, impossible values
Non-Destructive Cleaning & Audit Trail
Can clean without destroying the original, leaving a documented transformation trail
Cross-Sheet Matching & Row Accounting
Can match records across two sheets and account for unmatched, excluded and multiplied rows
Grain Aggregation & Source Reconciliation
Can aggregate to any grain and reconcile the total back to the source
Separated Inputs & Formula Models
Can build a calculation model whose inputs are visibly separated from its formulas
Question-Driven Chart Selection
Can select a chart form from the question being asked rather than the data's shape
Track 3 · Relational Grain & SQL
Prevent join fan-out, write grain-preserving aggregations, and verify row-count invariants in SQL.
Single-Table Filtering & Row Prediction
Can filter, sort and limit a single table and predict the row count before running it
GROUP BY & Result-Set Grain
Can aggregate with GROUP BY and state the grain of the result set
Primary Key Discovery & Uniqueness Proofs
Can identify the primary key of an unfamiliar table and prove its uniqueness
Join Mechanics & Row Fate Prediction
Can predict which rows each join type keeps, drops and duplicates, before running it
Detecting & Repairing Join Fan-Out
Can detect and repair join fan-out that silently inflates a sum
Filtering Aggregates: HAVING vs. WHERE
Can filter on aggregates and articulate why WHERE cannot do it
Readable Multi-Step Queries with CTEs
Can compose a multi-step query with CTEs that another analyst can read unaided
Window Functions & Period-over-Period Analysis
Can rank, offset and window over partitions to answer period-over-period questions
Conditional Logic & Defensible Bucketing
Can branch logic with CASE and bucket a continuous variable defensibly
Pre-Delivery Query Validation Checks
Can validate a result with row-count, total and spot checks before anyone else sees it
Undocumented Schema Exploration & Data Dictionaries
Can explore an undocumented schema and produce a data dictionary for the tables in scope
Track 4 · Analytical Data Modeling
Design clean fact and dimension grains, snapshotting rules, and metric definitions that survive scale.
Reading & Reconstructing ER Diagrams
Can read an entity-relationship diagram, and reconstruct one from an undocumented database
Normalising Flat Files into Entities
Can decompose a flat file into entities and relationships, and say what that buys
Facts, Dimensions & One-Sentence Fact Grain
Can distinguish facts from dimensions and declare the grain of a fact table in one sentence
Slowly Changing Attributes & Historical State
Can explain how history is preserved when an attribute changes, and what breaks when it isn't
End-to-End Lineage & Pipeline Failure Points
Can trace the path data takes from source to report and name where it can break
Unambiguous Metric Specifications
Can define a metric precisely enough that two analysts compute it identically
Loud-Failing Data Assumption Tests
Can write data tests that fail loudly when an assumption stops holding
Track 5 · Python for Reproducible Analysis
Automate validation, cohort checks, and reproducible notebooks over messy operational datasets.
Scripts vs. Notebooks as Analysis Containers
Can run an analysis as a script and as a notebook, and say when each is the right container
Dataset Loading, Profiling & Shape Reconciliation
Can load and profile a dataset and reconcile its shape against what was expected
Dataframe Selection, Filtering & SQL Equivalents
Can select, filter and derive columns, naming the SQL equivalent of each operation
Grouped Aggregation & Source Total Verification
Can group and aggregate to a declared grain and verify totals against the source
Dataframe Joins & Fan-Out Prevention
Can join dataframes and catch the same fan-out failures learned in SQL-05
Reshaping Long vs. Wide Form
Can reshape between long and wide form and say which the analysis actually needs
Temporal Alignment, Resampling & Timezones
Can parse, align and resample dates, and survive timezones and period boundaries
Statistical Computation & Substantive Interpretation
Can compute the Track 1 statistics in code and interpret the output, not merely produce it
Descriptive Linear Models & Scope Limits
Can fit and read a linear model as description, and state what it cannot support
Clean-Machine Reproducible Packaging
Can package an analysis so a colleague reproduces it on a clean machine
Track 6 · Executive Analytical Communication
Translate reconciled numbers into decision memos that state grain, uncertainty, and next owner action.
Comparison-Driven Visual Encoding
Can select an encoding from the comparison the question demands
Preattentive Attributes & Visual Hierarchy
Can use preattentive attributes to direct attention to the finding, not decorate around it
Honest Axes, Scales & Annotations
Can design an axis, scale and annotation set that cannot mislead a hurried reader
Exploratory Dashboards vs. Explanatory Narratives
Can build an exploratory dashboard and keep it distinct from an explanatory narrative
The Five-Part Analysis Memo
Can write an analysis memo: question, method, finding, caveat, recommendation
Decision-Preserving Caveats & Limitations
Can state a finding's limitations without dissolving the recommendation
Non-Technical Stakeholder Briefing & Defence
Can present to a non-technical stakeholder and hold up under the challenge questions
Cross-Tool BI Rebuild & Reconciliation
Can rebuild one prior analysis in a second BI tool and account for what changed
Track 7 · AI-Assisted & Agentic Analytics
Audit LLM-generated SQL/Python against ground-truth grain invariants and build reliable verification harnesses.
Schema- & Grain-Grounded Model Briefing
Can brief a model with schema, grain and constraints so its first draft is worth reading
Diagnosing AI Analytical Failure Modes
Can recognise the standard failure modes: invented joins, wrong grain, silent filtering, confident bad statistics
Verifying Generated SQL by Independent Recomputation
Can verify generated SQL by independent recomputation rather than by reading it
Rewrite vs. Repair Judgement
Can judge when generated code is faster to rewrite than to repair
AI Generation & Verification Audit Trails
Can keep an audit trail of what was generated and how each part was checked
Setting Hard Boundaries on Model Delegation
Can name the steps in an analysis where a model should not be used at all
Track 8 · End-to-End Capstone
Deliver a complete, peer-defended analytical investigation from raw operational tables to executive action.
Stakeholder-Backed Domain Question Sourcing
Can source a question from a domain you know, with a named person who wants the answer
Wild Data Acquisition & Gap Documentation
Can acquire and document data of unknown quality, including what could not be obtained
End-to-End Modelling, Analysis & Gate Validation
Can model, analyse and validate end to end against the Track 3–5 gates
Workflow-Native Stakeholder Delivery
Can deliver the finding in the form the stakeholder actually uses
Adversarial Analytical Defence
Can defend the analysis under adversarial review without retreating from it
Parallel Track · Professional Transfer & Portfolio
Turn completed analytical units into public artifacts, interview walkthroughs, and peer teaching sessions.
The 90-Second Project Write-Up
Can write a project up so a reader grasps question, method and finding in 90 seconds
Decision-Indexed Portfolio Curation
Can maintain a portfolio where every item names the decision it informed
Targeted Outreach & Answerable Asks
Can write targeted outreach with a specific, answerable ask and document respectful follow-through
Unrehearsed Technical Project Walkthrough
Can hold a technical conversation about a project you did not rehearse for
Live Narrated SQL Problem Solving
Can solve a live SQL problem while narrating your reasoning aloud
Ambiguous Case Interview to Recommendation
Can work a case interview from ambiguous prompt to defensible recommendation
Range- & Walk-Away-Grounded Offer Negotiation
Can negotiate an offer with a stated range and a walk-away number