A FIELD GUIDE TO BETTER DATA

Follow your curiosity.

24 notes on preparing better material, one useful question at a time.

Start with the question

Ask a better question ↗

Turn a broad request for insight into a question your records can actually answer.

2 min read

Define the decision ↗

State what someone will decide before spending time producing an answer.

2 min read

Choose a useful measure ↗

Give every business number a definition that another person can reproduce.

2 min read

Set the boundary ↗

Say exactly which records, dates and parts of the business an answer covers.

2 min read

Know your sources

Data inventory ↗

Make a short map of the records your business holds and who understands them.

2 min read

Source of truth ↗

Agree which record governs a particular fact when systems disagree.

2 min read

Provenance ↗

Keep enough information to trace a claim back to the material it came from.

2 min read

Permission and purpose ↗

Decide what information is needed and who may use it before sharing it with a tool.

2 min read

Put things in order

One row one thing ↗

Decide what a row represents, then keep that meaning consistent throughout the table.

2 min read

Unique identifiers ↗

Use stable references to distinguish records and connect them without guessing from names.

2 min read

Data dictionary ↗

Write down what your fields mean so people and tools can interpret them consistently.

2 min read

Consistent categories ↗

Give repeated labels a shared meaning while preserving distinctions that matter.

1 min read

Check the quality

Missing values ↗

Treat a blank as a question to investigate before deciding what it means.

2 min read

Duplicates ↗

Separate repeated records from real events that happen to look alike.

2 min read

Freshness ↗

Check whether information is recent enough for the decision you want to make.

2 min read

Reconciliation ↗

Compare records that should agree and explain the differences instead of hiding them.

2 min read

Give AI context

Context ↗

Supply the purpose, definitions and limits that make a piece of information interpretable.

2 min read

Chunking ↗

Divide material into useful sections that retain enough meaning to stand on their own.

2 min read

Retrieval ↗

Find the source material that is relevant to a question before trying to answer it.

2 min read

RAG ↗

Use retrieved source material to support an AI answer, then check whether the answer stays within that evidence.

2 min read

Keep it useful

Human review ↗

Give a responsible person a clear way to check evidence, assumptions and consequences.

2 min read

Version history ↗

Keep a recoverable record of what changed, when it changed and why.

2 min read

Feedback loops ↗

Turn real questions, corrections and failed searches into specific improvements.

2 min read

Publish with care ↗

Make publishing an explicit editorial step between private working notes and a public site.

2 min read