Learn
Most of what goes wrong with a data file goes wrong in a handful of ways, and each of them has a name. This is where I write those down: what the symptom looks like, what is actually causing it, and which tool fixes it.
Everything here is hand-written, dated, and signed. Where a page states a number, it is a number from the code or from a file I looked at, not a round figure that sounded right.
Fix what is broken
Fourteen troubleshooting guides. Symptom, cause, fix, and a link into the tool that repairs it.
CSV file too large to open
Excel silently drops rows past 1,048,576 and Sheets stops at 10 million cells. What actually breaks, and five ways past it.
CSV has extra blank rows
A blank row between every row, the whole file on one line, or empties at the end. All three are line-ending problems.
Why your CSV opens wrong in Excel
Six things Excel does to a CSV without asking, from one-column imports to numbers stuck as text, and the fix for each.
How to fix CSV encoding problems
Mojibake, the replacement character and the BOM are three different failures. One is recoverable, one is not, and one is cosmetic.
CSV delimiters explained
Comma, semicolon, tab and pipe, why sniffers get it wrong, and how to choose a separator that survives the trip.
How to keep leading zeros
02134 becomes 2134 and still looks like a plausible zip code. How to keep them, and when they can honestly be rebuilt.
Fix 1.23E+15 in CSV files
The 15 significant digit ceiling, how to tell a display problem from real rounding, and which identifiers are at risk.
Excel changed your CSV dates
Locale guessing, gene names that become dates, serial numbers, and a distribution check that catches a misparse instantly.
Excel formulas lost in CSV
A full accounting of what Save As CSV discards: formulas, sheets, formats, validation, and the merged-cell mess it leaves.
Quotes, commas and escaping
The RFC 4180 rules in four lines, then the four ways writers break them, including the space after the comma.
JSON to CSV with nested arrays
Objects flatten. Arrays force a choice between exploding and joining, and picking wrong is how a revenue report doubles.
Parquet vs CSV: when to switch
Five signals that the format is costing you something, and four cases where plain CSV is still the better answer.
Excel's 1,048,576 row limit
Where the number comes from, what Excel does when a file exceeds it, and five routes past it that keep all the data.
Why my CSV has one column
Six causes ranked by likelihood, and three terminal commands that tell you which one you have in under a minute.
Understand the formats
Reference material for the vocabulary and the format choices behind the guides.
Data file glossary
Eighty-seven terms across seven groups: the format itself, types and values, columnar formats, shaping, analysis, engines, geospatial. Sixty-five of them link to a tool.
Parquet vs CSV, side by side
The storage models and type systems compared, with a table of what each format can and cannot express.
Every format we read
CSV, TSV, Excel, JSON, JSONL, XML, Parquet, Arrow, Avro, SQLite, DuckDB, PDF, DOCX and six container formats.
Product glossary
The interface vocabulary: pipeline, operation, view, column explorer, condition builder. A different set of words from the file glossary.
Get something built
Longer walkthroughs that end with a working artifact rather than a fixed file.
Building a recurring CSV report
The monthly export that arrives slightly different every time. How to build a pipeline that survives it, and what to check before you send the numbers.
CSV variance analysis
Comparing this period against last, ranking the categories that drove the change, and handling the new and disappeared ones honestly.
Five ways to clean messy data
Missing values, inconsistent text, duplicates and wrong types, without writing any code.
Joining multiple CSV files
Matching two exports on a key column, choosing the join type, and spotting the row-count multiplication that means you chose wrong.
Auditing a file's data quality
Completeness, distinct rates, mixed value shapes, duplicates and outliers, per column, before you build anything on top of the file.
Profiling a file column by column
Types, ranges, blanks and the values that appear most, laid out one column at a time, for the first ten minutes with a file you did not create.
Go deeper
How the tool is built, how it compares to the alternatives, and what changed when.
Engineering write-ups
Five posts from the actual codebase, with real code, real numbers and the bugs left in.
Alternatives and comparisons
Seven fair comparisons against other browser-based data tools, each one saying what the other tool is good at first.
Changelog
What shipped, month by month, reconstructed from the git history and written by hand. Includes what is next.
The blog
Seventy-one posts on cleaning, exploring, transforming and workflows, organized by what you are trying to do.