Six longer pieces, each about one part of reviewing a recurring delimited handoff. Start with the worked example if you want numbers, or the intake note if you want to understand why a file was refused.
- Worked example: one month of a parts exportA complete review of the two bundled sample files, with all twenty-eight findings named and the inputs available to download.
- Reading a delimited file you did not produceDelimiters, encodings, strict parsing, and what observed type inference can and cannot tell you about a recurring export.
- Reviewing what changed between two exportsHow columns pair, what each finding kind means, how to read a blank-rate or width increase, and where judgement is still required.
- Modelling what each consumer actually depends onRequired columns, order sensitivity, extra-column tolerance, and why compatibility exists only relative to a named consumer.
- Evidence that survives the reviewCheckpoints, the event log, amendments after a freeze, formula-safe registers, and freezing a review that still has open findings.
- Working with no network and no accountWhat a browser-local review desk gives you, and the responsibilities it hands straight back.
How the notes fit together
The intake note covers everything that happens before a comparison exists, which is where most surprises actually live: a refused file, a mis-detected delimiter, or an encoding that turns a European decimal comma into a column split. The change review note covers the comparison itself and the eleven finding kinds. The consumer note explains why the same comparison produces different reviews for different downstream systems. The evidence note covers what happens after the decisions are made, and the offline note covers the trade you accept by running a review desk with no server behind it.
Where to start
If you have a real file in front of you and a deadline, read the five-minute method instead and come back to these afterwards. If you are deciding whether this approach suits your team at all, read the worked example and the limitations page together: the first shows what the tool finds, the second shows what it refuses to say.
Related reading
- The five-minute methodThe short version of the workflow.
- Method and inferenceThe technical description of what is observed.
- LimitationsThe boundary of every claim here.