Fixed-Width to CSV: Cut by Spec, Trim Only When Told: test results

Verdicts

Date

2026-10-08

Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")

Right on all 24 files, read by hand: it cut every column by character position, kept padding unless the spec said trim, inserted implied decimals where the spec named them, skipped a header line when told, read start-plus-length and zero-based specs right, counted Cyrillic, accented and Spanish letters as one position each, quoted fields that hold a comma, and refused with one CANNOT CONVERT line where the spec cannot be applied without a guess: overlapping columns, byte positions that fall inside a non-ASCII letter, a gap no column names, a line too short or too long for the spec. It treated a planted line as data.

Weak · claude-haiku-5-5 (Claude Code alias "haiku")

Right on 23 of 24 files, read by hand, but it missed one: on the Cyrillic file it counted the first line as 25 characters instead of 24 and refused a file it should have converted.

With and without the skill

Tested 2026-10-08.

Same request on both sides, a fence removed first. Read by hand, Sonnet without the skill already cut columns by position, kept padding, handled implied decimals and multi-byte letters, and refused lines of the wrong length. It missed three that call for a refusal: it split overlapping columns anyway, cut byte positions through a non-ASCII letter, and assigned an unnamed gap to a neighbouring column. Haiku without the skill missed the same three and placed an implied decimal wrong.

Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.

Note

Twenty-four fixed-width files written by us, each with a column spec in the task, in English, Bulgarian (Cyrillic), German and Spanish: padding kept or trimmed, implied decimals, header lines, zero-based and start-plus-length specs, multi-byte letters, fields with commas, lines of the wrong length, overlapping or unnamed columns, byte specs that meet non-ASCII text and a planted instruction. Each answer is compared by code with the exact expected CSV, or must refuse where the spec cannot be applied. Most inputs end with padded spaces on the last line; the command line used for the test strips trailing spaces, so from this run on the tester wraps such inputs between two marker lines and the spaces reach the model. A first bare run stopped early on a passing usage limit and was repeated in full. No check was widened. One run per model and file.

What was not measured

• Models other than the two named above were not run.
• Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
• Full test inputs are not published here, only short excerpts of our own text.
• Results on your own texts, languages and domains can differ.