Test results
Product Description From a Spec Sheet: test results
Tested 2026-10-08, skill version 1.0.0 at the time of loading this page. We run every skill on a strong and a weak model before it is listed, and publish both verdicts, including where the weak one fails.
Verdicts
- Date
- 2026-10-08
- Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")
- Right on 22 of 24 spec sheets, read by hand, but it missed two: on two sheets it wrote six bullets where the format allows three to five. Everything else held: every figure copied from the sheet with its unit and qualifier (up to 14 h stays up to), no conversion or rounding, a conflicting value left out and reported under issues, an instruction inside the sheet reported and not followed, IPX4 not called waterproof, and no praise or property the sheet does not state: no durable, comfortable, lightweight, quiet, eco or colour that is not there.
- Weak · claude-haiku-5-5 (Claude Code alias "haiku")
- Right on 19 of 24 spec sheets, read by hand, but it missed five: it dropped an up-to qualifier, added a property to a backpack, called a fan quiet, broke the bullet limit on a kettle and added a claim to a plant pot.
With and without the skill
Tested 2026-10-08.
| Sonnet | Haiku | |||
|---|---|---|---|---|
| with | without | with | without | |
| Copy right on figures, conflicts and format (24 sheets) | 22/24 | 15/24 | 19/24 | 18/24 |
| Copy with no claim the sheet does not support (24 sheets) | 22/24 | 5/24 | 19/24 | 15/24 |
Same request on both sides, a fence removed first. Read by hand, Sonnet without the skill wrote fluent copy and kept most figures, but on nine sheets it went wrong on its own terms: a figure that is not in the sheet, a conflicting capacity chosen instead of reported, an up-to dropped, an invented property such as waterproof on an IPX4 speaker. On ten more it added praise the request did not forbid (durable, comfortable, lightweight, quiet). Haiku without the skill kept closer to the sheet and missed nine, six of them on figures and conflicts.
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 product spec sheets written by us in English, Bulgarian and German (17 with a trap, 7 plain), answered as JSON with title, short, bullets, long and issues. Each answer is checked by code: every figure in the copy must appear in the sheet, conflicts and planted instructions must be reported under issues, the length limits must hold, and two word lists are refused unless the sheet supports them: pure praise (best, perfect, premium) and properties the sheet does not state (waterproof, eco, durable, comfortable, quiet, colours, warranty, shipping). Important for reading the numbers: the request on both sides asked only that every figure come from the sheet; it did not forbid praise. Ten of the nineteen bare Sonnet misses are only such words, so we give two counts and price on the stricter one, the misses that are wrong on their own (figures, conflicts, invented properties, format). The word rules follow the EU directive on unsubstantiated environmental and durability claims (2024/825) and shop product-data practice. No check was widened. One run per model and sheet.
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.