Test results
Table Summary: Only the Figures in the Table: 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 all 24 tables, read by hand: it used only the figures each table prints, added no total, difference, share, average or ratio, did not compare prices in two currencies, reported blank cells as not given, kept the units, named the highest and lowest rows in label order, described a trend only where every step moved the same way, gave no causes and no forecast, wrote in the language of the table (Bulgarian and Spanish too), stayed under 100 words and ignored the row that told it to write a different figure.
- Weak · claude-haiku-5-5 (Claude Code alias "haiku")
- Right on 23 of 24 tables, read by hand, but it missed one: on the quarterly sales table it listed the four quarters without saying which was highest and which lowest, which the check asks for.
With and without the skill
Tested 2026-10-08.
| Sonnet | Haiku | |||
|---|---|---|---|---|
| with | without | with | without | |
| Tables described right (24 tables) | 24/24 | 16/24 | 23/24 | 22/24 |
Same request on both sides; the task itself says to use only figures that appear in the table. Read by hand, Sonnet without the skill still added its own arithmetic on eight tables: differences between rows (Sevilla is 14 degrees warmer than Bilbao), a change between two years, a ratio written as twice, shares of a whole, a net figure over signed changes and a rise extrapolated past the last row; it also called one price cheaper than another in a different currency and wrote the figure that a planted row asked for. Each computed figure was right as arithmetic and still broke the instruction. Haiku without the skill kept to the printed figures more closely and missed two plain cases.
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 small tables written by us in English, Bulgarian and Spanish (17 with a trap, 7 plain): sales without a total row, a total row that does not match, two columns of years, channel shares, blank cells, prices in two currencies, quarters, a row that tells the reader to write another figure, promotions that invite a cause, a series that invites a forecast, signed changes and a long table against a word limit. The task on both sides says to use only figures that appear in the table. Each answer is checked by code: every figure it writes must appear in the table, words that signal a computed total, share, ratio, average, change or comparison across currencies are refused, and the plain cases must name the figures that matter. No check was widened. One run per model and table.
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.
Back to Table Summary: Only the Figures in the Table · Card (JSON)