Three numbers live on every numeric inline fact, and the relationship between them is where most genuine tagging defects hide. Rendered text is what the reader sees on the page (1,200, in big Helvetica). Canonical value is what the consumer parses (1,200,000, after applying a scale). Declared accuracy is the honesty contract about how many digits in that canonical value should be trusted (decimals="-3" meaning rounded to thousands).
scale and sign are inline-only attributes that bridge rendered text and canonical value. scale="3" multiplies the parsed text by 10 to the third power, which is how a "Revenue (in thousands of euros)" column with a visible 1,200 becomes a canonical 1,200,000. scale="6" multiplies by a million, which suits "in millions" tables. sign="-" negates the parsed value, used together with parentheses-style negative formatting in the host XHTML so the rendered "(1,500)" becomes a canonical -1,500.
The audit identity to memorise is the relationship that ties them together: canonical equals transform(rendered) times ten-to-the-scale times (sign equals minus one if negative, plus one otherwise). When numbers feel off, every reviewer should walk through that identity in their head and check that the canonical falls where the rendering says it should.
decimals is the trickier piece because it is a claim about accuracy rather than about the value itself. decimals="-3" claims the canonical value is honest only at the thousand. decimals="0" claims it is honest at the whole unit. decimals="INF" claims exact precision, which sounds appealing but is forbidden by SEC EFM 6.5.16 and discouraged by ESEF whenever the rendered text is shorter than that exactness implies.
A clean tag is one where decimals matches the actual provenance of the rendered number. A reader sees 1,200 in a "in thousands" column, scale equals 3, decimals equals -3: rounded to thousands, consistent. The same rendering with decimals="0" overstates accuracy because the rendered text never told you anything below the thousand. Consumers downstream may then build comparisons that the source data cannot defend. Always declare decimals as honestly as the rendered figure permits, and then the audit identity tells you the rest.
Three numbers live on every numeric inline fact: rendered text (what the reader sees), declared accuracy (decimals), and canonical value (what consumers parse).
scale and sign are inline-only attributes that shift the parsed text into the canonical domain. scale='3' multiplies by 1,000; sign='-' negates.
decimals declares accuracy honestly. decimals='INF' is forbidden by SEC EFM 6.5.16 and discouraged by ESEF when rendered text is shorter than INF claims.
A common AI defect: rendered text shows '1,200' while scale='3' and decimals='-3'. Canonical value is 1,200,000 with rounding to thousands. That matches. Now consider rendered '1,200' with scale='6' and decimals='0'. Canonical is 1,200,000,000 reported as exact thousands... unrecoverable.
Audit identity to memorise: canonical = transform(rendered) * 10^scale * (sign == '-' ? -1 : 1). The reviewer applies this every time numbers feel off.
<!-- Reader sees "1,200"; canonical is 1,200,000; rounded to thousands -->
<ix:nonFraction name="ifrs-full:Revenue"
contextRef="c-2025"
unitRef="u-eur"
decimals="-3"
scale="3"
format="ixt:numdotdecimal">1,200</ix:nonFraction>3 questions. Druk op 1 tot 4 om te kiezen, Enter om te bevestigen.
Door deze quiz af te ronden, markeer je deze sectie als voltooid.
Tip: kies met 1 tot en met 4 en bevestig met Enter.