ContentGrapher
ContentGrapher
research/concept-validity-study/data
The concept validity studyData

The data

Per-language verdicts

The majority verdict for each language across the five concepts. Every cell resolved to go. The Pages column is the number of completed analyses behind each row; rows marked * (French and Chinese) reached four pages rather than five, below the corpus floor.

LanguagePagesBoundaryInteg.DepthRoleDim.
French4 *gogogogogo
German6gogogogogo
Spanish5gogogogogo
Italian6gogogogogo
Dutch5gogogogogo
Japanese5gogogogogo
Korean5gogogogogo
Chinese4 *gogogogogo

* Below the pre-registered five-page floor due to crawler blocking; verdicts shown for completeness.

Every page

All 40 native pages, with the retrieval role the tool detected, its coverage score, and the concepts that fell short of a clean go. Where the result reads “all go,” every one of the five concepts was a clean go; “loc” marks needs-localization.

LangSourceTypeRoleCov.Concepts below go
FReditions-tissot.frExplainerexplain25%all go
FRptitchef.comGuideguide55%all go
FRselectra.infoComparecompare58%all go
FRvidal.frExplainerexplain39%integration loc
DEeasycredit.deExplainerexplain68%all go
DEheise.deExplainerexplain42%all go
DEnetzwelt.deGuideguide7%integration, depth, dimensions loc
DEobi.deGuideguide57%all go
DEtechbook.deComparecompare75%all go
DEverbraucherzentrale.deCompareguide63%all go
ESadslzone.netGuideguide47%all go
ESconsumidorglobal.comGuideguide66%all go
ESocu.orgComparecompare63%all go
ESpccomponentes.comExplainerguide39%all go
ESredeszone.netGuideguide52%depth loc
ITaltroconsumo.itCompareevaluate43%all go
ITfocus.itGuideexplain43%role loc
ITgeopop.itExplainerexplain45%all go
IThtml.itGuideguide80%depth loc
ITmoney.itComparecompare45%all go
ITpunto-informatico.itExplainerexplain62%all go
NLgratissoftwaresite.nlGuideguide40%all go
NLid.nlExplainerexplain62%all go
NLthuisarts.nlExplainerexplain72%role loc
NLunitedconsumers.comComparecompare5%all go
NLveb.netCompareevaluate51%all go
JAfurusato-tax.jpExplainerguide57%role loc
JAhyponex.co.jpGuideguide67%all go
JApropane-npo.comCompareevaluate51%all go
JAsmbcnikko.co.jpComparecompare40%depth loc, role no-go
JAtaisho.co.jpExplainerexplain50%all go
KO10000recipe.comGuideexplain25%depth, dimensions loc
KObanksalad.comExplainerexplain75%all go
KOohou.seComparecompare48%all go
KOoppadu.comGuideexplain63%all go
KOtossbank.comComparecompare67%dimensions loc
ZH12sporting.comGuideguide33%all go
ZHcsdn.netExplainerexplain50%all go
ZHsspai.comGuideguide45%role loc
ZHzol.com.cnGuideguide28%all go

Coverage is the page’s overall completeness score, reported by the pipeline and independent of the validity verdicts. Eleven of the 40 pages carried at least one concept below go; the rest were clean across all five.

On confidence intervals

We did not compute bootstrap confidence intervals. With four to six pages per language, the sample is too small for a meaningful resampled interval, and the verdicts are categorical rather than continuous. We report raw majority counts and let the per-page evidence carry the weight. The verdicts also rest on a single Opus 4.8 judge with no human calibration, a trade-off set out on the methodology page.

← OverviewMethodology →