Proof standard
Every claim retains its task population, artifact and version, intervention, comparator, grader, failure states, human evidence layer, and evidence cutoff.
Researchers and evaluators
Match every claim to its task, grader, and evidence layer.
For visualization, HCI, and AI researchers choosing benchmarks, locating gaps, and designing studies that connect technical output to human consequence.
Decision to support
What is actually measured, what remains a demonstration, and which next study would change the synthesis?
Working output
A study or audit design that targets the first missing consequential receipt instead of adding another disconnected metric.
Composed research
5 source sections, assembled for this decision.
These are the original sections from the visual research experiences, not summaries or reference cards. Expand each step to read it here, or open its stable page for direct linking.
01 · SynthesisResearch synthesisWhat does current evidence establish—and what does it not?
Locate convergence, disagreement, counterevidence, and corpus limits.
Open this section as its own page →02 · EvidenceEvidence ladderWhich meaning of works has actually been observed?
Type each result by the layer it actually observes.
Open this section as its own page →03 · EvaluationWhat the benchmarks measureWhich task-and-grader result produced each score?
Compare task-and-grader pairs without turning them into one leaderboard.
Open this section as its own page →04 · Research agendaResearch gap ledgerWhich missing receipt would materially change the current answer?
Choose a bridge whose result would change a claim or decision.
Open this section as its own page →05 · ForecastForecasts as score movementsWhich score movements are plausible and resolvable?
Turn future claims into dated, resolvable tests rather than narrative extrapolation.
Open this section as its own page →Ask before acting
Three questions that keep the route honest.
- 01
Does the evidence join the exact artifact, version, exposure, grader, and human outcome being claimed?
- 02
Are unknown, blocked, partial, negative, and rejected results preserved in the denominator?
- 03
Would the proposed study change a decision, or only add another adjacent capability result?
Focused analysis
Complete arguments to keep nearby.
- Benchmark libraryStable task-and-grader profiles for the main evaluations in the synthesis.
- Benchmark crosswalkA detailed comparison of benchmark families and validity boundaries.
Named evidence
Reference pages connected to this route.
- ChartographyA deliberately difficult set of practitioner-authored professional chart-reading tasks.
- RealChart2CodeA chart-to-code benchmark built from real source data, multi-panel tasks, and iterative refinement.
- PolyChartQA: multi-chart figuresA multi-chart scientific-figure benchmark whose name collides with a separate multilingual chart benchmark.
- POLYCHARTQA: multilingual chart QAA multilingual chart-question-answering benchmark distinct from the similarly named multi-chart scientific-figure dataset.