Dataset Viewer
Auto-converted to Parquet Duplicate
id
string
family
string
title
string
tests
string
level
int64
n_series
int64
n_points
int64
axis_scales
unknown
image
string
truth
string
spec
string
file_name
string
A01
line
Rail-jump lattice
colour-based tracing that swaps identity at a crossing
2
8
3,200
{ "panel0": { "x": "linear", "y": "linear" } }
images/A01.png
truth/A01.truth.json
spec/A01.spec.json
images/A01.png
A02
line
Dense stochastic bundle
token budget and sampling honesty
2
12
24,000
{ "panel0": { "x": "linear", "y": "linear" } }
images/A02.png
truth/A02.truth.json
spec/A02.spec.json
images/A02.png
A03
line
Near-parallel pair
sub-stroke separation between two series
2
2
1,000
{ "panel0": { "x": "linear", "y": "linear" } }
images/A03.png
truth/A03.truth.json
spec/A03.spec.json
images/A03.png
A04
line
Hairline on dark ground
ink thresholding and halo attribution
2
3
1,800
{ "panel0": { "x": "linear", "y": "linear" } }
images/A04.png
truth/A04.truth.json
spec/A04.spec.json
images/A04.png
A05
line
Dash-pattern identity
series identity carried by stroke, not hue
2
3
900
{ "panel0": { "x": "linear", "y": "linear" } }
images/A05.png
truth/A05.truth.json
spec/A05.spec.json
images/A05.png
A06
line
Step and spline in one frame
corner preservation vs dense resampling
2
2
416
{ "panel0": { "x": "linear", "y": "linear" } }
images/A06.png
truth/A06.truth.json
spec/A06.spec.json
images/A06.png
A07
line
Spike train
peak preservation under resampling
2
1
1,200
{ "panel0": { "x": "linear", "y": "linear" } }
images/A07.png
truth/A07.truth.json
spec/A07.spec.json
images/A07.png
A08
line
Sawtooth near Nyquist
honesty about recoverable resolution
2
1
3,000
{ "panel0": { "x": "linear", "y": "linear" } }
images/A08.png
truth/A08.truth.json
spec/A08.spec.json
images/A08.png
A09
line
Four-decade log crossing
scale type before geometry
2
6
2,400
{ "panel0": { "x": "linear", "y": "log" } }
images/A09.png
truth/A09.truth.json
spec/A09.spec.json
images/A09.png
A10
line
Broken axis
piecewise transforms
2
2
50
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" } }
images/A10.png
truth/A10.truth.json
spec/A10.spec.json
images/A10.png
A11
line
Twin axes, mirrored ranges
axis binding
2
2
600
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" } }
images/A11.png
truth/A11.truth.json
spec/A11.spec.json
images/A11.png
A12
line
Inline value labels on the marks
datum vs annotation
2
3
48
{ "panel0": { "x": "linear", "y": "linear" } }
images/A12.png
truth/A12.truth.json
spec/A12.spec.json
images/A12.png
A13
line
Legend inside the plot box
furniture exclusion
2
4
1,200
{ "panel0": { "x": "linear", "y": "linear" } }
images/A13.png
truth/A13.truth.json
spec/A13.spec.json
images/A13.png
A14
line
Clipped excursion
gap honesty
2
1
500
{ "panel0": { "x": "linear", "y": "linear" } }
images/A14.png
truth/A14.truth.json
spec/A14.spec.json
images/A14.png
A15
line
Band over its own line
boundary vs fill
2
3
900
{ "panel0": { "x": "linear", "y": "linear" } }
images/A15.png
truth/A15.truth.json
spec/A15.spec.json
images/A15.png
A16
line
Iso-luminant palette
greyscale-based separation
2
6
1,800
{ "panel0": { "x": "linear", "y": "linear" } }
images/A16.png
truth/A16.truth.json
spec/A16.spec.json
images/A16.png
A17
line
Gridline collision
furniture vs mark
2
2
800
{ "panel0": { "x": "linear", "y": "linear" } }
images/A17.png
truth/A17.truth.json
spec/A17.spec.json
images/A17.png
A18
line
Abbreviated irregular time axis
tick coverage
2
1
9
{ "panel0": { "x": "linear", "y": "linear" } }
images/A18.png
truth/A18.truth.json
spec/A18.spec.json
images/A18.png
B01
scatter
Overplotted cloud
recall under merging
2
1
5,000
{ "panel0": { "x": "linear", "y": "linear" } }
images/B01.png
truth/B01.truth.json
spec/B01.spec.json
images/B01.png
B02
scatter
Bubble with two channels
the size channel surviving
2
1
60
{ "panel0": { "x": "log", "y": "linear" } }
images/B02.png
truth/B02.truth.json
spec/B02.spec.json
images/B02.png
B03
scatter
Shape-coded groups
grouping carried by shape
2
4
160
{ "panel0": { "x": "linear", "y": "linear" } }
images/B03.png
truth/B03.truth.json
spec/B03.spec.json
images/B03.png
B04
scatter
Jittered categorical strips
jitter read as data
2
6
270
{ "panel0": { "x": "linear", "y": "linear" } }
images/B04.png
truth/B04.truth.json
spec/B04.spec.json
images/B04.png
B05
scatter
Log–log decades
two scale calls at once
2
1
220
{ "panel0": { "x": "log", "y": "log" } }
images/B05.png
truth/B05.truth.json
spec/B05.spec.json
images/B05.png
B06
scatter
Fit line with envelope
three marks, one hue
2
4
690
{ "panel0": { "x": "linear", "y": "linear" } }
images/B06.png
truth/B06.truth.json
spec/B06.spec.json
images/B06.png
B07
scatter
Two-dimensional error bars
recording uncertainty
2
1
14
{ "panel0": { "x": "linear", "y": "linear" } }
images/B07.png
truth/B07.truth.json
spec/B07.spec.json
images/B07.png
B08
scatter
Touching markers
blob splitting
2
1
120
{ "panel0": { "x": "linear", "y": "linear" } }
images/B08.png
truth/B08.truth.json
spec/B08.spec.json
images/B08.png
B09
scatter
Far outliers
outliers kept, not trimmed
2
1
124
{ "panel0": { "x": "linear", "y": "linear" } }
images/B09.png
truth/B09.truth.json
spec/B09.spec.json
images/B09.png
B10
scatter
Connected scatter
ordering is data
2
1
40
{ "panel0": { "x": "linear", "y": "linear" } }
images/B10.png
truth/B10.truth.json
spec/B10.spec.json
images/B10.png
B11
scatter
Scatter with marginals
panel decomposition
2
1
400
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" }, "panel2": { "x": "linear", "y": "linear" } }
images/B11.png
truth/B11.truth.json
spec/B11.spec.json
images/B11.png
C01
area
Hundred-per-cent stack
cumulative vs segment, stack order
2
5
1,300
{ "panel0": { "x": "linear", "y": "linear" } }
images/C01.png
truth/C01.truth.json
spec/C01.spec.json
images/C01.png
C02
area
Streamgraph with wiggle baseline
a computed baseline
2
4
1,200
{ "panel0": { "x": "linear", "y": "linear" } }
images/C02.png
truth/C02.truth.json
spec/C02.spec.json
images/C02.png
C03
area
Overlapping translucent areas
composited colours
2
3
900
{ "panel0": { "x": "linear", "y": "linear" } }
images/C03.png
truth/C03.truth.json
spec/C03.spec.json
images/C03.png
C04
area
Signed area across zero
the datum edge flips
2
1
400
{ "panel0": { "x": "linear", "y": "linear" } }
images/C04.png
truth/C04.truth.json
spec/C04.spec.json
images/C04.png
C05
area
Ridgeline with overlap
panels that must not stack
2
8
2,400
{ "panel0": { "x": "linear", "y": "linear" } }
images/C05.png
truth/C05.truth.json
spec/C05.spec.json
images/C05.png
C06
area
Band between two curves
both boundaries recorded
2
2
600
{ "panel0": { "x": "linear", "y": "linear" } }
images/C06.png
truth/C06.truth.json
spec/C06.spec.json
images/C06.png
C07
area
Shaded integral region
hatching is annotation
2
1
400
{ "panel0": { "x": "linear", "y": "linear" } }
images/C07.png
truth/C07.truth.json
spec/C07.spec.json
images/C07.png
C08
area
Sliver layer
a thin layer surviving
2
4
1,200
{ "panel0": { "x": "linear", "y": "linear" } }
images/C08.png
truth/C08.truth.json
spec/C08.spec.json
images/C08.png
C09
area
Area on a log axis
fill area ≠ magnitude
2
1
300
{ "panel0": { "x": "linear", "y": "log" } }
images/C09.png
truth/C09.truth.json
spec/C09.spec.json
images/C09.png
D01
compound
Pressure–volume indicator diagram
a non-functional closed curve
2
1
400
{ "panel0": { "x": "linear", "y": "linear" } }
images/D01.png
truth/D01.truth.json
spec/D01.spec.json
images/D01.png
D02
compound
Psychrometric chart
a non-Cartesian frame
2
5
1,000
{ "panel0": { "x": "linear", "y": "linear" } }
images/D02.png
truth/D02.truth.json
spec/D02.spec.json
images/D02.png
D03
compound
Pump curve against system curve
preserving an intersection
2
3
601
{ "panel0": { "x": "linear", "y": "linear" } }
images/D03.png
truth/D03.truth.json
spec/D03.spec.json
images/D03.png
D04
compound
Bode plot
linked panels, shared log frequency
2
2
800
{ "panel0": { "x": "log", "y": "linear" }, "panel1": { "x": "log", "y": "linear" } }
images/D04.png
truth/D04.truth.json
spec/D04.spec.json
images/D04.png
D05
compound
Nyquist plot
the variable lives on the curve
2
1
500
{ "panel0": { "x": "linear", "y": "linear" } }
images/D05.png
truth/D05.truth.json
spec/D05.spec.json
images/D05.png
D06
compound
Phase diagram
boundaries meeting at a triple point
2
4
401
{ "panel0": { "x": "linear", "y": "log" } }
images/D06.png
truth/D06.truth.json
spec/D06.spec.json
images/D06.png
D07
compound
Titration curve with derivative
one series derived from the other
2
3
801
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" } }
images/D07.png
truth/D07.truth.json
spec/D07.spec.json
images/D07.png
D08
compound
Stress–strain curve
construction lines that look like data
2
2
409
{ "panel0": { "x": "linear", "y": "linear" } }
images/D08.png
truth/D08.truth.json
spec/D08.spec.json
images/D08.png
D09
compound
Moody diagram
a converging family of curves
2
11
3,300
{ "panel0": { "x": "log", "y": "log" } }
images/D09.png
truth/D09.truth.json
spec/D09.spec.json
images/D09.png
D10
compound
Smith chart
a curvilinear coordinate system
2
1
60
{ "panel0": { "x": "linear", "y": "linear" } }
images/D10.png
truth/D10.truth.json
spec/D10.spec.json
images/D10.png
D11
compound
Bars with an overlaid rate line
three geometries, two axes
2
3
40
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" } }
images/D11.png
truth/D11.truth.json
spec/D11.spec.json
images/D11.png
D12
compound
Candlestick with volume
colour encodes direction, not identity
2
7
420
{ "panel0": { "x": "linear", "y": "linear" }, "panel1": { "x": "linear", "y": "linear" } }
images/D12.png
truth/D12.truth.json
spec/D12.spec.json
images/D12.png
M01
market
Three Funds Against the Benchmark
identity through a scheduled crossing: BLC x TGR at 30% and 73%, RSP punching up through both at ~50% and ~90%
3
3
756
{ "panel0": { "x": "linear", "y": "linear" } }
images/M01.png
truth/M01.truth.json
spec/M01.spec.json
images/M01.png
M02
market
Fund and Its Index Tracker
two series that are meant to be identical — reporting them as two, or saying plainly that they could not be separated
3
3
756
{ "panel0": { "x": "linear", "y": "linear" } }
images/M02.png
truth/M02.truth.json
spec/M02.spec.json
images/M02.png
M03
market
One Breakaway, Four in the Pack
a 4x breakaway sets the y range, compressing four crossing series into the bottom fifth of the plot
3
5
1,260
{ "panel0": { "x": "linear", "y": "linear" } }
images/M03.png
truth/M03.truth.json
spec/M03.spec.json
images/M03.png
M04
market
Sector Constituents, One Year
series count and identity at ten, with the crossings clustered at two shared drawdowns rather than spread out
3
10
2,520
{ "panel0": { "x": "linear", "y": "linear" } }
images/M04.png
truth/M04.truth.json
spec/M04.spec.json
images/M04.png
M05
market
Sector Rotation: Leadership Changes Hands
ordering — the answer is a sequence of rank changes, not five curves
3
5
1,260
{ "panel0": { "x": "linear", "y": "linear" } }
images/M05.png
truth/M05.truth.json
spec/M05.spec.json
images/M05.png
M06
market
Four Funds Hugging the Benchmark
accumulated identity error: no single crossing is hard, there are simply a great many between series that look alike by construction
3
5
1,260
{ "panel0": { "x": "linear", "y": "linear" } }
images/M06.png
truth/M06.truth.json
spec/M06.spec.json
images/M06.png
M07
market
Shared Crash, Different Recoveries
behaviour through a shared event — every series turns in the same session, so direction cannot disambiguate the crossings
3
4
1,008
{ "panel0": { "x": "linear", "y": "linear" } }
images/M07.png
truth/M07.truth.json
spec/M07.spec.json
images/M07.png
M08
market
High Beta Against Low Beta
crossings as events with a common cause: all three turn at every market turn, so every crossing happens where the cue is shared
3
3
756
{ "panel0": { "x": "linear", "y": "linear" } }
images/M08.png
truth/M08.truth.json
spec/M08.spec.json
images/M08.png

ChartParse-Bench

58 synthetic charts whose exact values were written before the pixels existed.

Every gate in chart extraction measures self-consistency: redraw the extraction, check it lands on the same ink, call it verified. That passes a series traced confidently off the wrong axis. Because each chart here ships the array it was drawn from, the question stops being is this answer plausible and becomes is this answer right.

Contents

family charts series what it isolates
A — line 18 62 identity through crossings, resampling honesty, axis binding
B — scatter 11 22 recall under overplotting, size/shape channels, outliers
C — area 9 29 stacking, computed baselines, composited fills
D — compound 12 43 non-Cartesian frames, linked panels, mixed geometries
M — market 8 38 correlated series that all turn in the same session
total 58 194 77,929 ground-truth points

1–12 series per chart. 8 panels on log axes (3 log-y, 3 log-x, 2 log-log).

Each chart isolates one named failure, declared in its spec before it is drawn — A01 "colour-based tracing that swaps identity at a crossing", A14 "gap honesty", C07 "hatching is annotation", D10 "a curvilinear coordinate system", M06 "accumulated identity error: no single crossing is hard, there are simply a great many".

Files

images/<ID>.png          the figure, as an extractor receives it
truth/<ID>.truth.json    the exact values used to draw it, plus axis_scales and
                         `printed` flags on labels that actually appear
spec/<ID>.spec.json      what the chart tests, its difficulty level, and for the
                         M set the measured crossing schedule
metadata.jsonl           one row per chart
index.json               the same rows as a single array

Suggested metric

score = series_recall x coverage x accuracy x mark_recall

Kept as a product, not an average: recovering one of eight series perfectly must not read as a good parse. Three properties matter more than the formula:

  • Pair series by shape, not by name (Hungarian assignment), so a system cannot win on labelling.
  • Error relative to each series' own range, and compared in log10 space on a log axis — otherwise a four-decade residual is dominated entirely by the top decade.
  • Compare nothing where the extractor declared a gap. A declared absence is neither credited as coverage nor bridged and charged as error. Saying "I could not see this" has to be allowed to be the right answer.

Use RMS rather than median: a median hid a log-axis magnitude error at 12% where RMS showed 22%.

Known limitations

  • C05 is drawn with no numeric y ticks, so its density values are not recoverable from the image at all. It tests shape only.
  • A14 and C03 store values across a span wider than the figure can show (tails decaying to ~4e-6 of the peak). A coverage measure taken against the stored array rather than against visible ink will understate a correct parse on those two.
  • D10 is a Smith chart stored in reflection coefficient. A system answering in normalised impedance is not wrong; the conversion is arithmetic an extractor should not be expected to perform.
  • Synthetic throughout. These charts are matplotlib output, so they carry none of the compression artefacts, rotated scans, or hand-drawn annotation of a real document. They measure numeric correctness, not robustness to bad inputs.
Downloads last month
22