ResearchPlot 2.0

A local, source-backed preflight workspace for research figures.

PyPI package · documentation website · releases

ResearchPlot checks the files that researchers actually submit. It resolves an immutable venue profile, plans which evidence is needed, inspects live Matplotlib figures and saved artifacts, and reports both violations and gaps it could not establish. The project model also carries captions, descriptions, source data, deliverables, manuscript metadata, and profile locks.

ResearchPlot is a compliance assistant, not an acceptance guarantee. Its sources, caveats, coverage gaps, and skipped checks remain visible in every report.

flowchart LR
    P["Schema-v3 project"] --> C["Compliance plan"]
    V["Locked, source-backed profile"] --> C
    C --> L["Live figure evidence"]
    C --> F["Saved-file evidence"]
    C --> B["Bundle metadata"]
    C --> M["Compiled-PDF structure"]
    L --> R["Coverage-aware report"]
    F --> R
    B --> R
    M --> R
    R --> O["HTML / JSON / SARIF / bundle"]

Install

ResearchPlot 2.0 requires Python 3.11 or newer. The distribution is named researchplot-venues; the import package and command are both researchplot.

python -m pip install researchplot-venues
researchplot --version

Optional capabilities are installed only when needed:

python -m pip install "researchplot-venues[web]"       # local browser workspace
python -m pip install "researchplot-venues[registry]"  # signed profile sync
python -m pip install "researchplot-venues[plots]"     # deprecated plotting helpers

The base package works offline and requires neither LaTeX nor downloaded fonts. Checking, exporting, bundling, and the browser workspace do not contact the network.

Audit an existing figure first

Use a pinned profile coordinate for repeatable work:

researchplot audit figures/figure1.pdf \
  --profile nature@2026.08.0 \
  --role main \
  --width single \
  --content line-art

The same operation is available in Python:

import researchplot as rp

target = rp.target(
    "nature@2026.08.0",
    role="main",
    width="single",
    content="line-art",
)
report = target.audit("figures/figure1.pdf")

print(report.verdict)
for finding in report.failures:
    print(finding.rule_id, finding.message)

A saved file cannot prove every property of the figure that produced it. Required live-figure or bundle evidence that is unavailable makes a coverage-aware project result INDETERMINATE; it is never silently counted as a pass.

Create a strict v2 project

researchplot.toml uses schema version 3. Unknown keys, duplicate IDs, unpinned profiles, invalid formats, unsafe/out-of-root paths, and empty projects are errors.

schema_version = 3
profile = "nature@2026.08.0"
policy = "complete"
lock = "researchplot.lock.json"

[[figures]]
id = "figure-1"
number = 1
role = "main"
width = "single"
content = "line-art"
caption = "Response increases across the four measured inputs."
alt_text = "Line chart with a monotonic increase from input zero to three."
source_data = ["data/figure1.csv"]

[[figures.deliverables]]
id = "main"
format = "pdf"
path = "figures/figure1.pdf"
required = true
preferred = true

Resolve and write the profile lock, then check in frozen mode:

researchplot profile lock nature@2026.08.0 --output researchplot.lock.json
researchplot check --config researchplot.toml --frozen

The Python workflow exposes the same plan and coverage model:

import researchplot as rp

project = rp.Project.load("researchplot.toml")
plan = project.plan(frozen=True)
report = plan.check()

if report.verdict is rp.Verdict.COMPLIANT:
    bundle = project.bundle("dist/submission")

Project.bundle() currently writes a verified submission directory. Build a deterministic ZIP from that directory with create_deterministic_archive() after its manifest verifies. JATS and RO-Crate metadata converters are available for the emitted submission manifest; they do not invent missing captions or descriptions.

Make a Matplotlib figure at the venue width

import matplotlib.pyplot as plt
import researchplot as rp

project = rp.Project.load("researchplot.toml")
figure = project.figure("figure-1")

with figure.style(deliverable="main") as style:
    fig, ax = style.subplots(aspect=0.62)
    ax.plot([0, 1, 2, 3], [0, 1, 4, 9], marker="o")
    ax.set(xlabel="Input", ylabel="Response")

    report = figure.check(fig=fig)
    result = figure.export(fig, policy="violations")

print(report.verdict, result.paths)
plt.close(fig)

The style uses matplotlib.rc_context; global rcParams are restored even after an exception. ResearchPlot does not replace Matplotlib, Seaborn, SciencePlots, TUEPlots, R, Julia, or design software. Any tool may create the saved artifact; only live styling and live-artist inspection are Matplotlib-specific.

Understand the verdict

Rule strength and check status are independent. Recommendations warn; inferred guidance is informational. A manual attestation can satisfy only a rule classified as manual. A waiver records a workflow decision but does not turn a venue violation into venue compliance.

CLI exit codes are stable: 0 compliant, 1 non-compliant, 2 invalid/unsafe input or an operational capability error, and 3 indeterminate.

What 2.0 can inspect

• PDF page boxes, physical size, font resources, embedded images, color spaces, transparency, annotations, JavaScript, actions, and embedded files.
• SVG dimensions, text/font declarations, embedded and external references, scripts, handlers, and foreignObject content.
• PNG, JPEG, and TIFF dimensions, EXIF orientation, DPI, color mode, ICC data, bit depth, compression, alpha, and frame count where exposed by the format.
• EPS format integrity and bounding boxes, with unsupported properties reported as unresolved.
• Deterministic grayscale and color-vision previews plus advisory contrast, entropy, transparency, rendered text/legend clipping, label-overlap, whitespace, final-size font, and colormap-luminance diagnostics.
• Compiled manuscript PDF structure and conservative figure placement through embedded provenance IDs, exact raster fingerprints, or unique configured hints. It measures bounds, rotation, scale/effective DPI, and crop-box clipping; vector fingerprinting, caption/reference reconciliation, and venue-rule integration remain unresolved.
• Submission manifests, path safety, hashes, deterministic ZIP archives, JATS figure metadata, and optional RO-Crate metadata.

Parsers are bounded and active content is reported, never executed. Isolated artifact inspection is available for hostile or untrusted inputs, but no parser should be treated as a perfect sandbox.

Profiles and provenance

Profiles are declarative, immutable JSON data with coordinates such as cvpr-2026@2026.08.0, SHA-256 digests, rule applicability, typed probes, official source locators, verification dates, interpretation notes, review status, and caveats. Missing official guidance stays unspecified.

researchplot profile list
researchplot profile search vision
researchplot profile show cvpr-2026@2026.08.0
researchplot explain figure.width.single --profile nature@2026.08.0

ResearchPlot 2.0 ships 22 bundled profiles spanning journals, publisher guidance, and 2026 CS/ML conferences. See the generated profile evidence catalog for every rule and source. Generic publisher profiles are not presented as journal-specific guarantees, and narrow profiles cover only the rules their official sources establish.

Signed registry updates are opt-in. Only an explicit profile-sync operation may use the network; normal resolution consults installed data and project locks prevent silent updates or rollback. Registry clients require an explicitly configured trusted root.

Local browser workspace

python -m pip install "researchplot-venues[web]"
researchplot serve

The workspace binds only to 127.0.0.1, uses a per-launch token and origin checks, and deletes uploaded temporary artifacts after inspection. It supports local drag-and-drop artifact audits, installed-profile discovery, JSON export, reproducible CLI commands, and raster grayscale/color-vision previews. It does not upload files, send telemetry, edit scientific data, or update profiles.

Compatibility and migration

The v1 Target, target(), reports, project reader, CLI aliases, and researchplot.plots bridge remain available throughout 2.x with deprecation warnings. Install [plots] only while migrating; new work should compose figures with native Matplotlib. Run researchplot migrate to translate a v1 configuration into a separate schema-v3 file and review every untranslatable field before adoption.

No compatibility surface is scheduled for removal before 3.0.

Documentation

• ResearchPlot 2.0 overview
• Getting started
• Compliance and coverage
• Configuration
• Artifact and bundle formats
• Security and limitations
• Python API

ResearchPlot is MIT licensed. See CONTRIBUTING.md before proposing a profile or behavior change, report vulnerabilities through SECURITY.md, and cite the project using CITATION.cff.