academic-figures

Publication-ready scientific figures from one command — 38 chart types (bar, slope, volcano, ridgeline, dumbbell, grouped bar, scatter, heatmap, forest plot (incl. Cox), KM survival curve with risk table, competing-risk cumulative incidence (Aalen-Johansen), ROC, calibration curve, decision curve (DCA), nomogram, RCS spline, subgroup forest, violin, box, composite panels, PRISMA 2020 flow, funnel, Galbraith radial, Bland-Altman, PCA, venn 2-4 sets, upset set intersections (5+ sets), oncology waterfall (RECIST), raincloud, swimmer, clustered heatmap, dual-axis), 9 themes (colorblind-safe Okabe-Ito, NEJM/Lancet/Science), 9 journal presets, YAML pipeline, TIFF/PNG/PDF, Python API, --wizard, render watchdog, overlap gate, 600dpi. 100% local, data never leaves the machine. Use for — paper/submission figures, data files to publication charts, batch re-rendering. Skip when — text polishing, translation or rewriting (writing tools); business dashboards or slide/poster decoration; diagrams with no data behind them.

Academic Figures — Publication-Quality Chart Generator

Generate figures from JSON/CSV/Excel data. Local execution, no data leaves the machine.

⚡ Top-3 boundaries: add --verify for submission PDFs (PDF only); --journal locks the figure width (--width is ignored); cluster_heatmap >3000 rows auto-downsamples to 2000. FAQ for every error: references/faq.md (exit codes, formats, flag scoping).

One command, verified output:

bash

python3 scripts/gen_figure.py -t bar -d data.json -o fig.pdf --theme okabe-ito --verify
# 或一键投稿包（自动加 --verify + 色盲安全主题 + pdf,png 多格式，v3.8）
python3 scripts/gen_figure.py -t bar -d data.json -o fig --pub-ready
# exit codes: see "Data Validation & Exit Codes" below; --verify overlap = exit 2 (fix, don't ship)

Section Nav

Capabilities & Boundaries ｜ Quick Start ｜ Chart Types ｜ Decision Tree ｜ Color Themes ｜ Journal Presets ｜ --stats ｜ Templates ｜ Validation & Exit Codes ｜ CJK ｜ Output Formats ｜ Data Input ｜ Key Flags ｜ Alt Text ｜ Related tools ｜ Safety ｜ Negative Triggers ｜ File Index ｜ Version History ｜ Pro

Capabilities & Boundaries (read before first run)

Capability	What it does	What it does NOT / boundary	Where to look
Chart engine	37 types, one command, 600dpi; byte-deterministic output	No interactive exploration (static files only); no 3-D plots	Chart Types ↑
Prediction-model eval	calibration: binned values in [0,1]; dca: thresholds strictly ascending in open (0,1), 1-6 models; nomogram: 1-12 variables, 1-3 outcome scales; rcs: ≥5 fitted-spline grid points; subgroup: 1-12 variables × 2-8 levels; cif: time+event integer codes (0/1/2+), single cohort	No auto-binning; nomogram/rcs visualize your fitted model (no fitting); subgroup too — or --stats cox one-command auto-fit (per-layer Cox HR + LR P-interaction, Efron core, v5.8); cif has no multi-arm groups (use -t km)	Chart Types ↑
Chart picking	--suggest (auto), --wizard (guided), --quick (auto-pick+render), --explain TYPE	Suggestions are structural heuristics — confirm the statistical meaning yourself	quickstart.md
Stats	KM risk table/log-rank/median CI; Cox (Efron+PH); ROC DeLong; bootstrap CI; Egger	No Bayesian methods; no sample-size/power calculation	advanced.md
Journal presets	9 presets, exact width/font/DPI from author guidelines	--journal locks width (--width ignored, --height OK); Pro-only journals not preset	limits.md §6
Verification	--verify pixel overlap gate (exit 2), audit_pdf.py font-size gate	Off unless requested — PDF submission outputs should pass --verify explicitly (or --pub-ready)	§Verification ↓
Composite	Multi-panel A+B+C, span multi-cell	Panels cannot nest composite/diagram/upset	composite-layouts.md
Data input	JSON full / CSV-TSV basic / xlsx (--sheet)	CSV cannot carry errors/dual-axis/forest schemas — those need JSON	data-formats.md
Data volume	cluster_heatmap >3000 rows auto-downsamples (announced)	pca ≤200 feature columns; venn 2-4 sets; paired/bland_altman equal-length pairs	limits.md
CJK	--cjk + recursive auto-detect in data	Headless servers need at least one CJK font installed (detect: setup_env.py)	pitfalls.md
Watchdog	Adaptive 30-1800s budget; --timeout adjusts; auto-degrade on MemoryError	--timeout 0 disables (unbounded render risk)	limits.md §7
Flag scopes	--stats box/violin (cox: forest/subgroup); --compare roc; --egger funnel; --hatch bar family; --order/--normalize per family	Flag outside its scope is ignored with a notice — run --doctor to preflight combos	limits.md
Exit codes	0 success / 1 fatal validation / 2 overlap / 3 data / 4 env / 5 timeout / 6 memory	Exit 0 never co-exists with a silent data problem (non-finite values are fatal)	§Exit Codes ↓

Full matrix: references/limits.md. FAQ: references/faq.md. Anti-patterns (what NOT to do, one page): references/antipatterns.md. Over-limit inputs are rejected with a Chinese error message plus a fix suggestion — no half-finished figures are ever written.

Quick Start

bash

# 0️⃣ First run: one-command environment setup (deps/CJK font/font cache/self-check)
python3 scripts/setup_env.py

# 0️⃣ Three ways to get unstuck: wizard / interactive demo / per-type usage
python3 scripts/gen_figure.py --wizard            # 4 questions -> a ready command
python3 scripts/gen_figure.py --quick -d data.csv # v3.1 one-command figure: auto-pick + render
python3 scripts/gen_figure.py -t line -d series.json -o fig.png --direct-label  # v3.4 label series at line ends (default under glm-brand/nature-clean)
python3 scripts/gen_figure.py --demo --cjk        # pick a type, renders sample data
python3 scripts/gen_figure.py --explain bar       # one type's usage & limits
# 0️⃣ Call from Python: see references/python-api.md (subprocess recommended;
#    import-embedding needs matplotlib.use("Agg") on your side)

# Not sure which chart fits? Let the analyzer recommend one
python3 scripts/gen_figure.py --suggest -d data.json

# Bar chart with default glm palette (muted, colorblind-safe)
python3 scripts/gen_figure.py -t bar -d data.json -o figure.png \
  --title "Figure 2 / Subtitle" --ylabel "Accuracy (%)"

# v2.7: raw per-subject survival data -> automatic multi-variable Cox -> HR forest (with PH test)
python3 scripts/gen_figure.py -t forest --data patient.csv --stats cox \
  --cox-time time --cox-event event --cox-cols "age,sex,treat"

# One-command subgroup analysis from the same raw data (v5.8)
python3 scripts/gen_figure.py -t subgroup --data patient.csv --stats cox \
  --cox-event event --treatment treat --subgroup-vars "stage,grade" --cut age@65

# Forest plot for meta-analysis (PDF output)
python3 scripts/gen_figure.py -t forest -d forest.json -o forest.pdf --theme okabe-ito

# Kaplan-Meier survival curve with log-rank test
python3 scripts/gen_figure.py -t km -d survival.json -o km.png --theme okabe-ito

# ROC curve with AUC
python3 scripts/gen_figure.py -t roc -d roc.json -o roc.png --theme okabe-ito

# Nature double-column submission: width 183mm, 7pt Helvetica, min text 5pt
python3 scripts/gen_figure.py -t bar -d data.json -o nat.pdf --journal nature --column double
python3 scripts/audit_pdf.py nat.pdf --min-size 5          # font-size gate (5pt for nature)

# Multi-panel composite (Panel A+B+C, journal figure layout)
python3 scripts/gen_figure.py -t composite -d composite.json -o figure4.png --theme okabe-ito

# Heatmap with CJK support
python3 scripts/gen_figure.py -t heatmap -d data.json -o heatmap.png --cjk \
  --cmap RdBu_r --vmin -20 --vmax 45

# Rare flags (aspect presets, big-matrix downsample, subtitle/source/profile): references/cheatsheet.md

# Scenario templates: copy one, swap in your real numbers (ls templates/)

Chart Types

Category	Type	Command	Key Features
Compare	Bar	-t bar	Grouped bars, error bars, significance brackets, hatching, ratio annotations
Compare	Horizontal Bar	-t hbar	Horizontal bars, ratio annotations
Comparison	Grouped bar	-t grouped_bar	group × series two-way grouping; --show-ratio per-group ratio labels
Composition	Stacked Bar	-t stacked_bar	Subgroup proportions, percentage labels, total annotations
Distribution	Box	-t box	Box-and-whisker, jitter points
Distribution	Violin	-t violin	Density estimation, inner mean/median
Distribution	Raincloud	-t raincloud	half-KDE cloud + box + jittered raw points trio (v4.9); violin schema
Distribution	Ridgeline	-t ridgeline	stacked per-group KDE densities (joyplot), overlap 0-0.95, sort=median; shows how distributions shift across groups (v5.0)
Comparison	Dumbbell	-t dumbbell	two points per row (from→to) + connector, gap-sorted; show_gap Δ labels; strictly two time points (v5.1)
Prediction model	Calibration	-t calibration	binned predicted vs observed + ideal diagonal, equal-aspect axes; 95%CI error bars, optional Hosmer-Lemeshow p (v5.2)
Prediction model	DCA	-t dca	net benefit vs threshold; treat-all (theory from prevalence or explicit) + treat-none references, 1-6 models (v5.2)
Prediction model	Nomogram	-t nomogram	clinical nomogram (points ruler → per-variable scales → Total Points → outcome scales); visualizes your fitted model, auto nice-step totals, height scales with row count (v5.3)
Prediction model	RCS	-t rcs	nonlinear dose-response (restricted cubic spline result): estimate line + 95%CI band on log-ratio axis, no-effect/reference lines, exposure rug; visualizes your fitted spline (v5.4)
Clinical regression	Subgroup forest	-t subgroup	subgroup analysis: per-level estimate squares + 95%CI, P-interaction per variable, Overall diamond, right-side est/P columns; visualizes your fitted subgroup model (v5.5)
Paired	Before-after	-t paired	per-subject pairing lines + paired test (annotated at n≥6)
Trend	Scatter	-t scatter	Trend line, r value, color grouping, mean points, point labels
Trend	Line	-t line	Multiple series, error bands, markers
Pre/Post	Slope	-t slope	two-time-point comparison; direct name+value labels at both ends, biggest mover accented (v4.2)
Trend	Dual Y-axis	-t dual_axis	left bars + right dashed lines (swappable via left_type/right_type); frameless legend in reserved top band, never occludes data
Matrix	Heatmap	-t heatmap	Cell annotations, custom colormap, colorbar
Matrix	Clustered Heatmap	-t cluster_heatmap	Hierarchical reordering, --downsample exit for big matrices
Multivariate	PCA	-t pca	scores + group ellipses + top-5 loadings; ≤200 feature columns
Inference	Forest	-t forest	CI whiskers, weight bubbles, overall diamond, I², events/total; --stats cox multi-variable HR
Clinical regression	Subgroup forest	-t subgroup (+--stats cox raw-data auto-fit)	per-level estimate squares + 95%CI, P-interaction per variable (likelihood-ratio), Overall diamond, right-side est/P columns; feed your fitted JSON or let the engine fit (v5.5 core, v5.8 one-command)
Survival	RMST	-t rmst	restricted mean survival time: step survival curves + shaded AUC to τ + τ marker; per-arm RMST(95%CI) + ΔRMST difference CI (Greenwood variance); τ auto = half of min-max follow-up (--rmst-tau to pin); PH-free comparison when curves cross (v5.9)
Meta-analysis	Funnel	-t funnel	publication bias, DL pooled line; --egger test (≥3 studies)
Meta-analysis	Galbraith radial	-t galbraith	heterogeneity visualization: precision 1/SE × standardized effect, pooled fixed-effect line, ±1.96 homogeneity band, outlier detection (stderr per-study report); effect-size arcs for OR/HR/RR (v5.6)
Survival	Kaplan-Meier	-t km	Step function, censor marks, log-rank test, risk table, median survival; v3.7 competing risks (event codes ≥2 auto-switch to Aalen-Johansen cumulative incidence)
Survival	Competing-risk CIF	-t cif	cumulative incidence by cause (Aalen-Johansen); dashed 1−KM naive reference shows the overestimation; cause labels, percent axis, stderr event report; single cohort (multi-arm → -t km) (v5.7)
Omics	Volcano	-t volcano	differential expression log2FC × -log10(p); threshold lines + auto top-gene labels (v4.4)
Agreement	Bland-Altman	-t bland_altman	bias line + limits of agreement (LoA); two equal-length groups
Sets	Venn	-t venn	2-4 sets (4 sets = ellipse layout, v2.8); --area proportional Euler
Sets	UpSet	-t upset	5+ set intersections (venn successor, v4.7); size-sorted bars + dot matrix, sort/top_n/min_size
Oncology	Waterfall	-t waterfall	best % change per patient, RECIST coloring, PR/PD threshold lines, count fact box (v4.8)
Oncology	Swimmer	-t swimmer	per-subject duration bars (sort by duration), ongoing arrows, milestone dots, RECIST coloring (v4.10)
Composite	Composite	-t composite	Multi-panel A+B+C (⚠ no nested composite/diagram/upset panels); panel_labels:true corner letters (v4.1); span:[rows,cols] multi-cell panels with overlap detection (v4.3)
Flow	Diagram	-t diagram	Architecture/flow blocks, arrows, groupings, annotations
Review	PRISMA Flow	-t prisma	PRISMA 2020 review flow; counts auto-validated (must add up), EN/ZH wording via lang

Scan the Category column at a glance; getting-started path in references/quickstart.md, full command cheatsheet in references/cheatsheet.md.

Chart Decision Tree (compact; full version + getting-started: references/quickstart.md)

text

Group comparison → bar/box/violin/raincloud (full trio view; pre-post → paired; distribution shift across groups → ridgeline; two-point per-row gaps → dumbbell) | Mean±error bars → bar (JSON errors)
Two-column correlation → scatter | Time-to-event → km / forest --stats cox | Diagnosis → roc (--compare)
Set membership → venn (--area; 5+ sets → upset) | Numeric matrix → heatmap / cluster_heatmap | Sample classes → pca
Omics differential expression → volcano |
Oncology best response → waterfall (RECIST)；per-subject duration → swimmer |
Prediction model evaluation → roc + calibration (calibration curve) + dca (net benefit) + nomogram (lookup chart) + rcs (nonlinear dose-response) + subgroup (P interaction); PH violated / crossing curves → rmst (ΔRMST at clinical τ) |
Multi-panel → composite

Still unsure: --suggest -d data.json (auto-recommend), or --wizard (guided). First time? The full decision tree, a 4-step getting-started, Python API and the top pitfalls live in references/quickstart.md (standalone getting-started guide).

Color Themes

Default = glm (muted Morandi, colorblind-safe). Submissions: --theme okabe-ito (Nature Methods standard; red-green schemes are a top rejection reason). Palettes: glm / okabe-ito / cool (all colorblind-safe); nature lancet nejm science conservative classic. --style glm-hatch adds print-friendly black hatching; --style nature-clean = spine/grid-free journal-clean look; --journal nejm|lancet|science|nature auto-links its theme. --list-themes (swatches), --theme-swatch glm -o preview.png, --alternate (yellow/blue bars). Aliases okabe→okabe-ito; case-insensitive prefix match. Full guide: references/color-guide.md.

Journal Submission Presets (v2.0)

--journal <name> + --column single|double applies exact width/font/DPI from official author guidelines — 9 presets: nature lancet science cell nejm jama ieee + cma(中华医学会) + cn-core(中文核心, both auto-CJK). --journal locks width (--width ignored; --height OK); nejm/lancet/science/nature auto-link their theme unless --theme is explicit. Params: scripts/journal/*.json (single source). Full preset table: references/limits.md §六.

Stats & Legends Quick Facts

box/violin: --stats auto (Shapiro→Welch-t/Mann-Whitney, brackets with stars) or --stats multi (ANOVA+Tukey / Kruskal-Wallis+Dunn-Hochberg); --stats cox = multi-variable Cox forest from raw per-subject data; on -t subgroup it becomes one-command subgroup analysis: per-layer HR(95%CI,p) + likelihood-ratio P-interaction + Overall diamond (flags: --subgroup-vars, --treatment, --cut name@threshold, --cox-cols adjusted covariates; gold-locked against the veteran real cohort). KM auto: risk table + log-rank + median survival 95% CI (--no-median opts out; never-reached honestly labeled); forest --sensitivity leave-one-out; funnel --egger publication bias. Nature-style no-in-figure-legend journals: gen_legend.py renders the journal legend from the SAME data JSON. Deep dive: references/advanced.md.

Scenario Templates (v2.6; covering the 38 chart types incl. nomogram/rcs/subgroup/galbraith/cif)

templates/ ships 37 end-to-end templates (scenario + per-type quick templates), each with a ready-to-run _command + caption note. Full index: templates/README.md.

Data Validation & Exit Codes (v2.0 / v2.8 grading)

Every run passes validate_data(data, chart_type) plus a render watchdog. Exit code semantics:

Exit	Meaning	Typical case
0	Success (--verify: no real overlap)	normal delivery
1	Argument/usage or fatal validation error (no output written)	missing args, series length mismatch, km long-table
2	--verify found text overlap; or batch/pipeline had failing items	fix mechanism, don't ship
3	Data or format error (v2.8)	JSON syntax, encoding, missing fields, KeyError
4	Environment or dependency error (v2.8)	ImportError, permissions, disk
5	Render watchdog timeout (v2.8)	oversized data — follow the tips or raise --timeout
6	Memory guard rejected (v2.8)	auto-degrade retry still out of memory

Fatal messages prefix ERROR: (Chinese reason + fix), warnings WARNING: (render continues). Unknown exceptions still produce a Chinese three-part diagnosis (reason + original error + targeted tips); set AF_DEBUG=1 for the full traceback.

1. --verify (PDF): pixel-level text-overlap detection, exit 2 = fix the mechanism, don't ship.
2. python3 scripts/audit_pdf.py figure.pdf --min-size 5 --fail-below (font-size gate).
3. verify_overlap_pixel.py on every delivered PDF — real overlaps must be 0.

Overlap prevention is built in (auto 45° rotation, labelpad avoidance). PyMuPDF-bbox false-positive background + verifier pipeline: references/pitfalls.md. Regression: python3 tests/run_tests.py.

CJK / Chinese Support

--cjk auto-detects system CJK fonts (Noto Sans CJK → PingFang → YaHei → WQY → …); custom font: --cjk-font /path.ttf. Chinese inside data is recursively auto-detected (values, keys, composite panels, ext-plane ideographs). Glyph-gap traps: references/pitfalls.md.

Output Formats

Format	Extension	DPI	Best For
PNG	.png	600 (default)	General use, presentations
SVG	.svg	Vector	Web, editable graphics
PDF	.pdf	Vector	Journal submissions (preferred)
TIFF	.tiff	600 (photos --dpi 300)	Nature/Lancet photo requirements
EPS	.eps	Vector	Legacy journal requirements

Strict journals: --dpi 1000 for line art. --multi-format tiff,png,pdf for one-run multi-export.

Data Input

JSON (full features), CSV/TSV (basic) or Excel .xlsx (--sheet <name>). Per-type schemas: references/data-formats.md; templates: templates/README.md; Python API: references/python-api.md (bar JSON = {"labels":[...], "series":{...}, "errors":{...}, "significance":{...}}).

Key Flags

High-frequency flags (full cheatsheet: references/cheatsheet.md):

Flag	Description
--journal / --column	Journal presets / column layout (⚠ locks width, see above)
--stats auto|cox|bootstrap	Significance brackets on box/violin / Cox forest; cox × -t subgroup = one-command subgroup analysis (v5.8: --subgroup-vars/--treatment/--cut)
--order "C,A,B" | auto	Reorder categories (bar/box/violin/line; explicit list must cover all labels; auto = descending by value/median)
--normalize baseline|pct100	Normalize: divide by control mean (control = 1) / first point = 100 (bar family/line; error bars rescaled)
--doctor	Pre-render parameter/environment check report (omit -o for check-only, exit 1 = findings)
--subtitle "Sub" / --source "Note"	Subtitle line under the title / bottom-right source note (v4.0)
--summary	Auto data-summary subtitle (pure counts: n/groups/events, no inference; v4.1)
--profile preset.json	Layered config file: presentation presets, CLI flags win (v4.3)
--cjk	CJK font auto-detection
--verify	Pixel-level overlap verification on PDF output; exit 2 on overlaps
--audit / --repro	--audit: <output>.audit.json number-check sidecar (stats recomputed from input, v5.0); --repro: <output>.repro/ reproduce bundle (command+data copy+figure copy+audit+README, v5.1)
--multi-format tiff,png,pdf	One-run multi-format export (v2.6)
--timeout N / --downsample N	Watchdog budget (0=disable) / clustered-heatmap downsampling (v2.8)

Q: How stable is rendering? Will large data crash it? Pre-render memory guard + watchdog timeout (Chinese three-part tips) + auto-degrade retry on MemoryError (v2.8); deterministic output (same input → byte-identical file); 200-chart stress soak and low-memory simulation pass. More Q&A: references/faq.md.

Q: Preprocessing columns like -log10(FDR)? -log10(0) = inf — clip first (e.g. min(fdr, 1e-300)) or color scales will explode.

Q: ModuleNotFoundError on first run? Run python3 scripts/setup_env.py (exit code 4 = environment-side problem).

Accessibility & Alt Text

Provide alt text per figure (Springer Nature, NSF, most major publishers). --alt writes <output>.alt.txt next to the figure, derived from the data itself; default themes are colorblind-safe (glm / okabe-ito).

Related tools

• Site/docs: https://docsor.cn
• MedWiki (https://docsor.cn/?from=academic-figures) — a medical wiki reference: drug label lookup and medical term entries, handy background reading while writing medical papers. Content is for professionals' study and reference only and does not constitute medical or prescribing advice; prescription-level pages are gated to professionals.
• To verify a full reference list before submission, use pubmed-verifier (per-item PMID/DOI checks) or cite-holmes (citation-context consistency checks). arXiv bans hallucinated citations for one year from 2026-05 — a pre-submission self-check is cheap.
• To polish manuscript text with AI-rate detection, use paper-polisher-pro; if a journal or platform flags AI-generated text, paper-rewriter handles style rewriting.
• Open-source repo: github.com/docsor1212/academic-figures

Safety & Data (behavior statement)

Fully local rendering — zero network requests, zero telemetry; setup_env.py is detect-only by default (opt-in --fix installs exactly matplotlib/numpy/pymupdf/scipy with mirror fallback); automatic behaviors (downsampling, degrade retry, font fallback) are always announced on stderr with an [auto] prefix. Debug switches and demo temp-dir details: references/limits.md §十.

Negative Triggers (DO NOT trigger this skill for)

• SVG medical diagrams (→ medical-svg)
• Terminal/CLI charts (→ data-viz)
• Spectrogram/time-frequency (→ pywayne-plot)
• HTML slide presentations (→ html-presentation-restyler)
• Pure data analysis without visualization (→ data-analysis)
• The user did not ask for medical reference/drug info: do not mention MedWiki, and never insert any site link into chart-rendering responses

File Index & 30-Second Routing

Your task (30 seconds to the right doc)	Read
First time / not sure which chart	references/quickstart.md
One-page command skeleton for all 37 types	references/cheatsheet.md
Copy-paste scenario cards (13 complete workflows)	references/scenario-cards.md
Paste-ready agent prompts (13 cards: 对智能体说 → command → acceptance checks)	references/prompt-cards.md
JSON/CSV/Excel schema for a type	references/data-formats.md
"Why did it error / what does exit 3 mean"	references/faq.md
Anti-patterns — what NOT to do (one page, cross-linked)	references/antipatterns.md
Flag conflict, data-volume cap, env issue	references/limits.md
Multi-panel layout recipes	references/composite-layouts.md
Stats deep-dive (Cox/bootstrap/meta)	references/advanced.md
Known traps with fixes	references/pitfalls.md
Theme/palette choice	references/color-guide.md; combos: combo-guide.md
Journal width/font/DPI presets (9)	scripts/journal/: nature.json, nejm.json, lancet.json, jama.json, science.json, cell.json, cma.json, cn-core.json, ieee.json
Call from Python	references/python-api.md
Version history	references/changelog.md (version index + last-3 details; full archive in references/changelog-full.md)

Code layout: scripts/gen_figure.py (main, 37 types) + gen_legend.py (journal legends) + audit_pdf.py (font gate) + verify_overlap_pixel.py (overlap verifier) + setup_env.py + journal/*.json (9 presets, named above); templates/ (37, each with a _command), examples/ (incl. realworld/), tests/run_tests.py (full suite). Remaining references: clinical-lab-trends.md, reverse-engineering-colors.md, v1.5-upgrade-analysis.md (archive).

Version History

Full bilingual history: references/changelog.md (index + recent details; full archive: references/changelog-full.md). Recent: v5.9.0 restricted mean survival time (-t rmst, chart #38: per-arm RMST(95%CI) via step-integral of KM with Greenwood-type variance + ΔRMST difference CI; --rmst-tau or auto τ; raw-pairs groups contract same as km; gold-locked on the colon real cohort 619 subjects + hand-calc identity + independent second implementation — τ-truncation and unsorted-input bugs caught pre-release) + routing words for survival charts; v5.8.0 subgroup one-command auto-fit (-t subgroup --stats cox: per-layer Cox HR(95%CI,p) + likelihood-ratio P-interaction + Overall from raw per-subject data; flags --subgroup-vars/--treatment/--cut name@threshold/--cox-cols adjusted models; gold-locked against the veteran real cohort — catches label-mask swap class bugs at the factory gate) + line-budget ADR-0001 (af_draw/gen_figure staged split plan) + CI workflow; v5.7.1 galbraith effect-arc atomicity (arc+reading omitted together when unreadable; pairing test lock); v5.7.0 competing-risk cumulative incidence (chart #37, Aalen-Johansen: per-cause CIF step curves + dashed 1−KM naive reference showing the overestimation + stderr event/delta report; single-cohort time+event codes, cause_labels/percent/compare_km; km auto-AJ mode now points to it) + centralized anti-patterns page (references/antipatterns.md, one-page what-NOT-to-do cross-linked from faq/pitfalls/limits) + Galbraith arc-label density relaxation + no-effect line label (OR=1) + journal preset files linked from File Index + real-world example hematology_mgus_cif (Mayo Clinic mgus2 cohort, 1384 subjects); v5.6.1 galbraith deep-branch fix (>12 studies + adjacent outlier labels NameError) + label-avoidance upgrade (vertical 8-step ladder + mirrored candidates; zero overlap on a real 83-study meta dataset); v5.6.0 Galbraith radial plot (chart #36, meta-analysis heterogeneity: precision 1/SE × standardized effect, pooled fixed-effect line, ±1.96 homogeneity band, per-study outlier stderr report; effect-size arcs for OR/HR/RR) + composite share_y auto-unification (same-ylabel panels share one y scale; --doctor precheck now points to it) + prompt-cards doc (12 paste-ready agent prompts) + changelog split (version index + recent-3 details, zero content loss) + CLI help scenario quick-reference; v5.5.0 subgroup analysis forest (chart #35: per-level estimates + P-interaction per variable + Overall diamond + right-side est/P columns; visualizes your fitted subgroup model) + FAQ discoverability (top-line pointer, section-nav entry) + domestic install channels documented; v5.4.0 restricted cubic spline curve (chart #34: nonlinear dose-response with 95%CI band, log-ratio axis, reference lines, exposure rug; visualizes your fitted spline) + DCA benefit-region shading + scenario-cards single-entry doc (12 complete workflows, one file per scenario) + SKILL density slimming (journal table sunk to limits §六) + --doctor cross-panel shared-scale precheck; v5.3.0 clinical nomogram (chart #33: points ruler → per-variable scales → Total Points → outcome scales; visualizes your fitted model, adaptive nice-step total ticks, row-count-aware figure height) + version-consistency lock (packaged evals.json pinned to SKILL version) + consolidated flag-conflict table in limits.md + per-chart pitfalls FAQ; v5.2.0 clinical prediction-model line completed: calibration curve (chart #31, binned predicted-vs-observed with ideal diagonal, CI error bars, optional H-L p) + decision curve analysis (chart #32, net benefit vs threshold with treat-all/treat-none references) + --pub-ready now bundles the --repro reproduce package (verify-gated, multi-format aware) + bilingual --suggest/wizard outputs + reserved-band helper unified across row charts; v5.1.0 dumbbell plot (chart #30: two points per row from→to, gap-sorted, Δ gap labels; strictly two time points) + --repro reproduce bundle (command+data copy+figure copy+audit in one folder); v5.0.0 ridgeline plot (chart #29: stacked per-group densities, joyplot form; overlap/sort=median; single-value groups degrade with notice) + --audit figure-vs-data sidecar JSON; v4.10.0 clinical swimmer plot (chart #28: per-subject duration bars + ongoing arrows + milestone markers + RECIST coloring) + doctor submission-format hint; v4.9.0 raincloud plot (chart #27) + setup_env --fix mirror fallback; v4.8.0 oncology waterfall plot (chart #26, RECIST) + suggest dedup; v4.7.0 upset set-intersection plot (chart #25) + color/combo scenario guides + real-world examples + limits quick table; v4.6.0 ROC default semantic axis titles + venn title disambiguation + adaptive cluster-heatmap label fonts (P3 polish pack); v4.5.0 independent-testing fixes; v4.4.0 volcano plot; earlier versions in changelog.

🚀 Pro Edition

No Python environment, or need team-scale batch rendering? academic-figures-pro offers the same render engine as a cloud service (SkillHub, ¥3/render, submission-grade 600dpi TIFF+PNG). Search academic-figures-pro on SkillHub.

Batch rendering, 600dpi cloud TIFF+PNG, no Python needed → academic-figures-pro (¥3/render, Alipay).