MoMA's collection

A tour of the Museum of Modern Art's open collection data (CC0) with Polars, Observable Plot and the jupyter display API. It fetches MoMA's CSV files from GitHub and imports packages from npm, so the first run downloads both. See Example notebooks for the others.

A notebook run on the cdp-node kernel, shown with its saved output.Download moma-collection.node.ipynb · View on GitHub

Every museum is a record of buying decisions. The Museum of Modern Art publishes its entire cataloged collection — 160k+ works and their artists — as open data, including the date each work was acquired. That turns the museum itself into a dataset: what was bought when, how long after it was made, from whom, and in what.

This notebook reads like a visit: we load and clean the data with Polars, chart it with Observable Plot, build a small interactive widget on cdp-kernel's comm protocol, and end with a one-work-per-decade tour rendered as museum wall labels.

It runs on cdp-kernel with a Node target: the kernel host speaks the Jupyter wire protocol while every cell executes over CDP in a separate Node process — TypeScript and TSX are transformed on the way in.

Sources and inspiration

The analysis approach here is inspired by Trevor Manz's post Exploring Art with TypeScript, Jupyter, Polars, and Observable Plot (notebook source), which explores the National Gallery of Art's open data under Deno's Jupyter kernel. The structure of that piece — tidy data, static plots, an anywidget-style table, a custom SSR component — is a lovely template for what a notebook-driven runtime should be able to do, so this notebook deliberately exercises the same breadth.

All code and prose here are original, the dataset is different, and the questions asked are different (the inspiration focuses on when art was made; we mostly ask when the museum acquired it). No code or text is copied from the original.

Data — MuseumofModernArt/collection on GitHub, released into the public domain (CC0). Per that repository, images are not part of the dataset, and much of the metadata is "as is" research data, not curator-approved. Links in this notebook go to the museum's own collection pages.

Specifiers below pin what needs pinning (react and react-dom ranges must agree — React enforces version parity between the pair).

Load

import * as pl from "npm:nodejs-polars@0.18.0";

const baseUrl = "https://github.com/MuseumofModernArt/collection/raw/refs/heads/main/";

// MoMA ships CSVs with a UTF-8 BOM; strip it so the first column name is clean.
const stripBom = (s: string) => s.replace(/^\uFEFF/, "");

let artists: pl.DataFrame = pl.readCSV(
  stripBom(await (await fetch(baseUrl + "Artists.csv")).text()),
  { quoteChar: '"' },
)
  .select(
    pl.col("ConstituentID").alias("artistId"),
    pl.col("DisplayName").alias("artist"),
    pl.col("Nationality").alias("nationality"),
    pl.col("Gender").str.toLowerCase().alias("gender"),
  );

artists.head();

artistId	artist	nationality	gender
1	Robert Arneson	American	male
2	Doroteo Arnaiz	Spanish	male
3	Bill Arnold	American	male
4	Charles Arnoldi	American	male
5	Per Arnoldi	Danish	male

let works: pl.DataFrame = pl.readCSV(
  stripBom(await (await fetch(baseUrl + "Artworks.csv")).text()),
  { quoteChar: '"' },
)
  .select(
    pl.col("ObjectID"),
    pl.col("Title"),
    pl.col("Artist"),
    pl.col("ConstituentID"),
    pl.col("Date").alias("dateText"),
    pl.col("Medium"),
    pl.col("Classification"),
    pl.col("Department"),
    pl.col("DateAcquired"),
    pl.col("CreditLine"),
    pl.col("AccessionNumber"),
    pl.col("URL"),
    pl.col("Height (cm)").alias("heightCm"),
    pl.col("Width (cm)").alias("widthCm"),
    pl.col("Duration (sec.)").alias("durationSec"),
  );

`${works.shape.height.toLocaleString()} works, ${artists.shape.height.toLocaleString()} artists`

'160,682 works, 15,934 artists'

Clean

Two messy columns do most of the damage. DateAcquired is a clean ISO date when present; the work's Date is free text (c. 1910, 1935-42, Published 1929…), so we take the first plausible four-digit year as an approximate work year.

works = works.withColumns(
  pl.col("DateAcquired").str.slice(0, 4).cast(pl.Int32, false).alias("acquiredYear"),
  pl.col("dateText").str.extract("(1[5-9][0-9][0-9]|20[0-9][0-9])", 1).cast(pl.Int32, false).alias("workYear"),
);

const knownAcquired = works.filter(pl.col("acquiredYear").isNotNull()).shape.height;
const knownWorkYear = works.filter(pl.col("workYear").isNotNull()).shape.height;
`acquisition year known for ${knownAcquired.toLocaleString()} / ${works.shape.height.toLocaleString()} works; approximate work year for ${knownWorkYear.toLocaleString()}`

'acquisition year known for 155,220 / 160,682 works; approximate work year for 156,257'

// Collaborations: some works list several constituent IDs. The first one is the
// primary artist (the Artist string lists names in the same order).
const collaborations = works.filter(pl.col("ConstituentID").str.contains(",")).shape.height;

works = works.withColumns(
  pl.col("ConstituentID").str.extract("([0-9]+)", 1).cast(pl.Int64, false).alias("primaryArtistId"),
  pl.col("acquiredYear").sub(pl.col("acquiredYear").rem(10)).alias("acquiredDecade"),
);

works = works.join(artists, {
  leftOn: "primaryArtistId",
  rightOn: "artistId",
  how: "left",
});

`join done — ${collaborations.toLocaleString()} works credit multiple artists; primary artist joined for `
  + `${works.filter(pl.col("artist").isNotNull()).shape.height.toLocaleString()} works`

'join done — 8,111 works credit multiple artists; primary artist joined for 159,425 works'

When the museum buys

import * as Plot from "npm:@observablehq/plot";
import { parseHTML } from "npm:linkedom@0.18.9";

// Observable Plot wants a DOM; on a Node target we hand it linkedom
// (the same trick most SSR renderers use).
const document = parseHTML("<html></html>").document;

const acquisitions = works
  .filter(pl.col("acquiredYear").isNotNull())
  .withColumns(pl.col("Department").fillNull("Unspecified"))
  .toRecords();

Plot.plot({
  width: 940,
  height: 300,
  y: { grid: true, label: "works acquired" },
  x: { type: "time", domain: [new Date(1925, 0, 1), new Date(2027, 0, 1)] },
  marks: [
    Plot.rectY(
      acquisitions,
      Plot.binX(
        { y: "count", filter: null },
        {
          x: (d) => new Date(d.acquiredYear, 0, 1),
          thresholds: 102,
          fill: "Department",
          fillOpacity: 0.9,
        },
      ),
    ),
  ],
  color: { legend: true },
  document,
});

Architecture & DesignDrawings & PrintsFilmFluxus CollectionMedia and PerformancePainting & SculpturePhotography

const busiest = works
  .filter(pl.col("acquiredYear").isNotNull())
  .groupBy("acquiredYear").len()
  .sort("acquiredYear_count", true)
  .head(10);

jupyter.md`## The ten busiest acquisition years

| Year | Works acquired |
| ---: | ---: |
${busiest.toRecords()
  .map((r) => `| ${r.acquiredYear} | ${Number(r.acquiredYear_count).toLocaleString()} |`)
  .join("\n")}`

The ten busiest acquisition years

Year	Works acquired
1968	20,528
1964	12,997
2008	7,443
2001	4,169
2005	4,062
2015	3,882
1974	3,656
2013	3,127
2012	2,962
2014	2,866

// What did the museum buy in its biggest year?
const busiestYear = busiest.toRecords()[0].acquiredYear as number;

const biggestYear = works
  .filter(pl.col("acquiredYear").eq(pl.lit(busiestYear)))
  .groupBy("Classification").len()
  .sort("Classification_count", true)
  .head(6);

biggestYear

Classification	Classification_count
Mies van der Rohe Archive	13622
Photograph	5457
Print	512
Illustrated Book	457
Drawing	253
Design	171

const samples = works
  .filter(pl.col("acquiredYear").eq(pl.lit(busiestYear)))
  .sort("Title").head(12)
  .select("Title", "Artist", "Classification", "CreditLine");

samples

Title	Artist	Classification	CreditLine
"Johnny"	Edward Weston	Photograph	Gift of David H. McAlpin
"M'Amenez-y"	Francis Picabia	Painting	Helena Rubinstein Fund
"The Museum of Modern Art Packed Project"	Christo	Drawing	Gift of D. and J. de Menil
"raum technik system rts" Shelving System	Hans Staeger, Manfred Malzacher	Design	Gift of Beylerian Ltd.
(Untitled)	Josef Raynefeld	Drawing	Gift of Jacques Lassaigne
(Untitled)	Josef Raynefeld	Drawing	Gift of Jacques Lassaigne
1 rue Brisemiche, Vieille Maison	Eugène Atget	Photograph	Abbott-Levy Collection. Acquired through the generosity of Shirley C. Burden, and Family of Man Fund
1/2 Brown Curved Area Series V	Robert Mangold	Print	John B. Turner Fund
1/2 Gray Curved Area Series X	Robert Mangold	Print	John B. Turner Fund
1/2 Manila Curved Area Series W	Robert Mangold	Print	John B. Turner Fund
10 de la Rue Mazet	Eugène Atget	Photograph	Abbott-Levy Collection. Acquired through the generosity of Shirley C. Burden, and Family of Man Fund
10 rue de Turenne, Vieille Maison	Eugène Atget	Photograph	Abbott-Levy Collection. Acquired through the generosity of Shirley C. Burden, and Family of Man Fund

The 1968 spike, explained

Nearly two-thirds of the museum's biggest year — 13,622 of 20,528 works — are classified Mies van der Rohe Archive. That's not a buying spree; it's a life's work arriving. Mies van der Rohe wanted his entire öuvre to go to MoMA, and the museum's trustees established the archive in 1968; the architect died in 1969. What the histogram records as a single-year spike is mostly one architect handing over ~19,000 drawings and prints — about a thousand of them by his longtime collaborator Lilly Reich. The runner-up year, 1964, doesn't have one clean story like that — it's a broad elevation across photographs and prints.

The interactive scrubber below lets you slice this yourself.

How long does art wait?

The lag between a work's creation and its acquisition is a quiet census of taste: how quickly a museum decides a medium matters.

const lagByClass = works
  .filter(pl.col("acquiredYear").isNotNull())
  .withColumns(pl.col("acquiredYear").minus(pl.col("workYear")).alias("lagYears"))
  .filter(pl.col("lagYears").gtEq(pl.lit(0)).and(pl.col("lagYears").lt(pl.lit(100))))
  .groupBy("Classification")
  .agg(pl.col("lagYears").count().alias("n"), pl.col("lagYears").median().alias("medianLag"))
  .filter(pl.col("n").gt(pl.lit(500)))
  .sort("medianLag", true);

// the extremes of the distribution, as numbers:
await jupyter.display(lagByClass.head(4));
await jupyter.display(lagByClass.tail(3));

Plot.plot({
  width: 940,
  marginLeft: 130,
  height: 300,
  x: { grid: true, label: "median years from creation to acquisition" },
  y: { label: null },
  marks: [
    Plot.barX(lagByClass.toRecords(), {
      x: "medianLag",
      y: "Classification",
      sort: { y: "-x" },
      fill: "#5b8db8",
    }),
  ],
  document,
});

Classification	n	medianLag
Notebook	2000	66
Periodical	768	54.5
Photograph	31446	41
Multiple	1102	36

Classification	n	medianLag
Video	2361	15
Installation	850	7
Print	32207	6

Who gets collected

MoMA's artist records include a gender field (as recorded by museum staff — imperfect and historical, but the museum publishes it and its gaps are themselves part of the story). The share of acquisitions credited to women, per acquisition year, is a slow-moving number worth staring at.

const genderShare = works
  .filter(pl.col("acquiredYear").isNotNull())
  .filter(pl.col("gender").eq(pl.lit("female")).or(pl.col("gender").eq(pl.lit("male"))))
  .withColumns(pl.col("gender").eq(pl.lit("female")).cast(pl.Int32).alias("isFemale"))
  .groupBy("acquiredYear")
  .agg(pl.col("isFemale").mean().alias("femaleShare"), pl.len().alias("n"))
  .filter(pl.col("n").gt(pl.lit(20)))
  .sort("acquiredYear");

await jupyter.display(genderShare.head(2));
await jupyter.display(genderShare.tail(2));

Plot.plot({
  width: 940,
  height: 300,
  y: { grid: true, domain: [0, 60], percent: true, label: "share of acquisitions by women" },
  x: { type: "time", label: null },
  marks: [
    Plot.areaY(genderShare.toRecords(), {
      x: (d) => new Date(d.acquiredYear, 0, 1),
      y: "femaleShare",
      fillOpacity: 0.15,
    }),
    Plot.lineY(genderShare.toRecords(), {
      x: (d) => new Date(d.acquiredYear, 0, 1),
      y: "femaleShare",
      stroke: "#b3589f",
    }),
    Plot.ruleY([0.5], { stroke: "#999", strokeDasharray: "3,3" }),
  ],
  document,
});

acquiredYear	femaleShare	n
1933	0	93
1934	0.016260162601626018	246

acquiredYear	femaleShare	n
2025	0.37948717948717947	780
2026	0.38415841584158417	505

The chemical century

Photography arrived as chemistry and left as software. The Medium column lets us watch the processes rise and fall — silver gelatin giving way to chromogenic prints, then inkjet.

const photoWorks = works
  .filter(pl.col("workYear").isNotNull())
  .withColumns(
    pl.when(pl.col("Medium").str.contains("(?i)gelatin silver"))
      .then(pl.lit("gelatin silver")).otherwise(pl.lit(null)).alias("process"),
  )
  .withColumns(
    pl.when(pl.col("Medium").str.contains("(?i)chromogenic"))
      .then(pl.lit("chromogenic")).otherwise(pl.col("process")).alias("process"),
  )
.withColumns(
    pl.when(pl.col("Medium").str.contains("(?i)inkjet|pigment print"))
      .then(pl.lit("digital inkjet")).otherwise(pl.col("process")).alias("process"),
  )
  .filter(pl.col("process").isNotNull());

await jupyter.display(
  photoWorks.groupBy("process").agg(
    pl.col("workYear").min().alias("firstWork"),
    pl.col("workYear").max().alias("lastWork"),
    pl.col("workYear").count().alias("works")));

Plot.plot({
  width: 940,
  height: 300,
  y: { grid: true, label: "works (binned)" },
  x: { type: "time", domain: [new Date(1890, 0, 1), new Date(2025, 0, 1)] },
  color: { legend: true },
  marks: [
    Plot.rectY(
      photoWorks.toRecords(),
      Plot.binX(
        { y: "count", filter: null },
        {
          x: (d) => new Date(d.workYear, 0, 1),
          thresholds: 68,
          fill: "process",
          fillOpacity: 0.85,
        },
      ),
    ),
  ],
  document,
});

process	firstWork	lastWork	works
gelatin silver	1838	2025	21942
chromogenic	1900	2025	2684
digital inkjet	1853	2025	1643

chromogenicdigital inkjetgelatin silver

The longest works in the collection

MoMA collects time-based media, and the dataset records runtimes. A nice excuse to pull in a package from JSR — @std/fmt's duration formatter.

import { format as formatDuration } from "jsr:@std/fmt@1.0.10/duration";

const longest = works
  .filter(pl.col("durationSec").isNotNull())
  .sort("durationSec", true)
  .head(5)
  .withColumns(pl.col("durationSec").cast(pl.Float64))
  .select("Title", "Artist", "durationSec", "URL")
  .toRecords();

const rows = longest.map((w) =>
  `| ${(w.Title as string).slice(0, 60)} | ${w.Artist ?? "Unknown"} | ${formatDuration((w.durationSec as number) * 1000, { ignoreZero: true })} |`)
  .join("\n");

jupyter.md`## The five longest works in the collection

| Work | Artist | Runtime |
| --- | --- | --- |
${rows}`

The five longest works in the collection

Work	Artist	Runtime
Interrupted Line	VALIE EXPORT	16m 38s
Cuts: Elements of Observation	VALIE EXPORT	16m 34s
Naked Doom	Edward Rankus	16m 31s
Delta: A Piece	VALIE EXPORT	16m 31s
Delta. A Piece	VALIE EXPORT	16m 27s

An interactive acquisition scrubber

Widgets are where cdp-kernel differs most from deno jupyter: the kernel side speaks the ipywidgets model protocol directly through jupyter.widget. The esm string below is the frontend — plain vanilla JS, no CDN, no bundler — and it talks back: dragging the sliders updates the model, the kernel listens for the change and pushes a fresh summary down.

const yearCounts = works
  .filter(pl.col("acquiredYear").isNotNull())
  .groupBy("acquiredYear").len()
  .sort("acquiredYear")
  .toRecords()
  .map((r) => [r.acquiredYear as number, Number(r.acquiredYear_count)] as const);

const totalAcquired = yearCounts.reduce((sum, [, n]) => sum + n, 0);

const summarize = (start: number, end: number) => {
  const inRange = yearCounts.filter(([y]) => y >= start && y <= end);
  const count = inRange.reduce((s, [, n]) => s + n, 0);
  return `${count.toLocaleString()} works (${(100 * count / totalAcquired).toFixed(1)}% of the collection) entered MoMA between ${start} and ${end}`;
};

const scrubber = jupyter.widget({
  state: {
    start: 1929,
    end: 2026,
    counts: yearCounts,
    summary: summarize(1929, 2026),
  },
  esm: `
export default {
  render({ model, el }) {
    el.style.font = "12px system-ui, sans-serif";
    const canvas = document.createElement("canvas");
    const draw = () => {
      const counts = model.get("counts") || [];
      const start = model.get("start"), end = model.get("end");
      const W = 640, H = 120, max = Math.max(...counts.map(([, n]) => n), 1);
      canvas.width = W * 2; canvas.height = H * 2;
      canvas.style.width = W + "px"; canvas.style.height = H + "px";
      const ctx = canvas.getContext("2d");
      ctx.scale(2, 2);
      const y0 = counts[0] ? counts[0][0] : 1929, y1 = counts.length ? counts[counts.length - 1][0] : 2026;
      for (const [year, n] of counts) {
        const x = ((year - y0) / (y1 - y0)) * W;
        const h = (n / max) * (H - 24);
        ctx.fillStyle = year >= start && year <= end ? "#5b8db8" : "#d8d8d8";
        ctx.fillRect(x, H - h, Math.max(1, W / (y1 - y0)), h);
      }
      ctx.fillStyle = getComputedStyle(el).color;
      ctx.fillText(y0, 2, H - 4); ctx.fillText(y1, W - 28, H - 4);
    };
    const row = document.createElement("div");
    row.style.display = "flex"; row.style.gap = "8px"; row.style.alignItems = "center";
    const mk = (key) => {
      const s = document.createElement("input");
      s.type = "range"; s.min = 1929; s.max = 2026;
      s.oninput = () => model.set({ [key]: +s.value });
      s.value = model.get(key);
      model.on("change:" + key, () => { s.value = model.get(key); });
      return s;
    };
    const start = mk("start"), end = mk("end");
    const label = document.createElement("span");
    const sync = () => {
      label.textContent = model.get("start") + " \u2013 " + model.get("end");
      draw();
    };
    const summary = document.createElement("p");
    summary.style.minHeight = "1.2em";
    model.on("change:summary", () => { summary.textContent = model.get("summary"); });
    summary.textContent = model.get("summary");
    el.append(canvas, row, summary);
    row.append(start, label, end);
    model.on("change:start", sync);
    model.on("change:end", sync);
    sync();
  },
};
  `,
});

// The kernel half of the loop: frontend slider changes arrive as comm updates;
// we recompute and push the summary back.
for (const key of ["start", "end"]) {
  scrubber.on("change:" + key, (value: number) =>
    scrubber.set({ summary: summarize(scrubber.get("start"), scrubber.get("end")) }));
}

// exercise the round trip from the kernel side:
scrubber.set({ start: 1964, end: 1980 });
console.log(`after kernel-side set(): ${scrubber.get("summary")}`);

scrubber

after kernel-side set(): 56,552 works (36.4% of the collection) entered MoMA between 1964 and 1980

Running in your browser from the saved notebook. With no kernel here, anything the kernel would do in reply doesn't happen.

One work per decade — a wall-label tour

Custom display objects are just [Symbol.for("Jupyter.display")] methods, so any renderer works. Here React SSRs a set of museum wall labels: the earliest-acquired work from each decade of the museum's life, ten works you could walk past in ninety years of buying.

import * as React from "npm:react@19";
import * as ReactDOMServer from "npm:react-dom@19/server";

const render = (node: React.ReactNode) => ({
  [Symbol.for("Jupyter.display")]() {
    return { "text/html": ReactDOMServer.renderToString(node) };
  },
});

const tour = works
  .filter(pl.col("acquiredYear").isNotNull().and(pl.col("URL").str.lengths().gt(pl.lit(0))))
  .sort("DateAcquired")
  .groupBy("acquiredDecade").first()
  .sort("acquiredDecade")
  .select("acquiredDecade", "Title", "artist", "dateText", "Medium", "CreditLine", "AccessionNumber", "URL");

function WallLabel(props: { work: Record<string, string> }) {
  const { work } = props;
  return (
    <div style={{
      border: "1px solid #ddd",
      borderRadius: "6px",
      padding: "14px 16px",
      background: "#faf9f7",
      color: "#222",
      fontFamily: "Georgia, serif",
    }}>
      <div style={{ fontSize: "11px", fontFamily: "system-ui, sans-serif", color: "#999", marginBottom: "6px" }}>
        acquired {(work.acquiredDecade ?? 0) as number}–{((work.acquiredDecade ?? 0) as number) + 9} · first purchase of the decade
      </div>
      <div style={{ fontStyle: "italic", fontSize: "15px", marginBottom: "4px" }}>
        {work.Title}
      </div>
      <div>{work.artist ?? "Unknown artist"}, {work.dateText}</div>
      <div style={{ color: "#555", fontSize: "12px", marginTop: "4px" }}>{work.Medium}</div>
      <div style={{ color: "#777", fontSize: "11px", marginTop: "6px" }}>
        {work.CreditLine}. {work.AccessionNumber} · {" "}
        <a href={work.URL} style={{ color: "#5b8db8" }}>moma.org</a>
      </div>
    </div>
  );
}

render((
  <div style={{ display: "grid", gridTemplateColumns: "repeat(auto-fill, minmax(280px, 1fr))", gap: "10px", maxWidth: 940 }}>
    {tour.toRecords().map((work) => (
      <WallLabel key={String(work.AccessionNumber)} work={work} />
    ))}
  </div>
));

acquired 1920–1929 · first purchase of the decade

Anna Peter

George Grosz, 1926–27

Pencil on paper

Gift of Paul J. Sachs. 4.1929 · moma.org

acquired 1930–1939 · first purchase of the decade

Railroad Gantry

Charles Burchfield, 1920

Gouache on paper

Given anonymously. 2.1930 · moma.org

acquired 1940–1949 · first purchase of the decade

Lend the Way They Fight

E. M. Ashe, 1918

Lithograph

Gift of Abby Aldrich Rockefeller. 1567.1940 · moma.org

acquired 1950–1959 · first purchase of the decade

Pacific Landscape

Ben Shahn, 1945

Tempera on paper over composition board

Gift of Philip L. Goodwin. 1.1950 · moma.org

acquired 1960–1969 · first purchase of the decade

Airplane Hangar, project

R. Buckminster Fuller, 1955

Wood and paint

Gift of the architect. 560.1960 · moma.org

acquired 1970–1979 · first purchase of the decade

Les Terrasses, Villa Stein-de-Monzie, Garches, France (Scale model, 1:33.3)

Le Corbusier (Charles-Édouard Jeanneret), 1926–1928 (model 1970)

Acrylic, wood, metal and paint

Lily Auchincloss Fund. 668.1970 · moma.org

acquired 1980–1989 · first purchase of the decade

Cantaloup

Steina, 1980

Video (black and white and color, sound)

Acquired through the generosity of Barbara Pine. 735.1980 · moma.org

acquired 1990–1999 · first purchase of the decade

Certain from That Is Right And Other Similarities

Edward Ruscha, 1989

One from a portfolio of twelve lithographs

. 216.1990.1 · moma.org

acquired 2000–2009 · first purchase of the decade

Jake at the New Viet Huong

Elizabeth Peyton, 1995

Oil on masonite

Gift of the Peter Norton Family Foundation. 15.2000 · moma.org

acquired 2010–2019 · first purchase of the decade

New York

Helen Levitt, 1984

Chromogenic print, printed c. 2005

Gift of Marvin Hoshino. 43.2010.x1-x2 · moma.org

acquired 2020–2029 · first purchase of the decade

Cover and Magazine Design (Obálka a úprava časopisu)

Ladislav Sutnar, c. 1938

Bound album with printed and colored paper

Gift of Merrill C. Berman and Dalia S. Berman. 856.2019 · moma.org

What we found

• A museum is its bequests. Two years tower over a century of buying: 1968 (20,528 works) and 1964 (12,997) — and 1968 is not curatorial appetite: 13,622 of those works are the Mies van der Rohe Archive, one architect handing over his life's work. The scrubber shows the scale: 1964–1980 alone accounts for 56,552 works, over a third of the collection.
• Prints are bought hot; photographs arrive cold. Median lag from creation to acquisition: 6 years for prints (n=32,207) against 41 years for photographs (n=31,446). Notebooks and periodicals wait longest (54–66 years) — research material ages in drawers before anyone thinks to collect it.
• The gender line moves at glacial speed. Among acquisitions whose primary artist's gender is recorded, the share credited to women goes from 0% in 1933 to about 38% in 2025–26. Two-thirds of a century to reach a little more than a third.
• The chemical century, then the digital one. Gelatin silver dominates the photographic record (21,942 works), chromogenic color runs through the collection's middle age (2,684), and digital inkjet arrives last (1,643) — the relay baton-pass is visible in the chart above.
• Even the longest works are short. The longest cataloged runtimes here are about 17 minutes — VALIE EXPORT leads the list — because MoMA's feature-film holdings live in a separate film database, not in this artworks dataset.
• The decade tour ends where the museum began. Its earliest label is Anna Peter by George Grosz, accession number 4.1929 — the fourth work the museum ever acquired, in its founding year.

Caveats, honestly stated

• MoMA publishes this data as "research in progress": it is not curator-approved, and fields like Gender (as recorded by museum staff) and free-text Date are approximations. The work year, in particular, is a regex, not scholarship.
• Images are not part of the dataset; every link above goes to the museum's own collection page.

Licenses

• Data: MoMA collection dataset, CC0 (public domain) — see the source repository.
• This notebook: part of cdp-kernel, MIT, same as the rest of the repository.
• Analysis approach inspired by Trevor Manz's Deno post linked at the top; no code or text is reproduced from it.