The Periodic Table of Taylor Swift

Status

live

Surfaces

5

Demo

open

Hard part

schema before viz

why

In 2003 chemistry class I drew a Periodic Table of Milk Mustaches: atomic numbers, columns, rows of properties, the format used as a joke and outliving the joke. Twenty-three years later, looking at the Swift catalogue, that frame was the one that finally fit.

Every Swift project on the internet is a tier list, a ranking, or a chronology. They all answer the same question (what’s best, in what order) and they all flatten the catalogue into a line. The catalogue isn’t a line. It’s a structure: songs cluster by era, by emotional charge, and by something like reactivity, which is how readily a song bonds with another song years away. That’s a periodic logic, not a chronological one.

The Periodic Table of Taylor Swift (ts.hosette.net) is the site that takes the metaphor seriously. 283 songs as elements with stable numbers, symbols, columns, rows, and bonds, plus a working lab notebook of the decisions that shaped the grid.

what I built

Songs

283

Eras

7

Families

18

Bonds

234

Findings

13

Five surfaces, one dataset:

• The Grid: an 18-column table over 7 chronological periods. Eras as rows, emotional families as columns, vault tracks shelved at the bottom.
• Insights: 13 R-series findings rendered as scrollytelling charts (vocabulary curves, producer fingerprints, controversy timelines, a 3rd-person POV ledger).
• The Mirror Clock: a 12-spoke radial visualization of 234 thematic connections (reflections, evolutions, echoes, contrasts) drawn as chord bundles between songs that answer each other across years.
• The Lab Notebook: a dated, numbered, never-edited record of every method change and anomaly, in the project’s own voice.
• The Reading Guide: legend, glossary, orientation tour, scoped to the lab’s typography.

The demo loads with no signup and reseeds itself. The catalogue is the pitch.

The grid is the surface. Underneath, every song carries four nested rings of properties: from identity (atomic number, title) to atomic (era, family, mass, reactivity) to relational (bonds, mirror pair, group) to set (traits, controversy score, recording-era flag). The schema is the project; this is its shape:

the chemistry

The metaphor is doing more than decoration. The actual periodic table works because of three load-bearing ideas, and the Swift catalogue maps onto each:

• Atomic number, the stable ID. In chemistry, atomic number is the count of protons that makes hydrogen hydrogen regardless of isotope or compound. The Swift analog is chronological-release position 1–283, picked specifically because it’s the one fact about a song that doesn’t move when the catalogue does. Taylor’s Version re-recordings keep the original’s number, vault tracks slot in around their parent album, and a future re-release wouldn’t renumber the universe. Anything else can change. The number can’t.
• Periodicity, properties that recur. Group I in chemistry (lithium, sodium, potassium, the alkali metals) is meaningful as a column because every member has one electron in its outer shell and behaves similarly. A column you can’t make a prediction from is a label, not a structure. The Swift analog is the eighteen emotional families. Family on this grid means a column you could ask a question of: which Reputation-era song answers the Lover-era song in the same family? The grid is doing real work whenever the answer is a specific second song.
• Predictive gaps. Mendeleev’s 1869 table left blank cells where the schema said an element ought to exist, and predicted gallium and germanium by position alone before either was isolated. For Swift, the gap principle produced a weekend with fan wikis: the schema said certain era × family cells should have more songs than the automated source had, and 49 of those gaps turned out to be real songs the streaming catalogues hadn’t surfaced. The schema knew before the data did.

The bond layer sits on top of all three. In chemistry, bonds are how elements combine (covalent, ionic, hydrogen), each a different relationship determined by where the elements sit on the table. The 234 bonds in the Mirror Clock are typed the same way: reflection, evolution, echo, contrast. The strongest bonds cross eras and stay in the same family, the way the strongest chemical bonds cross row boundaries while staying inside a group. The grid predicts the bonding; the bonding confirms the grid.

That’s why the metaphor isn’t decoration. The chemistry is borrowed for the organizing discipline that made chemistry a science instead of a list: atomic number as stable ID, periodicity as predictive, gaps as features. A pretty grid of song tiles wouldn’t have caught the 49 missing songs.

decisions I’d defend

Treat the metaphor as a constraint, not a skin. I could have shipped a pretty grid and called it a day. Instead, every observation about the catalogue has to be expressible in the same coordinate system: era × family, with bonds as a separate layer. If a finding can’t be drawn on the table, the finding probably isn’t real. Same rule that runs through Golden Cell. That rule killed three “interesting” charts and forced one anomaly (Ronan) to become a trait flag instead of a new row.

Build a lab notebook before building features. The Notebook page exists because I kept making the same data decision twice. Now every method change (promoting TV-only vault tracks onto the grid, splitting controversy into public + fandom sub-scores, choosing fit-width as the default zoom) gets a permanent dated entry. Rejected ideas stay rejected. Readers see the working, not just the result. It’s the most-loved page on the site and the cheapest one to maintain.

Commit to one identity, not a theme system. The “Lab” voice (chartreuse on near-black, Instrument Serif italic display, JetBrains Mono for atomic numbers, 2px sharp corners, small-caps section markers) is enforced project-wide via a memory file the AI consults on every change: no second saturated color, no rounded pill badges, no purple gradients. Visual budget stays small on purpose, so each era’s palette can take over a drawer without fighting the chrome.

what I learned

Schema before viz. Atomic number, atomic mass, reactivity, family, group: deciding what each meant for a song was the project. Once the schema stabilized, the layout fell out of it in an afternoon. Most data-viz projects skip this step and end up with a beautiful chart of an unconsidered dataset.

Bonding is a sharper question than ranking. Every Swift project on the internet sorts the catalogue. None ask which song bonds with which, across eras. Reactivity (will this song chemically react with that one) turned out to be more revealing than any ranking system, and harder to fake.

The dataset is the product. I spent a weekend cross-referencing four fan wikis and BMI registrations to find 49 missing songs. That work is invisible in the interface. It’s also the entire reason the site is trustworthy. Same hand-edited-after-AI pattern in Pocketbook, where the model writes the first-pass caption and I catch the misreads.

Show your mistakes. The Notebook started as an internal scratchpad and became the page that earns the most trust, because it admits the catalogue used to be wrong and explains how it got fixed. The project keeps a record of itself, in its own voice, that the project then has to stay consistent with.

what would prove it

The project deliberately doesn’t measure itself. If it did, three things I’d want it to show:

• The Lab Notebook is the trust metric. The case study claims the Notebook is the page that earns the most trust. The proxy: do users who read a Notebook entry spend more time on the rest of the site, and come back more often, than users who don’t? If they do, the show your mistakes discipline has measurable weight. If they don’t, the Notebook is a flourish.
• Bonds retain; findings alone plateau. The 13 R-series findings are scrollytelling. The Mirror Clock is the bond layer that turns a finding clickable into another song. Users who jump from a finding into a bond visualization should come back more than users who read findings and leave; if they don’t, the surfaces are siloed and the integrated story is one I’m telling myself.
• The schema generalizes, or the periodicity claim is decoration. The Mendeleev test was that periodic logic predicted gallium and germanium before they were isolated. The TS analog is that the era × family × bonds model would surface gaps in another long catalogue (Beyoncé, Springsteen, Mitski) the same way it surfaced the 49 missing Swift songs. If it does, the schema is doing real work. If the model breaks the moment the artist changes, the chemistry is borrowed for ornament, not for organizing discipline.

Two risks the project has to keep budgeting for:

• Catalogue maintenance is unbounded. Every new album, vault track, re-recording, or controversy update is a catalog event. 283 songs is a snapshot, not a fixture. The Lab Notebook is part of the answer; the rest is a versioned update workflow that doesn’t currently exist. Without it, findings get stale, bond counts shift quietly, and the dataset’s trust posture starts to leak.
• The metaphor has a learning curve, and the first thirty seconds matter. First-time visitors who don’t know that atomic number = chronological release position can bounce before they reach the Reading Guide. The hedge is making the grid pleasurable to look at before it has to be parsed; if first-session bounce stays high, the metaphor is too thick to enter and the orientation needs more weight, not less.

the receipts

The Lab Notebook is the most-loved page on the site. The discipline that produces it is portable: the public starter is at /lab-notebook-starter.md, with the load-bearing rule from TS quoted at the top:

The premise: most research and data projects pretend to a polished finality. A lab notebook makes the messy middle visible by recording, in chronological order, every methodology change and every anomaly. The rule is simple. When something changes, you write a new entry. You do not rewrite the old one. Reversals are noted in place (“reversed YYYY-MM-DD because X”), not deleted.

The polished result is trustworthy because the messy middle is readable.

try it yourself

Drop /lab-notebook-starter.md at the root of a research, data, or content project. Rename to whatever fits (notebook.md, lab.md, findings.md) and start adding numbered, dated entries. Sections cover method changes, anomalies, findings, and reversals, noted in place rather than overwritten, so the file ages legibly. Sister to decisions-starter.md: that one records what you decided not to build, this one records what you observed and how the method changed. The never quietly edit history clause is the actual mechanism.

method extracted: Schema-first interfaces

The Periodic Table of Taylor Swift is the project. Schema-first interfaces is the reusable method underneath it. This project turns a cultural corpus into a structured interface. The work is not just visual theming; it is taxonomy design, relationship mapping, and interaction modeling.

Method: (1) define the smallest meaningful units, (2) map relationships, (3) design surfaces that let people explore without needing to understand the underlying model. Before the interface could be playful, the schema had to be serious.

See the Methods index.

what’s next

• Open the dataset: publish the 283-song corpus as a downloadable CSV with the full schema (eras, families, bonds, mirror pairs, traits) so other people can run their own analyses against it.
• Generalize the engine: the era × family × bonds model isn’t Swift-specific. It works for any artist with a long catalogue and enough fan-driven taxonomy to argue about. Beyoncé, Springsteen, Mitski on the same grid.

Live site: ts.hosette.net.