Deep Time

→ emal-avala.github.io/deep-time

An interactive timeline of 617 events spanning 13.8 billion years — from the Big Bang through the dinosaurs, the ice ages, the Buddha and the ancient empires, to generative AI — readable on both a linear and a logarithmic time scale.

The point of the two scales: on a linear axis, all of recorded human history is a sliver a few pixels wide at the right edge. On a logarithmic axis that sliver opens up and the deep past compresses, and you can see the shape of the whole thing at once. Switching between them is the tool.

Run

Use the hosted version, or clone and open the file directly:

git clone git@github.com:emal-avala/deep-time.git
cd deep-time && open index.html

No server, no build, no dependencies.

Reading it

Lanes run from substrate to superstructure: Cosmos & Earth → Life → Human Origins → Science & Tech → Culture & Art → Religion & Ideas → Empires. The physical world first, then life, then humans, then what humans made and believed, and last how they organised themselves into states. The order isn't decorative — the staircase of first appearances down the left side is itself the story, and the things that depend on everything above them sit at the bottom.

The strata band at the top carries the named ages and eras — geological eons through the Bronze Age, the Middle Ages, the Information Age — stacked so longer eras sit above the shorter ones nested inside them.

The density rail at the bottom counts events per interval. Read it as attention density, not ground truth: it reflects what this dataset records and what history preserved, both of which lean heavily toward the recent and the literate.

Controls

Scale	Linear / Logarithmic, plus a slider to scrub continuously between them
Window	Preset zoom ladder — All time, Dinosaurs, Sapiens, Farming, Century…
drag · scroll	pan · zoom
L · 0 · Esc	toggle scale · reset · close panel
Search	type to dim everything else; Enter jumps to the match
Table	every event as sortable text, the accessible path to the same data

Deep links

State lives in the URL, so a view can be bookmarked or shared:

index.html?scale=linear&window=Sapiens
index.html?from=-3000&to=500&scale=log     # from/to are calendar years
index.html?q=buddha&theme=light

Three things to look for

1. Switch to linear at "All time." Everything after the first cell divides collapses into the last few pixels. That compression is the honest picture.
2. Watch the density rail as you switch scales. Under log time, human events spread into a fairly even band — the appearance of steady progress is partly an artifact of the scale.
3. Zoom the Science & Tech lane from "Farming" to "Century." The gap between milestones shortens by orders of magnitude; the lane goes from readable to a solid bar of overlapping marks.

Data

data/events.js is generated. Rebuild it with:

node tools/build-dataset.js

It reads data/sources/ and applies the whole cleaning pipeline: validation, the adjudicated corrections, de-duplication, sorting.

How it was compiled. Twelve domain-historian agents (deep time, human evolution, the ancient Near East, India and East Asia, classical Mediterranean, world religions, the medieval world, early modern, industrial and modern, plus a technology throughline) each produced their era, deliberately overlapping. Four gap-fill agents then deepened the thinnest lanes. Fifteen auditors re-derived every date independently and proposed corrections; an adversarial adjudicator reviewed each proposal with instructions to defend the original data. Of thirteen proposed corrections, the four date changes were all rejected as within accepted scholarly range and nine genuine duplicates were removed. Raw agent output is preserved in data/sources/raw-events.json.

Dates are consensus estimates. Deep-time and prehistoric figures are approximate by nature and some are actively debated — the Buddha's dates, the peopling of the Americas, when hominins controlled fire. Each event carries a confidence field of exact, approximate, or debated. Treat this as a tool for seeing shape and proportion, not for citing dates.

Decisions worth remembering

Time is stored as bp (years before present, present = 2026), not as calendar years. Both scales map the same bp window onto the plot width, so switching redistributes the interior without moving the endpoints — which is what makes the morph between them legible rather than disorienting.

Logarithms can't reach zero, so "now" is clamped to 1 year ago. The right edge is the present, and events in the last year pile onto it.

Canvas, not SVG or DOM. 617 events with labels re-laid-out on every frame of a pan, zoom, or scale morph; canvas holds 60fps where a few thousand DOM nodes would not. The cost is that hit-testing and the accessible view had to be built by hand — hence the table view.

Labels are packed per lane on every frame into three sub-rows, degrading gracefully: labelled → unlabelled mark → a tick on the lane's density rail. Every event always leaves a mark, even where no label fits, so the shape of a dense period stays honest.

Colour was validated, not chosen by eye. The seven category hues clear CVD-separation, contrast, and lightness-band checks against both the light and dark surfaces. Text drawn on a coloured bar picks black or white by computing both contrast ratios — a lightness threshold picks white on mid-tone pink, which measures 2.7:1. There is deliberately no UI accent colour: the data owns all the colour on the page.

Reckonings of time

The timeline is not tied to one cosmology. The renderer knows nothing about years, kappas or baktuns - it only knows distance from now on a log axis. Each tradition supplies its own units, threads and data as a drop-in atlas.

Three ship today:

Atlas	Shown as	Unit	Span
physical	Modern	years	13.8 Ga to now
kappa	Buddhist	mahakappas	10^6 to 10^-13 kappas
yuga	Hindu	years	311 trillion years to now

Switch with the Modern / Buddhist / Hindu toggle in the header, or with ?atlas=kappa. Buttons name the tradition rather than its unit, since the unit is already on the axis.

Adding a tradition

Create data/atlases/<id>.js and add a <script> tag for it in index.html. Nothing else in the app changes.

window.DEEP_TIME_ATLASES = window.DEEP_TIME_ATLASES || [];
window.DEEP_TIME_ATLASES.push({
  id: 'yuga',
  label: 'Yuga',
  tradition: 'Hindu cosmology',
  thesis: 'One sentence shown as the page subtitle.',
  unit: 'years',
  nowLabel: 'now',

  scale: 'unitsBeforePresent',   // or 'signedYears'
  field: 'year',                 // record field holding the position
  format: 'years',               // which formatter labels the axis

  ceil: 3.11e14, floor: 1,       // outer bounds of the axis
  home: [3.11e14, 1],            // the view it opens on
  ageKey: 'Ages',                // category routed to the strata band
  tickGap: 96,                   // px between axis labels

  categories: [ { key: 'Yugas', label: 'Yugas', v: '--c1' } /* up to 7 */ ],
  ladder:     [ { label: 'All of it', max: 3.11e14, min: 1 } ],
  events:     [ /* records */ ]
});

Scale strategies turn a tradition's own units into distance from now:

• signedYears - records carry start/end as signed calendar years (negative BCE). Used by physical.
• unitsBeforePresent - records already count backwards from now in the tradition's unit, named by field (kappa/kappaEnd, baktun/baktunEnd). A negative value means still to come: it is plotted by how far ahead it lies and flagged, because a logarithm has no negative side.

Formatters live in FORMATS in app.js. years and kappas exist; a new unit adds one function. Below one unit the kappa formatter switches to exponents (10^-6 kappa), since the whole point of these axes is that the numbers stop being readable in decimal.

Each record needs name, category, the position field, kind (moment/period/age), description, significance and confidence. Everything else - search, the table, the detail panel, mobile, linear/log - is inherited.

On honesty across traditions

These cosmologies do not agree, and the app must not pretend they do. Each atlas keeps its own confidence vocabulary: the scientific one uses exact/approximate/debated; the Buddhist one uses canonical/traditional/derived, because the distinction that matters there is whether a claim is in the suttas, in the commentaries, or a modern calculation.

The Hindu scheme is the opposite case, and worth stating plainly: where the Pali canon refuses a number, the Puranas are exact and internally consistent. The build script for that atlas refuses to emit unless the chain closes - 1,728,000 + 1,296,000 + 864,000 + 432,000 = 4,320,000; 1,000 mahayugas = one kalpa; 14 manvantaras plus 15 junctions tile that kalpa exactly; 100 years of Brahma = 311,040,000,000,000 years. A tradition that gives you a checkable number should be checked.

That last category earns its keep. The Pali canon refuses to put a number of years on a kappa - asked directly, the Buddha answered with two similes instead (SN 15.5, a mountain worn away by a passing silk cloth; SN 15.6, a city of mustard seeds emptied one seed a century). Every year-figure in the kappa atlas is therefore a modern derivation and is marked as one. Placing a tradition on an axis should not quietly convert its refusals into numbers.