◀ Deep Time Games

About

How this was built

A study tool for geology and anthropology, made by one student working with an AI — and checked, model by model, against the published values it claims to reproduce.

Deep Time Games began as one undergraduate’s way of remembering the record in relation to itself — how the periods, the sediments, the water and the ancestors all connect — instead of memorising a scale and hoping. It grew into fourteen games covering two disciplines, and it is free for anyone to use.

14games
70sections
56live models
31drawn figures
336explained answers
0dependencies

Every page is a single HTML file with no frameworks and no libraries. The site counts anonymous usage — which games and sections get opened, and how often questions are answered correctly — with no names, accounts or cookies. Your personal record never leaves your browser. The heaviest page is 52 KB over the wire.

Who did what

This was a collaboration, and it is more interesting described accurately than flattened in either direction.

The student

  • Set the goal: a relational study tool, not a flashcard drill
  • Supplied every piece of source material — syllabi, lecture notes, chapters, lab assignments, field-school archives
  • Made the editorial calls: fictional sites only in archaeology, anthropology before geology, which topics belonged and which did not
  • Tested the games and reported bugs, including ones that had shipped
  • Decided what stayed off the site entirely

The AI

  • Wrote the code and the prose
  • Built and numerically verified the physical models
  • Audited coverage against each course and the licensure exam
  • Found and fixed its own defects — and several times reported a finding wrong before re-checking

Every model was checked

This is the part that took the time. A simulation that looks plausible and is quietly wrong teaches confident nonsense, so each one was run against an independent calculation or a published value before it went live. A sample:

Genetic drift

The page claims a neutral allele’s fixation probability equals its starting frequency. Tested across 4,320 simulated populations at three starting frequencies.

observed 0.211 / 0.492 / 0.813  vs  expected 0.2 / 0.5 / 0.8

Earth’s hypsometric curve

The curve’s anchor points were tuned until integrating the rendered curve reproduced the published means, then re-derived from the page’s own readouts as an independent check.

mean continental +845 m (published 840)  ·  mean ocean depth −3,802 m (published 3,800)

Heat flow through the crust

Fourier’s Law with standard crustal values, checked against measured global averages.

computed 75 mW/m²  ·  continental average ≈65, global ≈87

Ternary diagram geometry

Plotted points were read back off their own pixel positions into barycentric coordinates and compared with the inputs — at the three apexes, the three edge midpoints, the centroid and an interior composition.

exact at 8 of 8 test points

And where they fail, it says so

A model that hides its own limits is worse than no model. Where an equation breaks down, the page that uses it says where and why.

Kozeny–Carman permeability

Reproduces measured values well for sands and silts. For clay-rich soils it overpredicts by three orders of magnitude or more, because it takes no account of electrostatically bound water or platy particle shape. The page says to treat a clay result as an upper bound and get a lab test.

The plasticity chart

It cannot distinguish organic soils from inorganic silts — they occupy the same region below the A-line. Separating them needs a second test, and the page explains which one.

Seafloor age and depth

Depth follows the square root of age only to about 80 million years. The curve is drawn solid inside that range and dashed beyond it, where real seafloor flattens and the half-space model stops applying.

Errors it caught in itself

Verification is only worth anything if it sometimes fails. Three examples that shipped wrong and were caught by checking rather than by reading:

An inverted convention. Folk & Ward sorting came out negative, and glacial till — the textbook example of very poor sorting — was reported as very well sorted. The phi scale runs opposite to millimetres, so φ84 corresponds to the D16 diameter. The site now teaches that trap explicitly, because it is the one everybody hits.

Sliders that lied. Several range inputs had step values that put the number you wanted off the grid, so the browser silently snapped it — a sea-level preset of 0 m became −5 m, and a peaty-organic soil preset of 2,200 kg/m³ was silently clamped to 2,400, the bottom of its range, so the page reported a porosity for a soil nobody had asked for. Caught by a test that clicked every preset and read the slider back to check it held the value requested.

A false alarm, reported honestly. One audit flagged 70 unused CSS rules. Acting on it would have deleted the styling for four SVG diagrams — the detector could not see classes set from JavaScript. The real figure was 35.

What it covers

Geology — deep time and the stratigraphic record, minerals and thin sections, maps and structures, sediment and weathering, groundwater and surface hydrology, engineering ground, ore and energy. Mapped to the eight domains of the ASBOG Fundamentals of Geology licensure exam.

Anthropology — all four fields: human osteology and the biological profile, primates and human variation, archaeological field method and law, and kinship, language and ritual.

Content is built from real course material — the International Chronostratigraphic Chart, the Hjulström curve, Folk & Ward statistics, Darcy’s law, the Unified Soil Classification System, the Powers roundness chart.

Every archaeological site, case and individual on this site is fictional. No real excavation, collection or person is described, by deliberate choice.

Who it is for

Students taking these subjects, and anyone curious about them. It was built to be shared with university geology and anthropology departments, and it is free, requires no account, and stores nothing about you beyond a saved progress marker in your own browser.

If something here is wrong, it is worth telling us — the point of showing the working is that the working can be checked.