Archived exploration · sourced, approximate dates
The civilization clock
Stone tools fill almost the whole day. Agriculture arrives about five minutes before midnight, writing two and a half, industry seven seconds, electronic computing two. The foundation-model era is the last sixth of a second.
This exploration is archived. It is kept, with its sources and the film The Last Second, as earlier work; the current explorations are here.
The mark
The Aperture Clock is the reason for the Ostrium logo.
The mark is this clock at midnight, reduced to three strokes. The arc is the last stretch of the day descending to one instant, the present. The rising line is the instrument's line of sight, meeting it on the horizon. Both stop short of that instant, because the question it poses is still open. One short line carries on past it: what comes after midnight.
arc · the day's final stretchline · the line of sightgap · the open questiontail · after midnight
After midnight
A century into a second?
| Unit of the century | Passes in | Per 60 Hz frame | On the strip |
|---|---|---|---|
| second | 27.4 µs | 609 | only countable |
| minute | 1.64 ms | 10.1 | only countable |
| hour | 98.6 ms | < 1 | watchable |
| day | 2.37 s | < 1 | watchable |
| month | 1.2 min | < 1 | longer than the strip |
| year | 14.4 min | < 1 | longer than the strip |
| decade | 2.4 h | < 1 | longer than the strip |
| century | 1 day | < 1 | longer than the strip |
At ×36,525, a century passes in 1 day. Individual hours and days can be watched passing. Seconds and minutes complete more than once per 60 Hz frame — they can only be counted.
A replay of one reference century (3,155,760,000 s) at a chosen factor. Only its calendar units pass — no discoveries are invented. The strip shows the reference time that passed in the last four seconds of yours: units far enough apart are drawn singly, denser ones merge into tone, and longer ones fill as bars. Past one per display frame, a unit can only be counted. A ratio of durations, not a measured speed-up — as a research metric TC needs a defined capability, a baseline and an independent check (TC-H).
Fit one century of progress into
The ratio says how much shorter the interval would be, nothing about how. Progress is not one number: a model a billion times faster at text has not compressed a century of anything. As a research metric, TC needs a defined capability, a measured baseline and an independent check — which is what TC-H registers.
Time since the Big Bang, in seconds
435 084 631 461 305 861
The clock above folds 3.3 million years into a day. The universe is about 4,200 times older: 13.787 ± 0.020 billion years (Planck 2018). The measurement fixes about the first three digits; the other fifteen sit inside the uncertainty, though the last one changes every second.
What the clock shows
A change of scale. Years become positions on a ruler; a day is a frame that makes 3.3 million years legible at once.
Each milestone y years before 2026 sits 86,400 × y ÷ 3,300,000 seconds before midnight. The zoom is logarithmic, so every equal movement shows a window the same factor shorter.
What it does not prove
That the opening is settled. The 3.3-million-year anchor is the Lomekwi 3 stone tools, a date still debated; the Oldowan, about 2.6 million years ago, is the conservative alternative. Starting there would stretch every later milestone's share of the day by roughly a quarter — the shape stays, the numbers move.
That progress follows a law of acceleration. The dense final second is partly a property of which milestones were chosen; choose others and the picture changes. A capability's invention, diffusion and dependable use can be decades apart, and none of them arrived for everyone at once.
The question it motivates is prospective and narrower — whether one research loop can be made measurably faster — and it belongs to a registered protocol, TC-H, not to a picture.
Sources
Clock sources, scale and transcript
This is an illustrative rescaling of selected historical transitions, not evidence of a universal law of acceleration. The dates are approximate; transitions occurred unevenly across societies. The long sparse span does not mean that no technology developed between the selected milestones.
H = 3,300,000 years. Seconds before midnight = 86,400 × years before present / H. Present is fixed to 2026 for reproducibility. BCE conversion uses no year zero. Milliseconds describe the mapping arithmetic, not the precision of the historical dates.
| Transition | Historical anchor | Approximate clock | Source / definition |
|---|---|---|---|
| EARLY STONE TOOLMAKING | ≈ 3.3 million years ago | ≈ 00:00:00.000 | Harmand et al., Nature (2015) ↗ Lomekwi 3 stone artifacts; hominin technology predating Homo sapiens. This evidence anchors the selected horizon, not a definitive first invention. |
| AGRICULTURE / SETTLEMENT | ≈ 12,000 years ago | ≈ 23:54:45.818 | Smithsonian Human Origins ↗ A broad transition, occurring at different times in different regions; not a single invention date. |
| EARLY WRITING | ≈ 3300 BCE | ≈ 23:57:40.582 | British Museum, origins of writing ↗ An approximate anchor for early Egyptian and Sumerian writing, not the date of a universal transition. |
| INDUSTRIALIZATION | ≈ late 18th century | ≈ 23:59:53.036 | Science Museum, Steaming through the centuries ↗ 1760 is the illustrative anchor for a British transition spanning roughly 1760–1830; industrialization is geographically uneven. |
| ELECTRONIC COMPUTING | ≈ 1940s | ≈ 23:59:57.905 | Computer History Museum, 1946 ↗ 1946 and ENIAC anchor this selected transition. Electronic computing has earlier and competing milestones. |
| NETWORKED COMPUTING | 1983 TCP/IP transition | ≈ 23:59:58.874 | Internet Society, A Brief History of the Internet ↗ ARPANET TCP/IP cutover is the chosen definition. Packet networks, the Internet and the Web have different histories. |
| FOUNDATION-MODEL ERA | ≈ early 2020s | ≈ 23:59:59.843 | Brown et al., Language Models are Few-Shot Learners (2020) ↗ GPT-3 is the illustrative 2020 anchor, not the invention of AI or generative modeling. |