---
title: "Twenty-one terabytes under the permafrost"
url: "https://eternalos.com/blog/twenty-one-terabytes-under-the-permafrost"
description: "In 2020 GitHub put a snapshot of every active public repository on film in a mine on Svalbard, designed to last a thousand years. The film is the least interesting part."
---

# Twenty-one terabytes under the permafrost

September 9, 2026·2 min read·EternalOS

On 2 February 2020 GitHub took a snapshot of every active public repository on its platform. Over the following months the Norwegian company Piql wrote about 21 TB of it onto 186 reels of film, and on 8 July 2020 the reels were placed in the [Arctic World Archive](https://archiveprogram.github.com/arctic-vault), a vault in a decommissioned coal mine on Svalbard, deep in the permafrost. The design target is a thousand years.

It is easy to focus on the setting: the mountain, the cold, the remoteness. The interesting decisions were made elsewhere.

## The medium is chosen for the reader

piqlFilm is photosensitive archival film. Data is written as dense visual patterns that a computer can scan, while parts of each reel are plain text and images a person can read with a magnifying glass. Film of this kind is expected to last for centuries in stable cold, and reading it needs only light and a lens to start.

That is the right priority. The question for a thousand-year archive is not how many bits can be packed in but what a future reader needs to get started, assuming they have none of today’s software.

## The archive explains itself

Each reel opens with a _Guide to the GitHub Code Vault_ in five languages: what the archive is, how it is organised, how the data is encoded, and how to rebuild the tools to decode it. The software for retrieving the data is open source and stored in the archive alongside the code it reads.

This is the single most important feature of the project, and the one most personal archives lack entirely. A record that does not explain itself depends on someone who already understands it, and in a thousand years — or fifty — that person will not exist.

## What it chose not to be

**A snapshot, not a living archive.** It holds one moment in early 2020. Nothing written since is in it, and nothing in it is updated. That is deliberate: the vault is the cold, last-resort layer, while the working copies live elsewhere.

**One of several copies.** GitHub described the vault as part of a wider programme with other partners and formats, not as the only home for the code. A deposit like this protects against the worst case precisely because it is not the everyday case.

## What transfers to a single life

The scale is different by many orders of magnitude, and the lessons are not.

**Put the instructions inside the record.** An index that says what is there, how it is organised and how each format is read should travel with the material in plain text.

**Choose at least one form a person can read unaided.** Printed key documents and a written guide survive the loss of every device.

**Keep a cold copy that nobody touches.** Separate from the copy that gets used and updated, held somewhere else, on a different medium.

None of this needs a mine on Svalbard. It needs the same decisions, made deliberately.

-   arctic code vault
-   self-describing archives
-   film
-   svalbard

## Keep reading

-   [Who will verify the record is genuine in a hundred years](https://eternalos.com/blog/who-will-verify-the-record-is-genuine-in-a-hundred-years)
-   [The password that dies with you: access decades later](https://eternalos.com/blog/the-password-that-dies-with-you)