History & Archaeology Creator Directory

Live: https://www.openhistorymap.org/creators/

A directory of history / archaeology creators where every reel and video is indexed by the time period it talks about and the places it talks about — so you can ask "what has anyone made about the Aegean between 1300 and 1100 BCE?" rather than "what did this channel upload last month".

211 items from 30 creators in five languages, across 154 places in 41 countries, spanning 43 000 BCE to 2023.

The whole thing is three CSV files and a static page. There is no server-side code, no framework, and no build step.

Want your channel listed? See CONTRIBUTING.md — open an issue and a bot opens the pull request for you.

data/influencers.csv   who the creators are
data/places.csv        the gazetteer (id, name, lat, lon, wikidata)
data/videos.csv        the index: one row per reel / video
data/pending.csv       machine-written: new uploads with no period or place yet
index.html app.js style.css   the browser UI (Leaflet + a warped timeline)
tools/validate.py      referential-integrity check over the CSVs
tools/build_sqlite.py  optional: derive directory.db for SQL querying
tools/refresh_feeds.py poll the channels for uploads not yet indexed
tools/issue_to_row.py  turn a submission issue into validated CSV rows
tools/channel_videos.py read a channel's videos/shorts tab, past the 15-item feed
tools/build_pending.py  write data/pending.csv: new uploads, not yet indexed
tools/geocode.py        turn a place name into a real coordinate (Nominatim)
tools/build_site.sh     assemble _site/ for GitHub Pages

Run it

./serve.sh            # http://localhost:8088/

The page fetch()es the CSVs, so opening index.html over file:// will not work — it needs to be served over HTTP. Any static server does; serve.sh is just python3 -m http.server.

The data model

influencers.csv — one row per creator.

column	meaning
id	slug, referenced by videos.influencer_id
youtube_channel_id	canonical UC... id — the key tools/refresh_feeds.py polls
name, handle	display name and primary @handle
youtube, instagram, tiktok, facebook, website	profile URLs, any may be blank
language	ISO code of the main language
focus	free text — what they cover
avatar_url	optional image URL
verified	yes once the links have been machine-checked
notes	free text

places.csv — the gazetteer. Each place is entered once and referenced by id, so coordinates stay in one place when twenty videos mention Pompeii. wikidata is the Q-number, for later reconciliation against OHM.

videos.csv — the actual index. One row per item.

column	meaning
id	slug/serial
influencer_id	→ influencers.id
title	as published
platform	youtube, reel, short, tiktok, instagram, facebook, podcast, article
url	direct link to the item
thumbnail	optional image URL
published	upload date, YYYY-MM-DD, optional
year_start, year_end	the period the content is about, not the upload date. Integers; negative = BCE. Equal values mean a single year.
era	human label for that span, e.g. Late Bronze Age
places	;-separated places.id values
tags	;-separated free tags
verified	yes once the URL, title and creator are confirmed (see Provenance)
notes	free text

year_start / year_end are the join key to the rest of OHM: they are plain years on the same axis the tileservers use, so this directory can later be overlaid on OHM map layers without a conversion step.

Adding an entry

1. If the creator is new, add a row to data/influencers.csv.
2. If a place is new, add a row to data/places.csv (get lat/lon and the Q-number from Wikidata).
3. Add a row to data/videos.csv referencing both.
4. python3 tools/validate.py

The validator checks for duplicate ids, dangling influencer_id / place references, reversed or non-numeric year spans, out-of-range coordinates, relative URLs, and unknown platform values. It exits non-zero on errors; warnings (a video with no place, an unused gazetteer entry) do not fail.

Optional SQLite view

python3 tools/build_sqlite.py     # → directory.db
sqlite3 -header -column directory.db \
  "SELECT title, creator, year_start, year_end, places FROM v_directory
   WHERE year_start < -1000;"

directory.db is a derived, disposable artifact — the CSVs remain the source of truth. The schema normalises places and tags into video_places and video_tags join tables and exposes a flattened v_directory view.

The timeline

The year axis is cube-root warped: a linear axis from the Palaeolithic to now would compress everything after 3000 BCE into the last few percent of the bar. The warp gives recorded history a third of the bar while still reaching −50 000, which is where the oldest indexed item (the Aurignacian) sits. The histogram behind the slider counts items whose span overlaps each bucket, and responds to every filter except the year range itself.

Provenance of the data

The 211 items are real videos, verified end to end, from 29 creators in five languages (English, French, German, Italian, Spanish):

• Each creator's YouTube handle was resolved to a canonical youtube_channel_id by reading the rel="canonical" link on the channel page. Three of the handles first guessed from memory were wrong and were corrected this way (@miniminuteman is a different person; Metatron's history channel is @Metatronrealhistory; @Premodernist does not exist, it is @premodernist_history).
• Titles, video ids and publication dates come from each channel's own Atom feed (youtube.com/feeds/videos.xml?channel_id=...), so nothing is transcribed by hand and no item can be attributed to the wrong creator.
• Every row was then re-checked against YouTube's oEmbed endpoint: 210 of 211 URLs resolve, each title matches the feed exactly, and each author_name matches the creator the row is attributed to. The one exception (v138, Time Team Classics) has embedding disabled, which makes oEmbed return 401; it was confirmed live on its watch page instead and its notes say so.
• platform is set from whether the id resolves under /shorts/ — 69 of the items are Shorts or reels, 142 long-form.

What verified=yes does and does not mean. It means the link, title, creator and publication date are confirmed. It does not mean someone watched the video to confirm the period and place: year_start, year_end, era and places are editorial readings of the subject, and where the reading is loose or a place stands in for a region, the notes column says so. Twenty items are deliberately left with no place — a survey video, or a subject with no single site — and those simply do not appear on the map.

Archaeosoup is listed as a creator but has no items: the channel resolves, but its feed returns no recent uploads.

Language coverage. 17 anglophone creators (140 items), 5 French (30), 3 German (17), 2 Italian (13), 2 Spanish (11). influencers.csv carries the language code and the UI filters on it. The non-English creators are indexed with their original titles, untranslated — the title is what you would search for on the platform, so translating it would make the row harder to find, not easier. The era and tags columns stay in English so they remain a shared index across languages.

Note that reach is uneven by design of the sources, not of the field: French archaeology is well served by Inrap, the national preventive-archaeology institute, whose films are site-by-site and so map unusually cleanly. A few Inrap entries are Version LSF — French sign-language versions of a site film — which is recorded in notes.

Platforms other than YouTube. A creator row can carry instagram, tiktok and facebook links, but only YouTube exposes a public per-channel feed. Meta and TikTok put individual reel permalinks behind a login wall, so for a reel-first creator the profile is recorded on the creator row while the indexed items are whatever is publicly enumerable — usually their YouTube Shorts. Facebook and TikTok reels can still be added by hand, one row at a time, with a permalink copied from the app.

Keeping it current

python3 tools/refresh_feeds.py          # list uploads not yet indexed
python3 tools/refresh_feeds.py --csv    # same, as skeleton rows to paste in

It only reports; it never edits the CSVs, because deciding the period and place of a new video is a judgement call. It also re-checks that each feed is still titled what influencers.csv expects, which is what caught the wrong channel ids in the first place.

Pending uploads

data/pending.csv is the one machine-written file here. Every Monday the sweep rewrites it with recent uploads that are not in videos.csv, commits it, and redeploys the site. Each creator's profile panel then shows them under Latest uploads.

Those rows carry no year_start, no year_end, no era and no places, and they never reach the map or the timeline. Deciding what period and place a video covers is the editorial judgement this directory exists to record, so the robot does not guess at it — it only says "this exists and nobody has looked at it yet".

Promoting one is the normal contributionroute: open an Add a video or reel issue, or send a pull request.

Automation

Four workflows, in .github/workflows/:

Workflow	Trigger	What it does
validate.yml	every push and PR	runs tools/validate.py and the SQLite build
pages.yml	push to main	validates, assembles _site/, deploys to GitHub Pages
submission.yml	issue labelled add-channel / add-item	runs tools/issue_to_row.py, opens a PR, or comments the error back on the issue
refresh.yml	Mondays 06:17 UTC	rewrites data/pending.csv, commits and republishes it, and files the deep backlog as one issue

The submission bot prepares, a human merges. It resolves the channel id, reads the real title and upload date, works out whether the item is a Short, and refuses anything it cannot verify — but the period and place it writes are the submitter's claim, and that is what a reviewer is there to check.

The issue body is untrusted input: it is passed to the script through an environment variable and a file, never interpolated into a shell command. The parser is stdlib-only, so there is no third-party action in the path that writes to the data.