Skip to content
BranchesTags

Repository files navigation

robocon-archive

Every RoboCon website since 2019, served from archive.robocon.io/<year>.

There is no build step and no framework. The landing page is one hand-written index.html; the years are the original builds, copied in as they were. That is deliberate: anything with a toolchain rots, and this repository has to be deployable in ten years by copying files onto a host.

index.html          the landing page
serve.py            local dev server (mirrors vercel.json)
scripts/            prefix-paths.py, for adding a year
assets/             fonts + favicon for the landing page only
thumbs/<year>.webp  1000x600 screenshot of each year's front page
<year>/             the original build, essentially untouched
vercel.json         SPA rewrites + noindex for the archived years

Where the years came from

Each folder was exported from a branch of robotframework/robocon, except 2018, which predates that repository and was recovered from the Wayback Machine:

Year Branch Exported from
2018 none Wayback Machine, snapshot of 2018-02-18 — see below
2019 archive-2019 index.html + static/
2020 archive-2020 index.html + static/
2021 archive-2021 index.html + static/
2022 archive-2022 index.html, 404.html + dist/
2023 archive-2023 index.html, 404.html + dist/
2024 archive-2024 index.html, 404.html + dist/
2025 archive-2025 docs/ (the Vite build output)
2026 archive/robocon-2026 in samipe/rf-certs-page rebuilt, see below 
git -C ../robocon archive archive-2023 index.html 404.html dist | tar -x -C 2023

What was changed, and why

The builds were made for a site at the root of robocon.io. Two things break when the same files are served from /2023/ instead, and each needed one kind of edit. Beyond that, only two deliberate departures from the original exist: the Wayback cleanup in 2018, and the removal of Google Analytics from 2018–2021 (both below). Everything else is byte-identical — 2019, 2020 and 2021 have two changed files each plus that one script block, 2022 has two, 2023 and 2024 four.

1. Root-absolute asset paths

2019–2021 referenced /static/…, and 2025 referenced /assets/…, /img/…, /fonts/…. Under a year prefix those resolve to the domain root and 404.

They were rewritten to the full year-prefixed path (/2021/static/…), not to relative paths. A relative path would resolve against whatever route the SPA happens to be showing, so /2025/program would look for /2025/program/assets/…. An absolute path with the prefix cannot drift.

2019–2021 additionally carry webpack's runtime publicPath, which was "/" and is now "/<year>/". Only 2021 actually depends on it (one background video is assembled at runtime), but all three were changed so they behave alike.

2022–2024 already used relative paths (dist/…) and needed none of this.

2. The SPA router base

From 2022 on, every build is a Vue SPA whose router is created with createWebHistory() — no argument. At /2023/ such a router sees an unknown path and renders its own "404 – Not found" page, which is exactly what happened before this was fixed.

Vue Router's normalizeBase() falls back to the <base href> element when no base is passed, so a single line in the <head> sets it:

<base href="/2023/">

That is the whole fix for 2022–2025. No minified bundle was patched to change routing.

3. 2026 was rebuilt instead

RoboCon 2026 is a Nuxt site whose repository still builds, so it was generated for its own address rather than patched:

NUXT_APP_BASE_URL=/2026/ npm run build:archive

Nuxt then writes /2026/_nuxt/…, /2026/…/_payload.json and its route base itself. Two things it does not touch, and they were fixed here afterwards:

python3 scripts/prefix-paths.py 2026 --only favicon,archive --write
  • the favicon links written literally in nuxt.config.ts, and the asset URLs stored in content (/archive/assets/…). Served unfixed at /2026/, the Helsinki page loads 112 broken images; fixed, zero.

The navbar and the logo are plain <a> elements rather than link components, so a click is a real browser navigation and the router never joins the base path for them. They used to need a second pass (--links); the source repository now joins runtimeConfig.app.baseURL itself, so the build arrives with /2026/, /2026/online and /2026/helsinki already in place. --links remains in the script for a build that predates that fix.

Nothing under _nuxt/ is modified. Prefixing buildAssetsDir:"/_nuxt/" there is the same trap once more: Nuxt joins it with baseURL and the site then requests /2026/2026/_nuxt/… and loads nothing at all.

One cosmetic leftover: /2026/helsinki and /2026/online log "Hydration completed but contains mismatches". Vue reconciles the DOM, and content, images, links and URLs were all verified correct; the front page logs nothing.

4. 2018 was recovered from the Wayback Machine

There is no branch for 2018 — the site predates the robocon repository. The page was saved from a Wayback snapshot of 2018-02-18, which means the file arrived carrying the archive's own scaffolding rather than only the original site. Removed, in this order:

  • the injected toolbar, the playback engine (bundle-playback.js, wombat.js, ruffle.js, the __wm.* calls) and the "FILE ARCHIVED ON" note. The HTML went from 43 KB to 24 KB.
  • every web.archive.org/web/<timestamp>/ prefix, so the original URLs stand on their own again. They come in two shapes — with protocol and protocol-relative (//web.archive.org/…), the latter under a different timestamp — and both occur inside style.css and magnific.min.js as well as the HTML.
  • two counters that would otherwise have fired on every visit: athena.js and analytics.js, both archive.org telemetry rather than the site's own.
  • the WordPress emoji block, which pulled wp-emoji-release.min.js from the live domain and got a 403 there. That file is not in the archive and could not be recovered.

Three things had to be fetched, because they were referenced but never saved: 13 favicons, manifest.json, and the OCR font that style.css asks for. The video came in as a saved copy of the YouTube player; the iframe now points at the real playlist instead, so it plays again.

5. Google Analytics removed from 2018–2021

Those four years shipped the classic analytics.js snippet (UA-106835747-1, 2019 uses -2). It was taken out of each index.html. Universal Analytics was shut down in 2023, so the property is dead and the request pointless — and an archive that quietly reports its visitors to a third party is not what anyone is looking for here. 2022 onwards never had it.

Two 404s that were already there

2019–2021 request a site.webmanifest that was never committed to the RoboCon repository, and 2021 requests an LDR_LLL1_0.png that is likewise absent. Both were broken on the live site too, and are left alone — the archive shows what was actually shipped.

The same goes for MäRZ on the 2021 front page. That mojibake is in the original bundle (verified byte-identical), not something the migration caused.

Running it locally

python3 serve.py          # http://127.0.0.1:8099
python3 serve.py 3001     # another port

serve.py mirrors the two rules in vercel.json and nothing else, so what you see locally is what Vercel serves. python3 -m http.server works too, but it gets one thing wrong: from 2022 on each year is a single-page app that owns paths like /2023/sponsor, and opening one directly -- or just reloading after clicking through -- 404s there while it works in production.

Verified against serve.py: /2023/sponsor, /2025/program and /2026/helsinki render their real routes on a direct hit, assets still come from disk, the landing page is indexable and every year answers X-Robots-Tag: noindex.

The screenshots

Taken once at a 1000×600 viewport against a local server, then converted to WebP. They are committed, not generated at deploy time, and there is no <iframe> anywhere on the landing page — ten live conference sites on one overview would be tens of megabytes, and any of them breaking would leave a blank tile no one notices.

To retake them:

python3 serve.py                     # from the repository root
# then screenshot http://127.0.0.1:8099/<year>/ at 1000x600
cwebp -q 80 2023.png -o thumbs/2023.webp

Deployment

Vercel, static, no framework preset. vercel.json does two things:

  • rewrites every unmatched /<year>/<path> to that year's index.html, so deep links like /2023/sponsor survive a reload. Vercel checks the filesystem first, so real assets are unaffected.
  • sends X-Robots-Tag: noindex for everything under a year. The archived builds still carry canonical and og: tags claiming to be robocon.io; seven copies making that claim would compete with the live site in search. The landing page itself stays indexable.

Domain: archive.robocon.io (CNAME to Vercel).

Adding a year

Two situations, and which one you are in decides almost everything.

The build can still be rebuilt (a current Nuxt/Vite site, e.g. RoboCon 2026) — then let the generator produce the year prefix and do almost nothing by hand. Case A.

The build is frozen (old Node, no working toolchain — 2019 to 2025) — then the files get fixed where they lie. Case B.

Either way, the pieces vercel.json and serve.py need no change: their rule matches 20[0-9]{2}, so a new year is covered the moment the folder exists.

Case A — the site can be rebuilt

RoboCon 2026 lives in a Nuxt repository, and Nuxt reads its base path from runtimeConfig.app.baseURL — it strips exactly that prefix off window.location.pathname before routing. That is why the <base href> trick used for 2022–2025 cannot work here, and why rebuilding is not just the tidier option but the only clean one.

1. Build with the year as the base URL.

cd ~/Source/robotframework/samipe/rf-certs-page
# stop the dev server first -- build and dev share .nuxt/.data
NUXT_APP_BASE_URL=/2026/ npm run build:archive

2. Check that the prefix actually landed, before doing anything else:

grep -o 'href="/[^"]*"' .output/public/index.html | sort -u | head

Every /_nuxt/…, /_payload.json and internal link must now start with /2026/. If they do not, the environment variable did not take: set it explicitly in nuxt.config.ts instead and rebuild.

app: { baseURL: '/2026/' }

3. Copy the output in.

cd ~/Source/robotframework/robocon
rm -rf 2026 && mkdir 2026
cp -R ~/Source/robotframework/samipe/rf-certs-page/.output/public/ 2026/

4. Fix what the generator does not prefix. baseURL only governs paths the framework builds itself. Literal strings do not move: the favicon links written out in nuxt.config.ts (/favicon/…) and asset URLs stored in content (/archive/assets/…) both stay pointing at the domain root.

--only is not optional here. A site rebuilt with its own base URL also stores the pieces it joins at runtime, and prefixing those (_nuxt, the routes in the payload) yields /2026/2026/… — the page then loads nothing at all. Name only what no router and no config will touch:

python3 scripts/prefix-paths.py 2026 --only favicon,archive   # dry run
python3 scripts/prefix-paths.py 2026 --only favicon,archive --write

Then continue at "Checking a new year" below.

Case B — the build is frozen

1. Export it from its branch in robotframework/robocon:

mkdir 2026
git -C ../robocon archive archive-2026 index.html 404.html dist | tar -x -C 2026

Look at the branch first — where the deployed build lives has moved around over the years (static/, then dist/, then docs/; see the table above).

2. Prefix the root-absolute paths.

python3 scripts/prefix-paths.py 2026 --write

3. If it is a single-page app, give it a base. Open /2026/ (step 1 under "Checking" below); if the site renders its own "404 – Not found" instead of the front page, its router does not know about the prefix. For a Vue Router built with createWebHistory() and no argument, one line in the <head> of index.html and 404.html fixes it:

<base href="/2026/">

That works because Vue Router falls back to the <base href> element when it is given no base. A Nuxt build does not — see Case A.

4. Webpack builds may also carry a runtime publicPath. If assets are assembled in JavaScript rather than written into the HTML (2021 does this for one background video), find .p="/" in the runtime chunk and make it .p="/2026/".

Checking a new year

python3 serve.py

Open http://127.0.0.1:8099/2026/ with the browser console visible, and check four things:

  1. The front page renders — not the site's own 404. If it does, the router base is wrong (Case A step 2, or Case B step 3).
  2. No failed requests. A 404 on /img/… or /assets/… without the year prefix means a literal path the rewrite did not catch.
  3. The fonts are the conference's own, not a fallback — those come from CSS url() and are the usual thing to slip through.
  4. Click an internal link, then reload the page. The URL must stay under /2026/ and survive the reload; that is what serve.py reproduces from vercel.json and what plain http.server gets wrong.

404s that were already broken in the original build are worth keeping — the archive shows what was actually shipped. Note them here as with 2019–2021.

Putting it on the landing page

Screenshot at 1000x600, matching the others:

# with serve.py running, capture http://127.0.0.1:8099/2026/ at 1000x600
cwebp -q 80 2026.png -o thumbs/2026.webp && rm 2026.png

Add a card at the top of the <ul class="years"> in index.html, copying the newest entry and changing year, dates and location. Take those from the archived page itself rather than from memory.

<li>
  <a class="card" href="/2026/">
    <img class="shot" src="/thumbs/2026.webp" width="1000" height="600"
         alt="The RoboCon 2026 website">
    <span class="meta">
      <span class="year">2026</span>
      <span class="when">... Feb &middot; Helsinki<br>... Mar &middot; online</span>
    </span>
  </a>
</li>

The second-newest card carries loading="lazy" from now on; the top two do not, so move that attribute down one card.

Finally, note the new year in the table under "Where the years came from", and record anything you had to change to it under "What was changed, and why".

About

robocon-archive.vercel.app

Resources

Readme
Activity
Custom properties

Stars

10 stars

Watchers

9 watching

Forks

5 forks
Report repository

Releases

Packages

Used by

Contributors

Languages