Skip to content

About

48s code-rendered explainer film: the GitHub follow-vetting bot that followed 15 humans and rejected torvalds. Pure draw(frame), synthesized groove score, deterministic render pipeline.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

followbot-explainer

A 48-second motion-graphics explainer: "The bot that followed 15 humans on GitHub today (and rejected torvalds)" — the true story of an automated follow-vetting run: 214 candidates vetted, 28 passed the gates (>=30% follow-back, active <=92 days, human eyeballs), 15 followed at 2.2s intervals, 959 API calls at 37.5 req/s sustained, zero throttles. Torvalds fails the mutual gate. Both decisions were correct.

Rendered entirely from code: ink-on-graph-paper world, GitHub-dark palette, boil-4 line wobble, a synthesized electro/blues groove score locked to the same timeline as the picture (cuts on bar downbeats, SFX on action beats).

storyboard

Files

File What it is
film.html The whole film: draw(frame) pure function, CUES/SECTIONS/EVENTS data-driven timeline, window.SCORE (WebAudio groove builder), camera rig, claim()-based layout registry, #charsheet debug view
render_video.mjs Deterministic frame dump (playwright-core + cached chromium) -> ffmpeg x264 crf18; resumable (skips existing frames)
render_audio.mjs Offline-renders window.SCORE through an OfflineAudioContext -> 48kHz stereo WAV
render_still.mjs Single-frame stills at any frame numbers (for storyboard/QC passes)
qc_check.mjs Layout QC: sweeps every 3rd frame, flags straddling claims (clipped-at-edge = defect) and text overlaps; handoff windows excluded
render_char.mjs Character-sheet renderer (window.__CH = 1)
followbot2.py The production bot (v2): warm-orbit / followback-bio / neighbor sourcing, human+active vetting gates, climb→ratio governor, crash-safe ledger. Run: GH_TOKEN=<pat> python3 followbot2.py --budget 900
BOT2.md How v2 works: the three source pools, the five vetting gates, the 2.2× climb multiplier, the 0.93 ratio governor, budget-slice running — plus honest field notes on what automation can and cannot convert
followbot2.zig The production bot in Zig (0.14, stdlib-only, static binary). Dry-run by default; --live enables writes; same sources/gates/governor/pacing as the python bot. See ZIG.md

Render it

npm i playwright-core@1.57.0        # matches the sandbox's cached chromium-1200
node render_video.mjs               # frames/ + film-noaudio.mp4
node render_audio.mjs               # score.wav
ffmpeg -y -i film-noaudio.mp4 -i score.wav -c:v copy -c:a aac -b:a 192k \
  -shortest -movflags +faststart followbot-explainer.mp4

QC contract

Every visible element claims its screen-space AABB (claim()); the checker treats edge-straddling claims as defects and fully-off-screen pre-rolls as intentional (shot handoffs pre-roll 0.4s under the previous camera). The character sheet (node render_char.mjs film.html char.jpg) audits poses/eyes/mouth/prop. Sync is structural: picture and score read the same BPM/SECTIONS/EVENTS, so they cannot drift.

Facts depicted (from the actual run)

  • 2 followers, 0 following on day 1 -> follow-vetting bot with mutual/active/human gates
  • Candidate pool: followers, stargazers, neighbors, fork-owners seeds -> 214 vetted
  • 28 passed all gates; 15 followed, each 2.2s apart (politeness was the limiter)
  • HTTP 204 x 15, 959 total API calls, 37.5 req/s sustained, 4.5 calls/candidate, 0 throttles
  • Budget: 5000 requests/hour — GitHub never even noticed

About

48s code-rendered explainer film: the GitHub follow-vetting bot that followed 15 humans and rejected torvalds. Pure draw(frame), synthesized groove score, deterministic render pipeline.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages