| Question, direction, approval | The channel's creator (one person) | Chose the topic, a cat-shaped slice through the universe (2026-10-09); watched the full cuts and gave notes; asked for the guesses and the actual numbers to be shown side by side, and for the slices to be shown in 3D (2026-10-16); chose the ending (2026-10-16) and the music level; proofread the script line by line and reworded the line at 5:03.7 (2026-10-18). On 2026-10-18: approved the full cut v9 and its music, asking that the universe in the Chief's belly stay still while the Chief breathes in its sleep; chose the new poses and the Japanese and English titles; listened to the Japanese voice track, asked for a calmer Sescha and chose her settings, marked ten lines whose delivery or reading was off, chose among the retakes and approved the Japanese voices; asked for the seven other languages to use the same voice settings. watched the final English master with subtitles and checked the pronunciation; flagged one long silence after a line, which was fixed in six languages; approved all eight language tracks for upload (2026-10-19) | — |
| Research and fact ledger | AI agents | Found and read the sources and recorded in the fact ledger how much of each was read (full text, abstract, secondary). Computed the estimates in Method M1–M8. | Claude (Claude Code) |
| Script (Japanese master) | AI agents, with the creator's notes | Written and revised by AI agents over many drafts, tested with LLM readers given no context, then revised again. The creator's notes on the script were applied by one AI author agent, keeping their intent. | Claude |
| Fact checks | AI agents | Fact-check passes compared the script and the cards on screen with the fact ledger. This page was compared with the script, the fact ledger and frames of the v10 render. | Claude |
| Translations (English, French, Italian, German, Spanish, Chinese in simplified and traditional characters, Korean) | AI agents | Each language was translated scene by scene from the Japanese master by an AI agent, following a written style guide, notes on how each character speaks, a glossary of 147 terms, and the context and time limit of every line. In the first review round, two more AI agents checked each translation: one read it straight through as a native speaker would, without seeing the Japanese; the other compared it with the Japanese line by line. A second round checked the revised lines. When the Japanese voices were final, the lines whose wording or tone had changed were re-checked in all seven languages, and the revised lines were read again by a separate AI agent without context (2026-10-18). The translations have not been checked by human native speakers. If you speak one of these languages and a line sounds wrong or unnatural, please tell us in the comments under the video. | Claude (Claude Code) |
| Character art | AI image generation, edited by script; selected by the creator | As in Episode 1: poses were generated as sheets, cut out, and fixed with deterministic pixel-editing scripts. The jumping and back-view poses added for this episode were made the same way and chosen by the creator (2026-10-18). | Astra (gpt-6-astra), run through codex exec |
| Boards, room, animation, the 3D universe | Code written by AI agents | The paper-style boards, the room and all motion are drawn by code (JavaScript, Canvas 2D, three.js), recorded in headless Chrome and encoded with ffmpeg. No image generation was used for the boards, the room or the universe; the Milky Way and the other galaxies close up are drawn on a NASA/JPL-Caltech illustration, and the Earth uses NASA imagery (Data D1, D2). | Claude; headless Chrome; ffmpeg |
| Data processing | Code written by AI agents | Method M1–M8; the computed web of galaxies (Data D3). | Python (standard library); Node.js |
| Music | Score by an AI agent; sample libraries; rendered and mixed by script | The score was written by an AI agent and rendered without a graphical interface by our command-line music pipeline (final render 2026-10-18), using sampled instruments from the GarageBand and Logic sound libraries (strings, flute, clarinet, piano, French horn, bassoon, harp, and a hall reverb), Sonatina Symphonic Orchestra v4.0 (vibraphone, second violins, marimba; CC Sampling Plus 1.0; credited in the description) and VSCO 2 Community Edition (violin ensemble, timpani, percussion; CC0). The Apple sound libraries may be used in our own music and video, but the sounds may not be redistributed on their own. The music sits under the voices at one constant level, with no ducking, and stops for the two guessing times and one pause to think. | Claude; bgmctl 1.1.0 |
| Voices, Japanese track | Synthetic | The same two character voices as in Episode 1, designed from text prompts with ElevenLabs voice design. The lines were generated with Eleven v4, two takes per line, and every take was checked for misread words with ElevenLabs speech-to-text. After listening, the creator asked for a calmer Sescha: her lines were generated again with steadier settings and a calm delivery direction. The creator then marked ten lines whose delivery or reading was off; for those, we generated alternatives with other delivery directions and punctuation, and the creator chose among them by listening (2026-10-18). | ElevenLabs |
| Voices, English and the six other dubbed tracks | Synthetic | English is the video's original audio track; French, Italian, German, Spanish, Chinese and Korean, like the Japanese track, are extra audio tracks on the same video. Each of these languages uses its own versions of the two character voices, made from the Japanese-track voices with ElevenLabs Voice Remix (changing only the language and accent), as in Episode 1. The lines were generated with Eleven v4, with the same settings and delivery directions as the Japanese track (the creator's choice, 2026-10-18). For French, Italian, German, Spanish, Chinese and Korean, one take was generated per line, and a line was generated again only where ElevenLabs speech-to-text showed a misread word, a doubled reading or a direction read aloud. For English, Sescha's lines and seven of Comet's were generated this way; Comet's other English lines come from an earlier recording of two takes per line, chosen with speech-to-text. Code placed each language's lines into the Japanese timing slots: a line may run into the following picture-only pause, or be time-stretched by at most 5% without changing pitch. Lines that still did not fit, or left a much longer silence than the Japanese, were shortened or completed in translation and recorded again. Every track uses the same music and sound effects as the Japanese track, at the same levels. | ElevenLabs |
| Subtitles | Code | Nine subtitle tracks (Japanese, English, French, Italian, German, Spanish, Chinese simplified, Chinese traditional, Korean), generated by code from each language's spoken lines. The Japanese subtitles follow the Japanese voices, with their punctuation matched to the takes the creator chose. Subtitles are separate tracks and are not burned into the picture. The eight translated tracks are timed to each language's placed voices. | — |
| Text on screen | Code | The few words and numbers on screen are drawn by code in typefaces that come with macOS: Avenir Next (the distance labels, the counters, the cards and their words GUESS and ACTUAL) and Futura (the series name on the title card, the guessing cards, and the labels "1 mm" and "29 km"). The macOS license (macOS Tahoe 26, section 2E) lets users use the included fonts to display and print content while running macOS; the text is drawn into the video frames on our Mac, and no font files are embedded or shared (license text checked 2026-10-08 for Episode 1). | — |
| Sound effects | Stock sound library; synthesized | Paper, pin, clock, brass and stamp sounds from 効果音ラボ (soundeffect-lab.info) and 効果音辞典 (sounddictionary.info), from the set used in Episode 1: commercial YouTube use allowed, credit optional, no redistribution (terms checked 2026-10-03); trimmed and mixed by script. The dome's opening and closing sounds were assembled by script from the same set. The slicer's hum and the Chief's sleeping breath are synthesized. | — |