A Wikipedia article, explained

The HTML video

13-wikipedia-waggle-dance--waggle-dance.html is the single-file export, playing right here. Click it, then press ? for the keys, or open it on its own.

English: final.mp4 · 1920x1080 · 30 fps · 60.00 s · 15.0 MB · -14.1 LUFS / -1.5 dBTP · web page waggle-dance.html (11.3 MB, one file, plays offline, 7 chapters) Français : final-fr.mp4 · 1920x1080 · 30 fps · 64.20 s · 15.1 MB · -14.1 LUFS / -1.6 dBTP · web page danse-fretillante.html (11.4 MB)

Licence: CC BY-SA 4.0, not MIT. This video, its script and its project sources are adapted from a Wikipedia article, so they carry the article's share-alike licence (LICENSE.txt). The repository's MIT licence covers showtime's code only. The bee footage inside stays CC BY 3.0 (Su et al. 2008).

The request#

"Turn the Wikipedia article on the waggle dance into a 60-second animated explainer, and make a French version too."

Mode: quick, publish-bound (a researcher pass before the build, a critic pass on the final, a voice-director pass on the French). No questions asked. The opening line stated the assumptions:

Quick mode: 60 s, 16:9, canvas film with a procedural score, voice af_heart; French with ff_siwis, re-timed by its own narration. Source: English Wikipedia "Waggle dance", revision 1371912261 (CC BY-SA 4.0), so the video is CC BY-SA 4.0 too. Opening on real footage of a dancing bee (Su et al. 2008, CC BY 3.0), framed, not full-bleed.

Contract: a forager tells her nestmates where food is by dancing: the angle of her waggle run on the vertical comb is the flower's angle from the sun, and the length of the run is the distance.

The story#

Scene times are not typed anywhere: voice cues turns voice/timeline.json into a VO table, cues.js reads every cut from VO.lines.<id>.start and every reveal from the start of the word that names it (vword()), and the two DOM layers take their windows from the same table. The French cut re-times itself by running voice script + voice cues on its own narration.

Tracking the mark. The ring in the hook is measured, not drawn: 116 stills of the clip (one per 0.1 s, showtime snap), the red paint mark found by colour in each (project/data/track_red_mark.py), 52 positions kept for the 5.4 s shown (data/dance-track.js). The positions are in the frame's own pixels, and the camera follows the dancer (a whole-frame shift of up to ~60 px between stills 0.1 s apart), so they are not her path on the comb. The page therefore draws a ring on the mark and only the last 0.7 s as a trail, labelled "Her red paint mark, tracked (the camera moves with her)". Where the mark was not found (blurred, mostly during the waggle run) the ring is interpolated and drawn dimmer. An earlier cut drew the whole series and called it "her path"; the independent critic caught that (see Review).

The source scene is text only. The live article shows two figures under other licences (Bee dance.svg, CC BY-SA 2.5; Waggle dance.png, CC BY 2.5) and the Wikipedia logo. site record cannot hide elements, so capture/make_text_only.py saves the live page (checking it is revision 1371912261), strips its scripts and hides figures, images and the logo; site record --serve then scrolls that copy. Nothing else on the page changes.

Commands, in order#

showtime is skills/showtime/bin/showtime; <job> is showtime-out/waggle-dance-20260927-133213, <p> is <job>/project; <fr> is showtime-out/waggle-dance-fr-20260927-140728.

showtime doctor --quick
showtime job init waggle-dance --mode quick --platform youtube --goal "Turn the Wikipedia article ..." \
  --assumed "Source: English Wikipedia 'Waggle dance' rev 1371912261 (2026-08-29), CC BY-SA 4.0; output CC BY-SA 4.0" \
  --assumed "60 s, 16:9, canvas film, voice af_heart, procedural Synth score; FR with ff_siwis" \
  --assumed "Real footage: Su et al. 2008 Movie S2 (CC BY 3.0), framed, not full-bleed"

# real material: the live article, the CC BY footage
showtime site capture https://en.wikipedia.org/wiki/Waggle_dance <job>/capture/wiki --no-assets --max-shots 6 --max-height 8000
showtime new film <p> --title "The waggle dance" --duration 60
showtime assets media search "East Learns from West Asiatic Honeybees dance" --source commons --type video --allow-attribution
showtime assets media fetch commons:20989587 --project <p> --allow-attribution      # sha1 matches the Commons file
showtime footage probe <p>/assets/media/20989587.ogv                                 # 11.58 s, 640x480 shown 4:3
showtime footage trim <p>/assets/media/20989587.ogv --webm --no-audio --crf 30 -o <p>/media/dance.webm
showtime snap <p>/media/dance.webm --every 0.5 --cols 6                              # look at the dance
showtime snap <p>/media/dance.webm --at 0.017,0.117,...,11.517 -o track              # 116 stills for the trace
python project/data/track_red_mark.py track                                          # -> data/dance-track.js
python capture/make_text_only.py <job>/capture/wiki-text                             # rev 1371912261, text only
showtime site capture --serve <job>/capture/wiki-text <job>/capture/wiki-local --width 1120 --no-assets --no-full
showtime site record --serve <job>/capture/wiki-text <job>/capture/wiki-scroll --width 1120 --duration 6 --hold 1.3 --to 0.14 --dpr 2
showtime footage trim <job>/capture/wiki-scroll/scroll.mp4 --webm --no-audio --width 1600 --crf 34 -o <p>/media/wiki-scroll.webm

# voice first: the narration sets every cut
showtime voice ipa "Karl von Frisch shared the 1973 Nobel Prize ..."        # "1973" read as "nineteen hundred ..."
showtime voice ipa "Karl von Frisch, ... Apis mellifera." --lang fr         # "von" read as French "vont"
#   -> project/lexicon.json (von, Frisch per language); narration.md spells [1973](nineteen seventy-three)
showtime voice script <p>/narration.md -o <p>/voice --fit 58     # 59.37 s natural -> x1.04: cut 4 words
showtime voice script <p>/narration.md -o <p>/voice --fit 58     # 58.65 -> 58.00 s (x1.01, pauses -0.28 s)
showtime voice cues <p>/voice/timeline.json -o <p>/voice/cues.js

# music alone, in seconds
showtime score <p>        # -17.1 LUFS: 3 dB under the voice -> master gain -9
showtime score <p>        # -28.7 LUFS, but distance/source 8 dB quieter -> bass added there
showtime score <p>        # -28.1 LUFS, sections -24..-34 dB, voice about 14 dB over the bed

# first look, fix, look again
showtime snap <p> --at 0.5,3,5,...,59 --sheet --cols 6          # 3 rounds of stills
showtime check <p>        # 11 warnings: 6 still holds, "dark" 0.8 s, system font in the browser chrome, ...
showtime check <p>        # PASS, 1 warning (a component's 15 s load timer; see Tool issues)
showtime render <p> --job <job> --preview                        # 720p draft, 1 m 15 s
showtime snap <job>/preview.mp4 --every 2 --cols 6

# final, verify
showtime render <p> --job <job>                                  # final.mp4 (77 MB master, static grain)
showtime captions <p>/voice/vo.words.json --style clean --aspect 16:9 -o <job>/captions.ass --srt <job>/final.srt
showtime qa <job>                                                # PASS
showtime review-pack <job>                                       # round 1 -> review/round-1/FINDINGS.md

# French: same project, own voice, own timing
showtime job init waggle-dance-fr --mode quick --platform youtube --goal "French version ..." --assumed "..."
#   copy the project, lang.js = "fr", narration.fr.md (same line ids), strings.js carries both languages
showtime voice script <fr>/project/narration.fr.md -o <fr>/project/voice         # 61.82 s
showtime transcribe <fr>/project/voice/vo.wav --language fr --no-events          # round trip: "colauréat" misheard
showtime voice script <fr>/project/narration.fr.md -o <fr>/project/voice         # rewritten line only: 62.20 s
showtime voice cues <fr>/project/voice/timeline.json -o <fr>/project/voice/cues.js   # the film re-times itself
showtime snap <fr>/project --at ... --sheet                                      # overflow check: 2 fixes
showtime check <fr>/project; showtime render <fr>/project --job <fr>
showtime captions <fr>/project/voice/vo.words.json --style clean --aspect 16:9 --max-words 7 -o <fr>/captions.ass --srt <fr>/final.srt
showtime qa <fr>          # WARN: 2 frozen stretches, 45-char lines -> fixed (below), re-rendered

# round-1 fixes (both languages), then one full-res render each
showtime render <fr>/project --from 29.3 --to 33 -o <fr>/work/cut-distance.mp4   # the frozen-stretch fix, 3.7 s
showtime render <p> --job <job>                                  # final-2.mp4
showtime render <fr>/project --job <fr>                          # final (after one more label fix)
showtime qa <job>                                                # PASS (0 warn)
showtime qa <fr>                                                 # WARN: 1 caption at 21 chars/s (kept, see QA)
showtime review-pack <job>                                       # round 2 -> review/round-2/FINDINGS.md
showtime snap <job>/final-2.mp4 --at 45.04,45.5,52.5,53.5,54.2 --compare <job>/final.mp4

# deliver
showtime deliver exports <job> --targets youtube                 # 49.3 MB (FR 52.5 MB), not committed
showtime deliver exports <job> --targets original --max-mb 8     # 7.6 MB each, not committed (footage gets soft)
showtime deliver exports <job> --targets original --max-mb 16    # this folder's final.mp4 / final-fr.mp4
showtime qa <job>/exports/final-2.16mb.mp4 --project <p> --platform youtube     # PASS
showtime deliver thumb <job>/final-2.mp4 --at 4.6 -o <job>/thumb-1280x720.jpg   # (and the French one)
showtime export html <p> --audio score -o <job>/work/score-only-test.html       # 10.3 MB, but no voice: not shipped
showtime export html <p> --audio embed --target artifact -o <job>/waggle-dance.html
showtime export html <fr>/project --audio embed --target artifact -o <fr>/danse-fretillante.html
showtime job note <job> --stage deliver --verified "..." --next "..."   # (and <fr>)
showtime clean <job> -y; showtime clean <fr> -y     # freed 333.5 MB and 93.0 MB

# after the independent critic (see Review): fixes in scenes.js, layers.js, index.html, strings.js, then
showtime snap <p> --at 0,2.5,4.6,22.5,25,44.9,45.04,45.2,45.35,52.5 --width 960 --format jpg -o <scratch>/en1
showtime snap <fr>/project --at 0,4.6,22,24,26,28,49.2,49.4,49.6,56 --width 960 --format jpg -o <scratch>/fr1
showtime check <p>; showtime check <fr>/project                  # PASS, same 1 warning as before
showtime render <p> --job <job>                                  # final-3.mp4, 1 m 25 s
showtime render <fr>/project --job <fr>                          # final-2.mp4
showtime qa <job>                                                # PASS (0 warn)
showtime qa <fr>                                                 # WARN: the same 21 chars/s cue, in .ass and .srt
showtime snap <job>/final-3.mp4 --at 0,4.6,45.04,45.2,52.5 --compare <job>/final-2.mp4   # review/after-critic/
showtime snap <fr>/final-2.mp4 --at 4.6,28,49.4,49.6 --compare <fr>/final.mp4
showtime review-pack <fr>                                        # the French cut's own pack -> review/fr-round-1/
showtime deliver exports <job>/final-3.mp4 --targets original --max-mb 16    # final.mp4 (15.0 MB)
showtime deliver exports <fr>/final-2.mp4 --targets original --max-mb 16     # final-fr.mp4 (15.1 MB)
showtime qa <job>/exports/final-3.16mb.mp4 --project <p> --platform youtube  # PASS (and the French one: PASS)
showtime deliver thumb <job>/final-3.mp4 --at 4.6 -o <job>/thumb-1280x720.jpg   # (and the French one)
showtime export html <p> --audio embed --target artifact -o <job>/waggle-dance.html          # 11.3 MB
showtime export html <fr>/project --audio embed --target artifact -o <fr>/danse-fretillante.html  # 11.4 MB

Features shown#

export html --audio score (the brief's "tiny, live score" page) was made and measured, then not shipped: it plays only the Synth score, so the narration is gone (the exporter warns), and it is not tiny here anyway (10.3 MB): the two footage layers are most of the file. Both shipped pages embed the mixed soundtrack (AAC 96k, ~0.75 MB). They are --target artifact single files under the 16 MB artifact limit; nothing was uploaded from this session.

About the crew. This session had no sub-agent tool, so the director did each crew task from its brief in references/crew/ and wrote the same files a crew member would: crew/research.md (every claim against the pinned revision), crew/voice-director-fr.md, and the critic's review/round-*/FINDINGS.md, labelled as self-reviews. A second look by a person, and a native French speaker's read of project-fr/narration.fr.md, are the remaining steps.

Review (self-reviews)#

Round 1 (on final.mp4): ship after fixes. What worked: the measured trace on real footage as the hook; the same wedge in the field and on the comb (45° and 45°, then 20° and 20°, same side, checked with snap --at); the full attribution on a 6 s end card.

Round 2 (on final-2.mp4, fixes only): ship. Each fix confirmed on stills next to the old render (review/round-2/compare.jpg).

Review round note: independent critic on round 2 (a separate critic pass, not a self-review). Verdict: ship after fixes. It kept the facts, licences, loudness and three of the four round-1 fixes, and found what the two self-reviews missed. Everything below is fixed in the shipped files (English final-3, French final-2 in the jobs); before/after stills in review/after-critic/, each checked with snap --at at the cited time.

The critic's notes arrived cut off after a polish item at 25.00 s, so that last item is not known and was not acted on. Per the two-round protocol this did not start a third critic round: a person still needs to take a second look.

QA summary#

• English (final.mp4 = the job's final-3 capped at 16 MB): PASS (0 fail, 0 warn, 0 note), --platform youtube. Master final-3.mp4 (77.5 MB): PASS, -14.0 LUFS / -1.6 dBTP; voice about 14.4 dB over the score. h264 High, yuv420p, 1920x1080, 30 fps, faststart, BT.709; 60.00 s = showtime.json; no silent gaps, no black or frozen stretches; frame 0 has picture and flows into frame 1; captions: 23 cues, readable; credits: the one CC BY item is listed.
• French (final-fr.mp4 = the job's final-2 capped at 16 MB): the export alone is PASS (0 warn); the master final-2.mp4 (81.4 MB) with its captions is WARN (0 fail, 2 warn), -14.0 LUFS / -1.6 dBTP, 64.20 s. Both warnings are the same caption ("Plus la nourriture est loin,", 36.5 s), counted once in captions.ass and once in final.srt: it needs 21 characters/s, limit 20. Kept: the cue follows the voice's pace, and joining it to the next cue would pass 42 characters per line.
• showtime check: PASS, 1 warning in both (a 15 s safety timer inside the components' media loader, counted as a playback timer; the frames are unaffected) and 7 notes (browser-chrome decor text; labels near the edge during the push-ins).

Tool issues found (reported, worked around here)#

• site record / site capture cannot hide elements or inject CSS, so a page with images under other licences needs a saved, patched copy (capture/make_text_only.py) and --serve.
• assets media fetch copies the Commons "Author" field, which names the paper's editor as an author here; credits were corrected by hand.
• export html --audio score drops the voice mix of a narrated film (warned, by design); assets credits --all lists no fonts, voice or score, so credits.txt was written by hand.
• check/render warn about a setTimeout in components/core.js (mediaReady's 15 s fallback), which fires during playback in any page with a component video.
• review-pack builds its scene list from the DOM clips (2 layers) instead of the film's acts.
• captions --style clean writes 45-character French lines that qa then flags (limit 42).

Timings#

Shared 6-core Intel i5, other example builds running (load 14-23).

Files#

final.mp4 / final-fr.mp4        the videos (English qa PASS, French qa WARN: one 21 chars/s caption)
poster.jpg / poster-fr.jpg      the hook at 4.6 s (the tracked red mark on real footage)
thumb-1280x720.jpg (+ -fr)      YouTube thumbnails (deliver thumb)
captions.srt / captions-fr.srt  sidecar captions
waggle-dance.html               English web page (export html --target artifact, plays offline)
danse-fretillante.html          French web page
share.txt                       YouTube copy, EN + FR, with alt text and the licence lines
credits.txt / credits-fr.txt    article, footage, voice, fonts, music, output licence
LICENSE.txt                     CC BY-SA 4.0 for this folder (footage CC BY 3.0)
project/                        the English project: index.html, scenes.js, cues.js, strings.js (EN + FR), layers.js,
                                score.js, narration.md, lexicon.json, audio/mix.json, data/ (mark track + script),
                                media/ (footage and scroll as VP9), voice/ (timings, no WAVs)
project-fr/                     what differs in French: lang.js, showtime.json, narration.fr.md, voice/ timings
capture/make_text_only.py       the text-only article copy for the scroll
crew/                           researcher and voice-director notes (done inline)
review/                         round 1 and 2 findings (self-reviews), round-2 compare sheet, after-critic/
                                before/after sheets, fr-round-1/ (French pack sheet + self-review)

To rebuild English: showtime voice script project/narration.md -o project/voice --fit 58, showtime voice cues project/voice/timeline.json -o project/voice/cues.js, then showtime render project. French: copy project/, overlay project-fr/, run showtime voice script narration.fr.md -o voice and voice cues in it, then render.

Sources and licences#

• Text: "Waggle dance", Wikipedia contributors, English Wikipedia, revision 1371912261 (2026-08-29), https://en.wikipedia.org/w/index.php?title=Waggle_dance&oldid=1371912261, CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/); summarised, rewritten as narration, animated and (French) translated. The French article (https://fr.wikipedia.org/wiki/Danse_des_abeilles) was used for terminology only.
• Footage: Movie S2 of Su S, Cai F, Si A, Zhang S, Tautz J, Chen S (2008), "East Learns from West: Asiatic Honeybees Can Understand Dance Language of European Honeybees", PLOS ONE 3(6): e2365, https://doi.org/10.1371/journal.pone.0002365, via Wikimedia Commons, CC BY 3.0; trimmed, cropped, overlay drawn.
• Nobel fact: https://www.nobelprize.org/prizes/medicine/1973/summary/.
• Not used: the article's figures (Bee dance.svg, CC BY-SA 2.5; Waggle dance.png, CC BY 2.5), hidden in the scroll; the Wikipedia logo, hidden. The 45° diagram is drawn from the caption's facts, not traced.
• Voices: Kokoro-82M (Apache-2.0), af_heart and ff_siwis (trained on SIWIS, CC BY 4.0), synthesized locally.
• Fonts: Fraunces, Inter (SIL OFL 1.1), served locally by showtime.
• Music: composed in code with showtime's Synth (no samples, no library tracks).
• Output: CC BY-SA 4.0 (LICENSE.txt).

The project files on GitHub