Skip to content

PRD-005 — Video Editor

Status · Draft v0.3 · April 2026 (supersedes v0.2 — reshaped to match PRD template) Sources · OC-02 Surface 4 · OC-04 Video editor · OC-05 Video editor Audiences · climate-communicator, artist (PA audiences) Promises · accumulation, citation-travels, editorial-dignity (PA promises) Principles · aesthetic-traces-to-data, attribution-baked-in (PA principles) Why this is a PRD · A timelapse turns daily-render accumulation into something a person can actually show someone. Whether the result feels like data brought to life — or like a slideshow with music — is a product judgment, not an export-format decision.

You scrub the timeline back to August 2022. The thermal field on your North Atlantic Drift recipe is the deepest coral the palette can render. The audio panel says Record high · fr. 348. You toggle the four default overlays — they are already on. You hit export MP4. Thirty seconds later the file is on disk. You watch it back: the music swells exactly when +1.7° hits the screen, and a record flash pulses in the same beat. The year of the ocean compresses into thirty seconds, and the moment the data peaks, the film knows.

The problem

A recipe accumulates a frame a day. Three months in, you want to show someone what North Atlantic Drift has been doing. The fourteen-day strip in the gallery is too short. Scrolling through ninety individual renders is not a viewing experience. To make a film of the year, you would have to assemble it elsewhere — Premiere, ffmpeg, hand-built — and along the way you would lose the data attribution, the per-frame editorial framing, and the synchrony between what the data did and what the film says.

There is also a deeper version of the problem. When a year of ocean data is set to ambient music and titled with editorial overlays, the result feels like climate communication. But ambient music is decoration unless it is structurally connected to the data. A timelapse with a stock soundtrack is a slideshow. A timelapse with music that swells at the precise frame the anomaly peaks is a piece of work. The difference is invisible from the outside and total from the inside.

The experience

You open the video editor for North Atlantic Drift — the gallery's timelapse ↗ action drops you straight in. The preview takes the upper-left of the screen with the recipe's frames already animating at twelve frames per second. On the right, three sections: Sequence, Audio, Overlays.

You glance at Audio. The system has already detected three key moments in the year — El Niño peak fr. 54, Record high fr. 348, Acceleration fr. 378. Each is colour-coded. You pick a theme — Ocean. You set the energy arc — builds throughout. You leave sensitivity on reactive. You do not pick chord progressions, instruments, or effects. You pick mood and let the system score the rest.

You glance at Overlays. Eight are on by default — primary counter, anomaly indicator, named events, record flash, sparkline, timeline ribbon, date stamp, source attribution. Two are off — pull quote, projection ghost. The defaults give every frame a complete data identity; the off-by-default ones are deliberate creative choices. You leave them off.

You click export MP4. Thirty seconds. The file lands on disk: a year of ocean, scored to music, annotated with editorial dignity, ready to post anywhere. You watch it back. At frame 348, the music intensifies, the screen says +1.7°, the record flash pulses. The data and the feeling arrive at the same instant. They were never going to drift, because they came from the same signal.

Why now

Recipes accumulate from the day they are created. A recipe authored in March 2026 has a year of frames by March 2027 — the timelapse becomes interesting around the three-month mark, useful around the six-month mark, and the project's most shareable artefact by the twelve-month mark. Without the video editor, that accumulation has nowhere to go. The gallery shows the work; the timelapse moves it.

The video editor is also the surface that does the most to take OceanCanvas's work out of OceanCanvas. A render lives at a URL; a timelapse is an MP4 that lives anywhere a video can play. For climate communicators in particular, the export is the format that gets used. Without it, the project's reach stops at the gallery URL.

Success looks like

  • A viewer of the export feels the music swell at the same instant a record value appears on screen — and cannot articulate why the moment landed.
  • A climate communicator uses an exported timelapse in a deck or a post without modification. Citation, attribution, anomaly framing all already there.
  • An artist watching their own year-long timelapse describes it as the piece rather than the export.
  • The audio reads as part of the data, not part of the production. If a viewer thinks "nice soundtrack" but not "the music is the data," the surface has missed.

Out of scope

  • Recipe authoring. PRD-003 (Recipe Editor) holds that. By the time someone reaches the video editor, the visual story is already told.
  • Frame-by-frame video editing. No multi-track timeline, no clip splicing, no levels mixing. The video editor is enrichment and assembly, not editing.
  • Music composition. No chord progressions, no instrument selection, no effect chains. The user picks editorial controls (theme, energy arc, sensitivity); generation is delegated to RFC-006's audio system.
  • Hosting the export. The MP4 lands on disk; sharing happens elsewhere. (A landing page per timelapse is a Phase 2 question.)
  • Export formats beyond MP4. GIF, WebM, 4K — each is a separate trade-off, kept off the v1 list to ship.
  • Pull quote sourcing and projection ghost data. Both overlays are off by default; their content sources are Phase 2 questions.

The sharpest threat

The audio and overlays read as decorative rather than data-bearing.

If a viewer plays the export and concludes nice ambient music, pretty annotations, the surface has failed at its job. The product's promise is that the music and the overlays are the data, surfaced through a different sense or a different channel. The whole architectural decision to share the key-moment signal between audio and overlays exists to defeat this threat — same per-frame intensity signal feeds the audio's swells, the record flash, the timeline ribbon markers, and the anomaly indicator's contrast peaks.

What kills the threat: a viewer noticing — even unconsciously — that the music swells at exactly the moment +1.7° appears on screen. What feeds the threat: stock-feeling music, generic data-vis overlay aesthetics, key-moment detection that produces moments which are not actually audible or visually distinguishable from non-moments. The risk is highest at the audio side; the overlay side is more visually obvious. RFC-006's stem-based architecture and RFC-007's detection threshold both have to land or the frame collapses into "slideshow with music."

Open threads

  • RFC-006 — Audio system. Stem-based crossfading driven by data. Mubert vs Beatoven choice still open. Self-hostable fallback path required.
  • RFC-007 — Key moment detection. The shared signal both audio and overlays consume. Threshold definitions, user-facing moment names, false-positive handling.
  • UXS-005 — Video Editor. Static visual contract. Already enumerates the ten overlays (eight default-on, two deferred).
  • ADR — Audio API choice. Closes once RFC-006 lands.
  • Phase 2 — Pull quote and projection ghost. Both overlays exist as toggles in v1 but their content sources (curated quote library, CMIP6 projection dataset) are Phase 2.
  • Source — OC-02 Surface 4 · OC-04 Video editor · OC-05 Video editor
  • PA — audiences · promises · principles
  • Related PRDs — PRD-001 Recipe (the work the timelapse is of) · PRD-004 Gallery (the entry point — timelapse ↗)
  • Related UXS — UXS-005 Video Editor (visual contract, overlay enumeration)
  • Related RFCs — RFC-006 Audio system · RFC-007 Key moment detection

Changelog

  • v0.3 · April 2026 — Reshaped to match PRD template (consistent with PRDs 001–004). Replaced PA anchor / IA anchor / Status moves to Implemented when metadata block with template-shape header (Sources · Audiences · Promises · Principles · Why this is a PRD as a sentence). Added magazine-lede opening paragraph. Replaced "What changes for the person" + "The body of the argument" with template sections (The problem · The experience · Why now · Success looks like · Out of scope). Removed the ten-row overlay enumeration table — UXS-005 is the canonical home for that specification; PRDs argue, UXSes specify. Argument unchanged.
  • v0.2 · April 2026 — Added explicit overlay enumeration. Promoted "the overlays" from prose mention to its own listed section with default states. Refined the sharpest-threat paragraph to tie back to specific RFCs (006, 007) and the UXS. Superseded by v0.3.
  • v0.1 · April 2026 — Initial blog-post-format draft, replacing the earlier numbered-FR format. Folded audio + overlay enrichment into a single Video Editor PRD. Superseded by v0.2.