Motion authoring — from measured manifest to real motion

Runs only when the user chose a converted project and authored motion. The generator gives you structure, inert skeleton mounts, and motion.manifest.json (measured triggers, pageTriggers, initial-hidden hints). Authoring is the step where those mounts become real motion and the build is proven against the reference. This is the N2 dogfood recipe (nesil: one 31.21 s master scrub, exact 25,650 px parity).

Profile note: this skill's tools are analyze_page_structure, analyze_scroll_map, build_motion_manifest and retained build_project. Examples below use 1.x names (default full). On AWWWARDS_MCP_PROFILE=v2, use inspect_page {request:{kind:"structure", url}}, analyze_motion {request:{kind:"scroll", url, compact:false}} and build_motion {request:{kind:"manifest", bands, scrollCensus}}; build_project stays flat. The census must be compact:false — a compact census has no anims[] and cannot feed build_motion. Full mapping: v2-routing.md [blocked].

0. Read the reference, not just the census

The census tells you what moves and roughly when; its selectors are structural paths, not your ids/classes. Read the site's own motion script when available and treat the census as corroboration: real trigger count, pins, scrub ratio, duration, easing vocabulary, stagger delays, which animations are CSS keyframes vs GSAP vs native scroll timelines. Use anims[] to verify coverage — never to derive timelines.

1. Decide the topology before writing code

One question: one master timeline or independent section triggers?

• One master scrub (tall spacer, fixed stage, single scrubbed trigger) → author a page-level journey module with one labelled timeline; section modules become thin slice builders that add to it. Do NOT split into seven independent ScrollTriggers — overlap and hand-offs change.
• Independent section triggers → one timeline per section module keyed by the manifest's triggers[]; the skeletons already have the shape (gsap.context, measured constants, refresh).

Rows with start: null are CSS keyframes or unowned rows — provenance, not an instruction to build a timeline each.

2. Manifest contract

motion.manifest.json is evidence, not an authoring file:

• consume as generated; never add hand rows or flip measured
• record counts verbatim (sections, sectionTriggers, pageTriggers)
• record authored code separately (files + before/after) and keep the delivered manifest hash equal to a fresh generation; if you must add rows, mark them authored and say so — never measured: true
• pageTriggers belong to the page-level module, not to a section
• No build yet? build_motion_manifest({ bands, scrollCensus, sections? }) scaffolds this manifest shape straight from analyze_page_structure + analyze_scroll_map (+ optional {label, selector?} overrides). Its envelope is measured: false and DOM initial-state hints are empty (no markup), so author from the census triggers and record authored code as below — never flip the scaffold to measured: true by hand.

3. Intent map (what to port, what to skip)

4. Target surfaces (do not double-load GSAP)

Generated skeletons import npm gsap; replacing that with the kept global avoids two copies. An unused gsap dependency may remain — harmless, do not import it.

5. Authoring order

Page-level first (loader → journey → cursor → audio → ruler), then sections that own independent triggers. Keep the same lock discipline as the section loop: author a section, re-run its section gate (captured geometry + screenshots + explicit checks), lock it, next. verify_build is aggregate whole-page evidence, not a per-section score.

6. Verification gate (all must pass)

• Every target builds (vite build, astro build, next build + start probe).
• Desktop 1440×900 at the band anchors (8 stops): scrollHeight ±1 %, identical visible section ids, fonts, 0 console/page errors/404s.
• Mobile 390×844 smoke: no horizontal overflow, same anchors/visible sections, 0 errors.
• Reduced motion: head class applied, fixed stage disabled, no overflow, page readable.
• Authored proof: zero references to the dropped site scripts in source and build trees; only the kept vendor libs; SiteScripts before MotionRoot.

Failure modes