Player manifest reference

The manifest is the contract between the design plugin and the player. It declares what to render, how to route between variants for a given viewer, and what happens when a viewer interacts.

It is produced by the After Effects plugin at publish time — you do not normally write one by hand — but understanding it explains what a template can and cannot do.

{
  "version": "1.0.0",
  "lottie": { "defaultUrl": "https://…/animation.json" }
}

version and lottie are the only required fields.

Content

Field	Type	Purpose
lottie	object	The animation. defaultUrl plus optional variants
video	object	Background video. defaultUrl, optional variants, and mute
audio	array	Audio tracks. Each has id, defaultUrl, optional variants, and primary
poster	object	Still shown before playback. url plus optional variants
fonts	array	Vectorized glyph artifacts, merged into the Lottie before render

fonts

Each entry is a .font.json artifact the player fetches and merges into the animation before rendering. This is why personalized text renders in the correct typeface with no runtime webfont dependency — no @font-face, no FOUT, and no reliance on the recipient's machine having the font.

Key	Purpose
family	Font family name as referenced by the Lottie
fName	Font name used inside the animation
url	Where to fetch the glyph artifact
style	Optional style qualifier
ascent	Optional ascent override

Personalization

Field	Type	Purpose
dataSchema	object	Declares the data points the template expects
computed	array	Values derived from the supplied data before rendering

dataSchema

Each key describes one data point:

Key	Purpose
type	string, number, or boolean
required	Whether the viewer data must supply it
label, description	Human-facing text for the backoffice
fallback	Value used when the data does not supply one

fallback is what makes a template resilient: a missing data point renders the fallback rather than failing.

computed

Each entry is { key, expr, outputType? }. The expression is evaluated against the viewer's data before rendering, and key is written back into the data — so later computed fields can build on earlier ones.

Available operations include arithmetic and rounding (+, round, floor, ceil), logic (if, and, or, in), string concatenation (cat), and formatting helpers:

Operation	Purpose
formatCLP	Chilean peso currency formatting
formatNumber	Number formatting
formatPercent	Percentage formatting
formatDate	Date formatting (YYYY, MMM, LL, short, long, numeric…)

This is what lets a template show "you saved $1.234.567 this year" from a raw number, without the client's system having to pre-format anything.

Routing between variants

lottie, video, audio, poster, textTracks and transcript all accept variants, and they all use the same rule shape:

{
  "variants": {
    "rules": [
      { "dataKey": "segment", "equals": "premium", "src": "https://…/premium.json" }
    ]
  }
}

A rule matches when the named dataKey in the viewer's data equals the given value. The matched rule supplies the asset — overlayUrl, videoUrl, audioUrl, posterUrl, src, or inline text for transcripts. A viewer matching no rule gets the default.

Comparison is strict, so a rule should key off a boolean or numeric computed field rather than a raw data point: host data often arrives as strings, and "true" does not equal true. The publisher emits a computed for exactly this reason.

The poster follows its own routing, independent of the video's. A template that routes its video by a data point normally wants the still to follow the same branch — otherwise the image advertises a version of the video the viewer will not be shown — but the two are declared separately, so they can diverge deliberately.

This is how one template serves many audiences: the routing happens per viewer, at view time, without rendering a separate video per segment.

Accessibility

Field	Type	Purpose
textTracks	array	Timed captions painted over the video
transcript	object	Full text alternative for the whole video

These are not the same thing, and the distinction matters for compliance.

textTracks are timed captions — each has id, label, srcLang, src, optional variants and default. They serve a viewer who can see the video but not hear it.

transcript is the WCAG 1.2.1 / 1.2.8 text alternative: the full text of the video, useful to a screen reader, which timed captions are not. It carries {dataKey} tokens that are interpolated with the viewer's resolved data — a transcript has to describe what this viewer saw — and it supports the same variants routing, so a viewer routed to another composition reads that composition's text.

The text is inline rather than a separate file. On a 40-variant template the transcripts measured 53 KB raw but 1.7 KB gzipped, because they are ~91% identical — external files would have bought nothing.

Interactions

interactions maps a CTA id to what happens when it is clicked. Keys are matched against the emitted CTA id, case-insensitively.

Key	Purpose
navigate	Opens the URL held in the data point named by linkDataKey, with optional target
setData	Merges values into the live data, re-resolves computed fields and variant rules, and re-renders in place
name	Human label, reported as cta_name on the cta.clicked event. Falls back to the CTA id
question	Groups several CTAs into one survey question

setData

setData re-renders in place — no navigation, no reload. Values may be literal scalars or JSON-logic expressions evaluated against live data at click time, which is what enables flows like a bounded quantity stepper.

The player evaluates each value defensively and keeps the prior data when a value is not evaluable. A malformed value is never rejected at manifest load, because doing so would fail the entire manifest instead of no-op'ing a single key.

When both are present

Both run, in a defined order: the data patch is applied first — so a link the patch itself sets is the one that opens — then the link opens synchronously inside the click, and the re-render is queued behind it. The click autopauses, as any navigation does.

question

Survey answers are authored as independent CTAs, so without question a click is just a click. Give two CTAs the same question and the player additionally emits interaction.engaged, with the question as interaction_id / interaction_name and the CTA's label as interaction_value — which is what lets analytics report a distribution ("62% answered Yes") rather than two unrelated counts.

It is analytics-only and does not affect playback.