Glossary

Use this page when a field name in a response or in the API reference looks unfamiliar. Terms are grouped by domain.

Data units

Term	Meaning
Video	The input unit of indexing. One media file or URL = one video, identified by video_id.
Scene	A timecoded segment inside a video, produced by automatic shot-cut detection and analyzed as one clip. Search returns matches at the scene level.
Job	One asynchronous indexing run for a video, identified by job_id. Transitions pending → running → done / failed. A re-index creates a new job.
Batch	A group of URLs or files submitted together under one batch_id. The batch endpoint is a submission receipt, not indexing progress.
Burst	In live monitoring, the set of HLS segments captured before an SSAI ad break, analyzed together with its own summary and labels.
Namespace	A tag that partitions data inside the same corpus. Exposed as a filter on live streams.
scene_index	0-based ordinal of the scene inside its video.
start_ms / end_ms	Scene start and end timestamps, in milliseconds.
duration_ms	Length of a video or ad creative, in milliseconds.
length_bucket	Video length category used as a filter: short (up to 2 min), medium (up to 30 min), long.
external_content_id	Your own content id, carried by videos ingested from an MRSS feed. Filter with content_id=.
Cursor	Opaque pagination token on list endpoints. Pass next_cursor back as cursor; null means the last page.
NDJSON	Newline-delimited JSON: one JSON object per line. Used by the scene stream, ad matching, live events and the search AI judge.

Search

Term	Meaning
Modality	Which embedding drives search: caption (narrative description), transcript (verbatim speech), omni (visual + audio), both (caption + omni), all (all three).
k	Videos returned per modality column. Default 20, max 100.
scene_pool	Scene candidates pulled from vector search before grouping by video. Default 60, max 200.
Deep search	POST /api/v1/search/deep: the same query with an extra LLM rerank pass over the top candidates. Better ordering, a few seconds more latency.
Multimodal search	POST /api/v1/search/multimodal: query by image or audio URL in addition to, or instead of, text.
AI judge	Opt-in judge: "text" on POST /api/v1/search (caption modality). Streams a 0–10 relevance and reason per scene and a summary keeping scenes at or above min_relevance (default 3).
score	Similarity score of a scene match; a video's score blends its best scene with the number of matching scenes.

Brand safety

Term	Meaning
GARM	The GARM Brand Safety Floor categories, a brand-safety framework published by the World Federation of Advertisers. Indexing uses a 12-slug vocabulary with levels; ad screening uses the 11-category Brand Safety Floor. See Taxonomies and verdicts.
GARM level	Severity of a category in a video: low, medium, high. Scene markers add none.
garm_min_level	Filter floor on search and the video list: only videos at this level or higher.
garm_categories	Scene-level GARM flags (sensitive, unreviewed).
garm_video	Video-level GARM assessment, one {category, level, reasoning} entry per confirmed category.
block_decision	Final ad-screening verdict: allow or block.
Rule	A natural-language policy line from your saved blocklist or the request's rules. Evaluated independently; reported in blocklist.hits.
Matched rule	A rule whose blocklist.hits[i].matched is true.
Fail-closed	If screening itself errors, the verdict is block with screening_failed: true, never allow.
Explanation	Free-form sentence citing the deciding cue. For display, not parsing.

Ad operations

Term	Meaning
Ad	A registered ad creative, identified by a UUID ad_id, reused across matching and screening.
Ad status	resolving (URL / VAST being fetched and validated), ready (media stored, usable), failed (error carries the reason). File uploads are ready immediately.
source_type	How the creative entered the registry: file, url, vast_url, vast_xml. VAST inputs are unwrapped to the media URL.
Ad analysis	GET /api/v1/ads/{ad_id}/analysis: what the creative sells (primary_brand, primary_product), everything that appears (brands, products), and its IAB / GARM labels.
Ad matching	POST /api/v1/ads/match: streams the scenes an ad fits into as NDJSON events — analyzing, ad_analysis, agent_step, meta, one result per scene, done (or error).
brand_hints	Optional manual brand keywords (case-insensitive) folded into matching when the captioner missed a logo.
Break point	A position in a video where an ad can be inserted. Ad matching result events give it as timestamp_ms (absolute, integer milliseconds).
score	Ad–scene fit from 0 to 10 on matching result events, with a reason string.

IAB taxonomies

Term	Meaning
IAB Content Taxonomy	IAB Tech Lab's content category standard. Tier-1 names are accepted by the iab_t1 filter; leaf labels are returned in iab_content_leaf.
iab_audience_interest_leaf	IAB Audience Taxonomy interest segments for the viewers a video suits.
iab_audience_purchase_intent_leaf	IAB Audience Taxonomy purchase-intent segments (for example auto or travel intenders).
iab_ad_product_leaf	IAB Ad Product Taxonomy categories of ads that fit next to the content.
Leaf	The most specific label in a taxonomy branch. The API returns leaf labels and filters by Tier-1 names.

Lifecycle

Term	Meaning
Job status	pending (queued) → running (pipeline in flight) → done (success) / failed (error).
total_scenes / done_scenes	Indexing progress on a job: scenes analyzed so far vs. total.
reason_code	Stable cause of an indexing failure (for example source_forbidden, url_unreachable, media_unreadable). Only meaningful when the video is failed.
Indexed	The job reached done: summary, labels, scenes and aggregations are available and the video is searchable.
Re-index	A later indexing attempt on an existing video. Produces a new job_id; index_trigger is reindex and reindex_run_id identifies the run.
indexed_at	When the most recent finished attempt ended, successful or not.
created	On URL submissions, true when the call created the video; false when the same URL was already being indexed and the existing ids were returned.
Custom summary	A summary of an indexed video written to your instructions, stored apart from the standard result. Status queued → ready / rejected / failed.
MRSS feed	A media RSS feed registered in the dashboard and polled on a schedule; each item becomes a video with an external_content_id.
storage_path	Internal object key returned on uploads and ad records. Not needed for any API call.
Quota	Per-organization limits on videos, minutes and ads, reported by GET /api/v1/usage.

Auth

Term	Meaning
API key	vuf_live_<keyid>.<secret>, sent as Authorization: Bearer ... on every request. See Authentication.
keyid	Public identifier half of the key, used server-side to look the key up.
secret	Private half of the key. Shown once at creation; the server stores only a hash.
Organization	The account that owns videos, ads, feeds and quota. Every key belongs to one organization, and list endpoints are scoped to it.