> ## Documentation Index > Fetch the complete documentation index at: https://docs.hebbianrobotics.com/llms.txt > Use this file to discover all available pages before exploring further. # Every analysed clip of an enriched corpus, with everything measured about it: where it was shot, what it contains, how its frames partition, and how long hands were visible and manipulating. > This is the row a buyer aggregates their own report from. Counts and seconds, never shares, so any subset of clips can be re-totalled without inheriting a denominator we chose. `frame_quality` and `hand_activity` are joined only when named, because each costs one stored read per clip on the page — the page bound is what keeps that finite. ## OpenAPI ````yaml https://api.hebbianrobotics.com/openapi.json get /api/quality/clips openapi: 3.1.0 info: title: squash description: >- Semantic search and debugging over LeRobot robot-episode datasets: register and index datasets, search episodes by natural language with frame-level evidence, inspect and annotate episodes, run episode-level analytics, export selections as new LeRobot datasets, and follow every operation live on the global event stream. Requests may select an organization's data by sending its id in the `X-Squash-Organization-Id` header; without it a request is scoped to public data. license: name: Apache-2.0 identifier: Apache-2.0 version: 0.0.0 servers: - url: https://api.hebbianrobotics.com description: Hosted squash API security: - {} - bearer_token: [] tags: - name: search description: Natural-language search over indexed episodes. - name: episodes description: >- Browse, inspect, and stream one episode's metadata, frames, and timeseries. - name: datasets description: Register datasets, follow indexing, and manage catalog entries. - name: analytics description: >- Corpus-level measurements: similarity, near-duplicates, coverage, and consistency. - name: quality description: >- Stored Quality Indexing reports: diversity and cleanliness scores over an enriched corpus. - name: annotations description: 'Human curation: episode annotations and mutually-exclusive curation flags.' - name: exports description: Export a selection of episodes as a new LeRobot dataset. - name: jobs description: 'Durable pipeline jobs: acquisition, conversion, and their records.' - name: backfills description: Queue derived-artifact backfills against an active index. - name: events description: Server-sent event streams of live activity. - name: access description: The effective capabilities of the calling principal. paths: /api/quality/clips: get: tags: - quality summary: >- Every analysed clip of an enriched corpus, with everything measured about it: where it was shot, what it contains, how its frames partition, and how long hands were visible and manipulating. description: >- This is the row a buyer aggregates their own report from. Counts and seconds, never shares, so any subset of clips can be re-totalled without inheriting a denominator we chose. `frame_quality` and `hand_activity` are joined only when named, because each costs one stored read per clip on the page — the page bound is what keeps that finite. operationId: quality_clips_route parameters: - name: dataset in: query description: Dataset or corpus id whose index root holds the enrichment. required: true schema: type: string - name: vocabulary in: query description: Stored vocabulary classification. Omit for the most recent. required: false schema: type: - string - 'null' - name: term in: query description: >- Only clips carrying this published term or a spelling that merged into one, in any section. Matching is evidence-based: a spelling no window ever produced selects nothing. required: false schema: type: - string - 'null' - name: frame_quality in: query description: >- Join each clip's frame counts from the most recent frame-quality run. A flag rather than an address, unlike the corpus-level routes: those name the run their whole answer is about, whereas this adds columns to rows the vocabulary already chose. Off by default because each joined column costs one stored read per clip on the page. required: false schema: type: - boolean - 'null' - name: hand_activity in: query description: |- Join each clip's hand-visibility seconds from the most recent hand-activity run. required: false schema: type: - boolean - 'null' - name: offset in: query description: Number of matching clips to skip (default 0). required: false schema: type: - integer - 'null' minimum: 0 - name: limit in: query description: Maximum matching clips returned (default 100, maximum 1000). required: false schema: type: - integer - 'null' minimum: 0 responses: '200': description: A page of analysed clips content: application/json: schema: $ref: '#/components/schemas/QualityClipsResponse' '400': description: Invalid dataset id or out-of-range page content: application/json: schema: $ref: '#/components/schemas/ApiErrorBody' '401': description: A valid bearer credential is required content: application/json: schema: $ref: '#/components/schemas/ApiErrorBody' '404': description: >- No such classification, one of its source runs is not stored, or `term` names nothing published content: application/json: schema: $ref: '#/components/schemas/ApiErrorBody' '409': description: The classification is stale for what its source runs now contain content: application/json: schema: $ref: '#/components/schemas/ApiErrorBody' security: - bearer_token: [] components: schemas: QualityClipsResponse: type: object description: Response of `GET /api/quality/clips`. required: - dataset - vocabulary - filtered_to - total - offset - limit - clips properties: clips: type: array items: $ref: '#/components/schemas/QualityClipDto' description: >- The `offset`/`limit` slice, ordered by episode index; clips whose runs disagree about an index sort last, by identity. dataset: type: string filtered_to: type: array items: $ref: '#/components/schemas/QualityTermFilterMatchDto' description: >- What `term` resolved to, per section it is published in — a query can name a term in more than one, and the clips are the UNION, so every contributor is stated rather than the first. Empty when no `term` was given. limit: type: integer minimum: 0 offset: type: integer minimum: 0 total: type: integer description: Clips matching the filter before pagination. minimum: 0 vocabulary: type: string description: >- The classification these rows were built from — the one asked for, or the most recent when none was named. ApiErrorBody: type: object description: Error payload returned by every failing API route. required: - code - message properties: code: type: string description: Machine-readable error code, e.g. `dataset_not_found`. message: type: string description: Human-readable message. QualityClipDto: type: object description: >- Everything the stored measurements say about one clip. COUNTS and SECONDS throughout, never shares. A share commits the reader to our denominator; counts let them re-aggregate over any subset of clips they care about, which is the whole reason this endpoint exists rather than only the corpus-level ones. required: - clip - analysed_secs - environments - objects - actions properties: actions: type: array items: type: string description: The clip's published action terms. analysed_secs: type: number format: double description: >- Footage assessed, in seconds. A duration, not an offset into the source. clip: type: string description: >- The clip identity, which no renumbering changes. Names the source file and the span within it, so it is the join key a reader already has. environments: type: array items: type: string description: The clip's published environment terms. episode_index: type: - integer - 'null' format: int32 description: >- The episode index the clip had when analysed. Absent when the source runs disagree about it — the corpus was renumbered between describes, and either run's number would address somebody else's footage. minimum: 0 frame_quality: oneOf: - type: 'null' - $ref: '#/components/schemas/ClipFrameQualityDto' description: >- Frame counts for this clip. Absent — never zeroed — when no `frame_quality` was named or this clip was never decoded, because a clip measured and found to have no active frames is a different fact. hand_activity: oneOf: - type: 'null' - $ref: '#/components/schemas/ClipHandActivityDto' description: >- Hand-visibility and manipulation seconds for this clip. Absent when no `hand_activity` was named or the clip carries none. objects: type: array items: type: string description: The clip's published object terms. QualityTermFilterMatchDto: type: object description: |- One published term a `term` filter resolved to (mirror of `squash_analysis::TermFilterMatch`). required: - section - term - clip_count properties: clip_count: type: integer minimum: 0 section: type: string term: type: string ClipFrameQualityDto: type: object description: >- One clip's frames, partitioned. Counts, so the reader picks the denominator. required: - frames - active - distorted - jerky properties: active: type: integer format: int64 minimum: 0 distorted: type: integer format: int64 minimum: 0 frames: type: integer format: int64 minimum: 0 jerky: type: integer format: int64 minimum: 0 ClipHandActivityDto: type: object description: >- One clip's hand activity, in seconds against its own measured footage. The repeated `_seconds` is the unit, and these are field names in somebody else's aggregation code; dropping it would leave `measured` and `hands_visible` with the unit documented only where they are not looking. required: - measured_seconds - hands_visible_seconds - two_hands_visible_seconds - active_manipulation_seconds properties: active_manipulation_seconds: type: number format: double hands_visible_seconds: type: number format: double measured_seconds: type: number format: double two_hands_visible_seconds: type: number format: double securitySchemes: bearer_token: type: http scheme: bearer description: >- Capability-bearing token issued by a squash operator. Required for curation, exports, dataset registration, and job control when the deployment runs in gated mode; public reads stay open. `GET /api/access` reports the capabilities a given credential actually carries. ````