> ## Documentation Index > Fetch the complete documentation index at: https://docs.mixpeek.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Create Interaction > Record a search interaction (view, click, feedback, etc.). Automatically computes and injects ``metadata.reward_value`` based on the interaction types and the retriever's reward_map configuration. This pre-computed value is used by the learned-fusion bandit aggregation query so it doesn't need to join against the retriever config at read time. **Deduplication (Phase 4e):** If ``(execution_id, feature_id, interaction_type)`` has already been recorded within the last 5 minutes the duplicate is silently dropped. **Session cache (Phase 1b):** When ``session_id`` is present the interaction is also written to the Redis session cache so the Thompson Sampling bandit can incorporate it in real time (before ClickHouse write-then-read latency settles). ## OpenAPI ````yaml post /v1/retrievers/interactions openapi: 3.1.0 info: title: Mixpeek API description: >- This is the Mixpeek API, providing access to various endpoints for data processing and retrieval. termsOfService: https://mixpeek.com/terms contact: name: Mixpeek Support url: https://mixpeek.com/contact email: info@mixpeek.com version: '0.82' servers: - url: https://api.mixpeek.com description: Production security: [] paths: /v1/retrievers/interactions: post: tags: - Retriever Interactions summary: Create Interaction description: >- Record a search interaction (view, click, feedback, etc.). Automatically computes and injects ``metadata.reward_value`` based on the interaction types and the retriever's reward_map configuration. This pre-computed value is used by the learned-fusion bandit aggregation query so it doesn't need to join against the retriever config at read time. **Deduplication (Phase 4e):** If ``(execution_id, feature_id, interaction_type)`` has already been recorded within the last 5 minutes the duplicate is silently dropped. **Session cache (Phase 1b):** When ``session_id`` is present the interaction is also written to the Redis session cache so the Thompson Sampling bandit can incorporate it in real time (before ClickHouse write-then-read latency settles). operationId: create_interaction_v1_retrievers_interactions_post parameters: [] requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/SearchInteraction' responses: '200': description: Successful Response content: application/json: schema: $ref: '#/components/schemas/InteractionResponse' '400': description: Bad Request content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '401': description: Unauthorized content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '403': description: Forbidden content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '404': description: Not Found content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' '500': description: Internal Server Error content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' components: schemas: SearchInteraction: properties: feature_id: type: string title: Feature Id description: >- ID of the document/feature that was interacted with. REQUIRED. This is the SAME value as the `document_id` returned in retriever execute results — you can pass it as either `feature_id` or `document_id` (the latter is accepted as an alias). Used to track which specific items users engage with. examples: - doc_abc123 - prod_xyz789 - feat_12345 interaction_type: items: $ref: '#/components/schemas/InteractionType' type: array minItems: 1 title: Interaction Type description: >- List of interaction types that occurred. REQUIRED. Multiple types can be recorded simultaneously (e.g., VIEW + CLICK + LONG_VIEW for a result the user engaged with). Use the InteractionType enum values. examples: - - click - - positive_feedback - click - long_view - - negative_feedback - - purchase - click position: type: integer minimum: 0 title: Position description: >- Position in search results where interaction occurred (0-indexed). REQUIRED. Critical for Learning to Rank - helps identify position bias. E.g., position=0 means first result, position=9 means 10th result. Higher engagement at lower positions suggests higher quality. examples: - 0 - 1 - 5 - 15 metadata: anyOf: - additionalProperties: true type: object - type: 'null' title: Metadata description: >- Additional context about the interaction. NOT REQUIRED. Can include device, duration, viewport info, etc. Use this to enrich interaction data with application-specific context. examples: - device: mobile duration_ms: 5000 page: search_results viewport_position: 0.75 - interaction_reason: not_relevant page_number: 2 results_count: 50 search_latency_ms: 150 user_id: anyOf: - type: string - type: 'null' title: User Id description: >- Customer's authenticated user identifier. NOT REQUIRED. Persists across sessions for long-term tracking. Enables personalization and user-specific metrics. Use your application's user ID format. examples: - user_abc123 - customer_456 - usr_xyz789 session_id: anyOf: - type: string - type: 'null' title: Session Id description: >- Temporary identifier for a single search session. NOT REQUIRED. Typically 30min-1hr duration. Tracks anonymous and authenticated users within a session. Use to group related queries and understand search journeys. Also used by learned fusion (auto-tune) for within-session real-time adaptation: interactions sharing a session_id allow the bandit to update feature weights mid-session without waiting for batch aggregation. examples: - sess_abc123 - session_xyz789_1234567890 execution_id: anyOf: - type: string - type: 'null' title: Execution Id description: >- ID of the retriever execution that generated these results. NOT REQUIRED but HIGHLY RECOMMENDED for training and optimization. Links the interaction back to the exact search query, pipeline configuration, and stage execution that produced the results the user saw. Essential for: fine-tuning embeddings, training rerankers, query understanding, and tracing which pipeline configs produce better user engagement. Retrieve from the retriever execution response and pass to interactions. examples: - exec_abc123xyz - exec_550e8400e29b41d4 retriever_id: anyOf: - type: string - type: 'null' title: Retriever Id description: >- ID of the retriever that was executed. NOT REQUIRED but RECOMMENDED for multi-retriever analytics. Enables comparing performance across different retriever configurations. If execution_id is provided, retriever_id can be inferred from the execution record. examples: - ret_abc123 - ret_product_search_v2 query_snapshot: anyOf: - additionalProperties: true type: object - type: 'null' title: Query Snapshot description: >- Snapshot of the query input that generated these results. HIGHLY RECOMMENDED for training optimization. Storing the query directly enables 10-100x faster training data extraction by avoiding expensive joins to execution records. Use the same format as retriever query input (e.g., {'text': '...', 'filters': {...}}). Essential for: embedding fine-tuning (query-document pairs), query expansion learning, and analyzing which query patterns lead to better engagement. NOT REQUIRED but strongly recommended for production use cases involving model training. examples: - text: wireless headphones - filters: price: lte: 1000 text: laptop - context: summer collection text: red shoes document_score: anyOf: - type: number - type: 'null' title: Document Score description: >- Initial retrieval score of this document when shown to the user. HIGHLY RECOMMENDED for Learning to Rank (LTR). This is a critical feature for reranker training - helps the model learn how to adjust initial scores based on user engagement. Should match the score from the retriever execution results. NOT REQUIRED but strongly recommended for LTR and reranker training. examples: - 0.95 - 0.87 - 0.62 result_set_size: anyOf: - type: integer minimum: 1 - type: 'null' title: Result Set Size description: >- Total number of results shown to the user in this search. NOT REQUIRED but useful for context. Helps understand interaction patterns - clicking position 5 of 10 results is different from position 5 of 100 results. Useful for position bias correction and CTR analysis. examples: - 10 - 20 - 50 - 100 feature_uri: anyOf: - type: string - type: 'null' title: Feature Uri description: >- Feature URI that produced the clicked result (e.g. 'mixpeek://text_extractor@v1/embedding'). Required for learned fusion — interactions without this field do not contribute to weight learning. examples: - mixpeek://text_extractor@v1/multilingual_e5_large_instruct_v1 - mixpeek://clip_extractor@v1/image_embedding occurred_at: anyOf: - type: string format: date-time - type: 'null' title: Occurred At description: >- When the interaction actually happened (ISO 8601). OPTIONAL. Omit for live interactions — the server stamps 'now'. Supply this ONLY to backfill historical interactions (e.g. migrating existing click logs) so learned-fusion temporal decay weights them by their TRUE age. A naive datetime is interpreted as UTC; a future value is clamped to now. Backfilled events bypass the real-time session cache. examples: - '2026-01-15T10:30:00Z' - '2025-12-01T00:00:00+00:00' type: object required: - feature_id - interaction_type - position title: SearchInteraction description: >- Records a user interaction with a search result. This model captures user behavior signals that can be used to improve retrieval quality. Each interaction represents a user action (click, view, feedback) on a specific document returned by a retriever. Use Cases: - Track which search results users actually click on - Collect explicit feedback (thumbs up/down) on result quality - Monitor engagement metrics (time spent viewing, sharing) - Identify problematic queries (zero results, immediate refinements) - Power Learning to Rank models with real user behavior Requirements: - feature_id: REQUIRED - The document/feature that was interacted with - interaction_type: REQUIRED - Type(s) of interaction that occurred - position: REQUIRED - Where in results list the interaction occurred (for LTR) - query_snapshot: HIGHLY RECOMMENDED - Query input for training optimization - document_score: HIGHLY RECOMMENDED - Initial score for LTR training - result_set_size: OPTIONAL - Total results shown (for context) - metadata: OPTIONAL - Additional context about the interaction - user_id: OPTIONAL - For personalization and user-specific metrics - session_id: OPTIONAL - For tracking multi-query sessions - execution_id: OPTIONAL - Link to full execution context - retriever_id: OPTIONAL - For multi-retriever analytics Related Concepts: - Retrievers: Interactions measure retriever performance - Evaluations: Interactions provide real-world complement to offline evaluation - Learning to Rank: Interactions train ranking models examples: - description: >- Minimal click interaction (query_snapshot and document_score optional) execution_id: exec_minimal_123 feature_id: doc_abc123 interaction_type: - click position: 2 session_id: sess_minimal - description: Simple click interaction with recommended fields for training document_score: 0.85 feature_id: doc_abc123 interaction_type: - click position: 2 query_snapshot: text: machine learning tutorials - description: Engaged view with metadata and execution tracking document_score: 0.92 execution_id: exec_abc123xyz feature_id: prod_xyz789 interaction_type: - view - click - long_view metadata: device: mobile duration_ms: 12000 viewport_position: 0.85 position: 0 query_snapshot: filters: price: lte: 100 text: wireless headphones result_set_size: 20 retriever_id: ret_product_search session_id: sess_abc user_id: user_123 - description: Negative feedback document_score: 0.78 feature_id: doc_bad456 interaction_type: - negative_feedback - return_to_results metadata: interaction_reason: not_relevant position: 5 query_snapshot: text: python tutorial result_set_size: 50 session_id: sess_xyz - description: >- Purchase conversion with full tracking (high-value signal for training) document_score: 0.88 execution_id: exec_purchase_789 feature_id: prod_789 interaction_type: - purchase metadata: conversion_time_seconds: 120 purchase_amount: 199.99 position: 1 query_snapshot: text: noise cancelling headphones result_set_size: 50 retriever_id: ret_product_search session_id: sess_purchase_001 user_id: user_456 InteractionResponse: properties: feature_id: type: string title: Feature Id description: >- ID of the document/feature that was interacted with. REQUIRED. This is the SAME value as the `document_id` returned in retriever execute results — you can pass it as either `feature_id` or `document_id` (the latter is accepted as an alias). Used to track which specific items users engage with. examples: - doc_abc123 - prod_xyz789 - feat_12345 interaction_type: items: $ref: '#/components/schemas/InteractionType' type: array minItems: 1 title: Interaction Type description: >- List of interaction types that occurred. REQUIRED. Multiple types can be recorded simultaneously (e.g., VIEW + CLICK + LONG_VIEW for a result the user engaged with). Use the InteractionType enum values. examples: - - click - - positive_feedback - click - long_view - - negative_feedback - - purchase - click position: type: integer minimum: 0 title: Position description: >- Position in search results where interaction occurred (0-indexed). REQUIRED. Critical for Learning to Rank - helps identify position bias. E.g., position=0 means first result, position=9 means 10th result. Higher engagement at lower positions suggests higher quality. examples: - 0 - 1 - 5 - 15 metadata: anyOf: - additionalProperties: true type: object - type: 'null' title: Metadata description: >- Additional context about the interaction. NOT REQUIRED. Can include device, duration, viewport info, etc. Use this to enrich interaction data with application-specific context. examples: - device: mobile duration_ms: 5000 page: search_results viewport_position: 0.75 - interaction_reason: not_relevant page_number: 2 results_count: 50 search_latency_ms: 150 user_id: anyOf: - type: string - type: 'null' title: User Id description: >- Customer's authenticated user identifier. NOT REQUIRED. Persists across sessions for long-term tracking. Enables personalization and user-specific metrics. Use your application's user ID format. examples: - user_abc123 - customer_456 - usr_xyz789 session_id: anyOf: - type: string - type: 'null' title: Session Id description: >- Temporary identifier for a single search session. NOT REQUIRED. Typically 30min-1hr duration. Tracks anonymous and authenticated users within a session. Use to group related queries and understand search journeys. Also used by learned fusion (auto-tune) for within-session real-time adaptation: interactions sharing a session_id allow the bandit to update feature weights mid-session without waiting for batch aggregation. examples: - sess_abc123 - session_xyz789_1234567890 execution_id: anyOf: - type: string - type: 'null' title: Execution Id description: >- ID of the retriever execution that generated these results. NOT REQUIRED but HIGHLY RECOMMENDED for training and optimization. Links the interaction back to the exact search query, pipeline configuration, and stage execution that produced the results the user saw. Essential for: fine-tuning embeddings, training rerankers, query understanding, and tracing which pipeline configs produce better user engagement. Retrieve from the retriever execution response and pass to interactions. examples: - exec_abc123xyz - exec_550e8400e29b41d4 retriever_id: anyOf: - type: string - type: 'null' title: Retriever Id description: >- ID of the retriever that was executed. NOT REQUIRED but RECOMMENDED for multi-retriever analytics. Enables comparing performance across different retriever configurations. If execution_id is provided, retriever_id can be inferred from the execution record. examples: - ret_abc123 - ret_product_search_v2 query_snapshot: anyOf: - additionalProperties: true type: object - type: 'null' title: Query Snapshot description: >- Snapshot of the query input that generated these results. HIGHLY RECOMMENDED for training optimization. Storing the query directly enables 10-100x faster training data extraction by avoiding expensive joins to execution records. Use the same format as retriever query input (e.g., {'text': '...', 'filters': {...}}). Essential for: embedding fine-tuning (query-document pairs), query expansion learning, and analyzing which query patterns lead to better engagement. NOT REQUIRED but strongly recommended for production use cases involving model training. examples: - text: wireless headphones - filters: price: lte: 1000 text: laptop - context: summer collection text: red shoes document_score: anyOf: - type: number - type: 'null' title: Document Score description: >- Initial retrieval score of this document when shown to the user. HIGHLY RECOMMENDED for Learning to Rank (LTR). This is a critical feature for reranker training - helps the model learn how to adjust initial scores based on user engagement. Should match the score from the retriever execution results. NOT REQUIRED but strongly recommended for LTR and reranker training. examples: - 0.95 - 0.87 - 0.62 result_set_size: anyOf: - type: integer minimum: 1 - type: 'null' title: Result Set Size description: >- Total number of results shown to the user in this search. NOT REQUIRED but useful for context. Helps understand interaction patterns - clicking position 5 of 10 results is different from position 5 of 100 results. Useful for position bias correction and CTR analysis. examples: - 10 - 20 - 50 - 100 feature_uri: anyOf: - type: string - type: 'null' title: Feature Uri description: >- Feature URI that produced the clicked result (e.g. 'mixpeek://text_extractor@v1/embedding'). Required for learned fusion — interactions without this field do not contribute to weight learning. examples: - mixpeek://text_extractor@v1/multilingual_e5_large_instruct_v1 - mixpeek://clip_extractor@v1/image_embedding occurred_at: anyOf: - type: string format: date-time - type: 'null' title: Occurred At description: >- When the interaction actually happened (ISO 8601). OPTIONAL. Omit for live interactions — the server stamps 'now'. Supply this ONLY to backfill historical interactions (e.g. migrating existing click logs) so learned-fusion temporal decay weights them by their TRUE age. A naive datetime is interpreted as UTC; a future value is clamped to now. Backfilled events bypass the real-time session cache. examples: - '2026-01-15T10:30:00Z' - '2025-12-01T00:00:00+00:00' interaction_id: anyOf: - type: string - type: 'null' title: Interaction Id description: >- Unique identifier for this interaction record. System-assigned UUID. Use this to reference the interaction in subsequent requests. Null when the interaction was deduplicated (see ``deduplicated`` field). examples: - int_abc123xyz789 - 550e8400-e29b-41d4-a716-446655440000 deduplicated: type: boolean title: Deduplicated description: >- True when this interaction was a duplicate of a recently recorded interaction with the same (execution_id, feature_id, interaction_type) tuple and was silently dropped. The ``interaction_id`` will be null. default: false timestamp: anyOf: - type: string - type: 'null' title: Timestamp description: >- ISO 8601 timestamp when the interaction was recorded. System-assigned. Used for time-based analysis, training data recency weighting, and temporal trends in user behavior. examples: - '2025-01-15T10:30:00Z' - '2025-01-15T14:45:30.123Z' warning: anyOf: - type: string - type: 'null' title: Warning description: >- Non-blocking warning about the interaction. Present when the request was accepted but may not behave as intended (e.g. missing feature_uri for learned fusion). type: object required: - feature_id - interaction_type - position title: InteractionResponse description: |- Response model for a stored interaction. Extends SearchInteraction with system-assigned fields. examples: - description: >- Minimal click interaction (query_snapshot and document_score optional) execution_id: exec_minimal_123 feature_id: doc_abc123 interaction_type: - click position: 2 session_id: sess_minimal - description: Simple click interaction with recommended fields for training document_score: 0.85 feature_id: doc_abc123 interaction_type: - click position: 2 query_snapshot: text: machine learning tutorials - description: Engaged view with metadata and execution tracking document_score: 0.92 execution_id: exec_abc123xyz feature_id: prod_xyz789 interaction_type: - view - click - long_view metadata: device: mobile duration_ms: 12000 viewport_position: 0.85 position: 0 query_snapshot: filters: price: lte: 100 text: wireless headphones result_set_size: 20 retriever_id: ret_product_search session_id: sess_abc user_id: user_123 - description: Negative feedback document_score: 0.78 feature_id: doc_bad456 interaction_type: - negative_feedback - return_to_results metadata: interaction_reason: not_relevant position: 5 query_snapshot: text: python tutorial result_set_size: 50 session_id: sess_xyz - description: >- Purchase conversion with full tracking (high-value signal for training) document_score: 0.88 execution_id: exec_purchase_789 feature_id: prod_789 interaction_type: - purchase metadata: conversion_time_seconds: 120 purchase_amount: 199.99 position: 1 query_snapshot: text: noise cancelling headphones result_set_size: 50 retriever_id: ret_product_search session_id: sess_purchase_001 user_id: user_456 ErrorResponse: properties: success: type: boolean title: Success description: Always false for error responses default: false status: type: integer title: Status description: HTTP status code for this error error: $ref: '#/components/schemas/ErrorDetail' description: Error details payload type: object required: - status - error title: ErrorResponse description: Error response model. examples: - error: details: id: ns_123 resource: namespace message: Namespace not found type: NotFoundError status: 404 success: false HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError InteractionType: type: string enum: - impression - view - click - dwell - positive_feedback - negative_feedback - purchase - add_to_cart - wishlist - long_view - share - bookmark - query_refinement - zero_results - filter_toggle - skip - return_to_results title: InteractionType description: >- Types of user interactions with search results. These interaction types are used to track user behavior and improve retrieval quality through Learning to Rank (LTR), collaborative filtering, and embedding fine-tuning. Values: IMPRESSION: Result was rendered on screen (passive signal, no user action) VIEW: User viewed a search result CLICK: User clicked on a search result DWELL: User lingered on a result (weaker than long_view, stronger than view) POSITIVE_FEEDBACK: User explicitly marked result as relevant/helpful NEGATIVE_FEEDBACK: User explicitly marked result as not relevant PURCHASE: User purchased the item (high-value conversion signal) ADD_TO_CART: User added item to cart (mid-funnel signal) WISHLIST: User saved item for later (engagement signal) LONG_VIEW: User spent significant time viewing (dwell time) SHARE: User shared the result (strong engagement signal) BOOKMARK: User bookmarked the result QUERY_REFINEMENT: User modified their search query ZERO_RESULTS: Query yielded no results (helps identify gaps) FILTER_TOGGLE: User modified filters (helps understand intent) SKIP: User skipped over result to click something lower (negative signal) RETURN_TO_RESULTS: User quickly returned to results (negative signal) Usage in Retrieval Optimization: - LTR (Learning to Rank): Train models to predict click-through rates - Collaborative Filtering: Find similar users/items based on interactions - Embedding Fine-tuning: Adjust embeddings based on what users actually click - Query Understanding: Analyze refinements and zero-result queries - Result Quality: Identify poorly-performing results via skip/return patterns ErrorDetail: properties: message: type: string title: Message description: Human-readable error message type: type: string title: Type description: Stable error type identifier (machine-readable) code: anyOf: - type: string - type: 'null' title: Code description: >- Fine-grained error code for programmatic handling (e.g., namespace_name_taken, feature_extractor_not_found). Present only when consumers may need to branch on a specific error condition. details: anyOf: - additionalProperties: true type: object - type: 'null' title: Details description: >- Optional structured details to help debugging (validation errors, IDs, etc.) type: object required: - message - type title: ErrorDetail description: Error detail model. ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type type: object required: - loc - msg - type title: ValidationError ````