Schemas

Logical identifier for the development across updates.

Canonical backtest system timestamp of this development version.

Most recent timestamp of any source used for the development.

Actual (estimate of) real world time that the development took place.

Confidence score in estimate of occurrence_timestamp.

Name of the development.

Description of the development.

Type of the development.

Status of the real world occurrence.

Explanation for all decisions made.

List of involved stakeholders and their impacts.

LLM heuristic of general development importance.

LLM heuristic of how surprising the development is to the market.

LLM heuristic of its confidence in everything it presents.

List of events this development is associated with, resolved to their most recent known state.

List of sources used for this development.

Semantic search query. If None, retrieves all developments matching the filters.

Filter for results >= this integer unix timestamp. None means realtime mode.

Filter for results <= this integer unix timestamp. None means realtime mode.

Filter for occurrence_timestamp >= this integer unix timestamp. None means no limit.

Filter for occurrence_timestamp <= this integer unix timestamp. None means no limit.

Filter by development type.

Filter by stakeholder symbol.

Minimum impact score filter in [-1, 1]. If both min and max are provided, matching is directional OR (impact >= min OR impact <= max).

Maximum impact score filter in [-1, 1]. If both min and max are provided, matching is directional OR (impact >= min OR impact <= max).

Minimum importance score in [0, 1].

Minimum surprise score in [0, 1].

Minimum confidence score in [0, 1].

Maximum number of results to return.

Logical identifier for the development across updates.

Canonical backtest system timestamp of this development version.

Actual (estimate of) real world time that the development took place.

Confidence score in estimate of occurrence_timestamp.

Name of the development.

Description of the development.

Type of the development.

Status of the real world occurrence.

LLM heuristic of general development importance.

LLM heuristic of how surprising the development is to the market.

LLM heuristic of its confidence in everything it presents.

Raw list of event IDs this development is associated with.

Logical identifier for the event.

Name of the event.

Description of the event.

Type of the event.

Most recent developments for this event (count controlled by developments_per_event param), ordered by occurrence_timestamp descending.

Semantic search query over event descriptions. If None, returns events ordered by recency of event creation.

Filter for events with development activity >= this unix timestamp.

Filter for events with development activity <= this unix timestamp. None means realtime mode.

Filter by event type.

Maximum number of events to return.

Number of most recent developments to embed per event result, ordered by occurrence_timestamp descending.

Whether to include developments whose occurrence_timestamp is in the future (relative to end_timestamp, or current time if end_timestamp is None) in the embedded per-event development list. Set to False to restrict to already-occurred developments.

Logical identifier for the event across updates.

Name of the event.

Description of the event.

Type of the event.

Logical identifier for the event.

Name of the event.

Description of the event.

Type of the event.

Developments associated with this event, ordered by occurrence_timestamp descending (real-world timeline).

Unique snake_case primary key for the factor. Name must be entirely self contained and describe the situation unambiguously. Name should be consistent and reusable across separate calls.

What the factor represents. This will be embedded for semantic retrieval, so write it as a clean, general definition.

Scale of factor.

Mandatory scoring rubric for consistency. Keep it short but explicit. This should ensure absolute consistency globally across separate calls.

Only for CATEGORICAL: enum of allowed category strings.

A brief explanation of the reasoning behind the assessment for this stakeholder given the factor.

Assessed value (type depends on factor_type). 1. BINARY -> bool; 2. CONTINUOUS -> float in [0,1]; 3. SIGNED_CONTINUOUS -> float in [-1,1]; 4. CATEGORICAL -> str in allowed_values.

List of sources that support this factor assessment.

Factor definition.

Map: stakeholder symbol -> assessment object for this factor.

The ID of the development to retrieve.

Filter for results <= this integer unix timestamp. None means realtime mode.

The event_id to retrieve developments for.

Filter developments with timestamp >= this unix timestamp. None means no lower bound.

Filter developments with timestamp <= this unix timestamp. None means realtime mode.

Maximum number of developments to return, ordered by occurrence_timestamp descending.

Whether to include developments whose occurrence_timestamp is in the future (relative to end_timestamp, or current time if end_timestamp is None). Set to False to restrict results to developments that have already occurred.

List of semantic search queries.

Best-effort target number of results to return after flattening and deduplication.

filter for results >= this integer unix timestamp.

filter for results <= this integer unix timestamp.

filters results for cosine similarity >= this value.

List of exact match keywords. If provided, returns results containing at least one.

List of tags to filter by. If provided, returns results containing all specified tags.

filters results for quality rating >= this value.

Ranking strategy for consolidating results across queries. Final output is always ordered by cosine_similarity. - 'cosine_similarity': rank purely by cosine similarity. - 'round_robin': interleave results from each query in a round-robin fashion to promote diversity across queries. - 'rank_sum': assign a score to each chunk based on its rank in each query's results and sum these scores across queries, promotes duplicates.

The semantic search query string.

filter for results >= this integer unix timestamp.

filter for results <= this integer unix timestamp.

Number of results to return.

filters results for cosine similarity >= this value.

List of exact match keywords. If provided, returns results containing at least one.

List of tags to filter by. If provided, returns results containing all specified tags.

filters results for quality rating >= this value.

Cosine similarity score between the query and the chunk embedding.

Unique identifier for the chunk.

Unix timestamp of the source document.

Identifier of the source document.

List of tags associated with the chunk.

Title of the source document.

Content of the chunk.

List of named entities extracted from the chunk.

Quality rating of the chunk between 0.0 and 1.0.

Source information.

Timestamp of the source item (unix epoch).

Table (Type) of the source item (e.g., 'news', 'earnings_calls', 'press_releases', etc.)

'id' of the source item

Additional metadata about the source item.

URL of the source material, if applicable.

Title of the source material, if applicable.

Publisher of the source material, if applicable.

Timestamp of the source item (unix epoch).

Table (Type) of the source item (e.g., 'news', 'earnings_calls', 'press_releases', etc.)

'id' of the source item

Additional metadata about the source item.

List of EXACT quotes from the source material.

Name of the stakeholder

The type of stakeholder. ONLY the following are allowed: [PUBLIC_COMPANY].

Ticker symbol of the stakeholder

The query / subject for which to identify stakeholders for.

Optional additional context to help identify stakeholders.

Optional additional instructions to guide stakeholder identification.

Optional start timestamp (unix epoch) to consider for time-sensitive queries.

Optional end timestamp (unix epoch) to consider for time-sensitive queries.

List of identified stakeholders with their impact details.

Factor matrix providing scoring for each stakeholder across defined factors.

The stakeholder being identified.

Impact score (-1.0 to 1.0) indicating the estimated magnitude and direction of the impact on the stakeholder.

Confidence score (0.0 to 1.0) indicating level of certainty in the stakeholder being materially impacted.

A brief explanation of why this stakeholder is relevant to the query

List of sources (with quotes) that support the identification of this stakeholder.

Confirmation message indicating successful creation of the webhook.

The unique identifier of the created webhook.

A sample of the data that will be sent to the webhook receiver URL upon execution.

The ID of the webhook.

The path of the desired endpoint.

A dictionary of arguments to call the endpoint with. Note: Do not include 'start_timestamp' or 'end_timestamp' as these are handled by the webhook.

The interval (in minutes) at which the webhook will check the endpoint. Must be a multiple of 5.

The URL to which the webhook payloads will be sent.

Optional headers to include in the webhook requests to your url.

Unique identifier for the webhook.

Current status of the webhook either 'active' or 'paused'.

User defined webhook configuration.

Timestamp indicating when the webhook was created.