Data Guide

Core concepts

Trend score

trendScore is the single project-ranking score. It ranges from 0 to 100 and estimates the probability that a project will sustain excess attention over the next six hours. It measures attention, not price direction or expected returns.

trendScoreDelta is the change from the project's preceding canonical score. Use momentum history for the underlying time series: every row exposes data[].timestamp and data[].trendScore.

When present, momentumContext adds computed context:

• trendTrajectory: 24-hour trend-score direction
• distinctClusters: number of represented clusters
• peakRank48h, bestRank14d, and rankedSince: rank history summaries
• leaderboard.top10 and leaderboard.top25: continuous presence and entry counts for each band

Trajectory labels are rising, stable, decaying, or new. Leaderboard stability labels describe continuous time in a band: flash (<2h), fresh (2-6h), persistent (6-24h), or durable (>=24h).

Intel

Intel items group related observations under one structured record. Important fields are:

• detectedAt: first observation time
• reinforcedAt: latest supporting observation time
• observationCount: number of observation entries
• activity: optional observation timeline, controlled by activity=all|observations|none
• hasOfficialSource: whether an observation came from an official project source
• clusters: represented source clusters

Discover current category names with GET /v2/intel-categories or the MCP category-discovery tool before filtering. Categories are service data, not a fixed documentation enum.

Projects and market data

Project objects include AIXBT scores and identifiers for external systems. Depending on the endpoint and projection, they can also include current market metrics, token addresses, recent signals, and momentumContext.

• metrics.usd24hChange is a percentage-point value: -9 means -9%.
• coingeckoData.apiId identifies the asset in CoinGecko.
• tokens[].chain and tokens[].address identify on-chain contracts.
• xHandle identifies the project's X account.

Topics and Frontier Reports

Topics are hourly ranked narratives. GET /v2/topics, /mcp/topics, the topics recipe action, and MCP get_trending_topics expose the current feed. A topic's frontierReport is either { url, preview } or null. Eligible account-backed clients can fetch a published full report through GET /v2/topics/:id/report or MCP get_frontier_report.

The deprecated REST /v2/grounding endpoint and MCP get_grounding tool are compatibility aliases for Topics.

Query patterns

Project research

Resolve a project by name, ticker, or contract address. Use project detail for current fields, momentum or rank history for time series, and the intel endpoint for a longer event history. Project detail includes only its recent embedded signals.

Market overview

Compare the Topics feed with the canonical project leaderboard. In recipes, use top-projects for the current ranking or projects with sortBy: trendScore when you also need general filtering and pagination.

Cluster or category filters

Discover clusters or intel categories first, then pass their current identifiers to the intel endpoint. Multiple category values use OR logic.

Time filters

• detectedAfter and detectedBefore filter by first observation.
• reinforcedAfter and reinforcedBefore filter by latest reinforcement.
• Momentum start and end parameters bound the time series.

Combining detected and reinforced ranges distinguishes when an item first appeared from when it most recently received evidence.

Reading fields together

• Read trendScore, trendScoreDelta, and trendTrajectory together to distinguish the current probability estimate from its recent direction.
• Use distinct-author and cluster breadth, recent intel, and score history as diagnostics for the canonical score, not as alternative rankings.
• Read a leaderboard stabilityLabel as elapsed presence in that band, not as a prediction of future rank.
• Read cluster count as source breadth, not independent confirmation.
• Use detectedAt, reinforcedAt, observationCount, and activity to describe the recorded reinforcement sequence without inferring causality.
• Compare market metrics with their lastUpdatedAt timestamp before combining them with event data.