Every entity in tech news carries meaning beyond its name. "OpenAI" isn't just a company — it's a bundle of associations: safety research, ChatGPT, Microsoft partnership, nonprofit-to-profit conversion, AI regulation debates. Those associations change over time. In 2020, "OpenAI" meant GPT-3 and a research lab. In 2026, it means a $300 billion advertising company with 800 million users. Semantic drift is the measurement of that change.
How It Works
TEXXR converts every article into a 1,536-dimensional vector using embedding models. These vectors capture meaning — not keywords, but the semantic space an article occupies. Two articles about different topics will have distant vectors. Two articles about similar topics will be close, even if they use completely different words.
To measure an entity's drift, TEXXR computes the average embedding of all articles mentioning that entity in each time period (monthly or quarterly). This average — called a centroid — represents what the entity "means" in the coverage during that period. The distance between successive centroids is the drift.
What Drift Reveals
Pivots. When an entity's drift suddenly spikes, it means the coverage fundamentally changed. Arm's Q1 2026 drift of 0.305 — the largest in its history — corresponded to its shift from designing chips for others to building its own AGI CPU.
Convergence. When two entities' drift trajectories move toward each other, they're becoming more similar. OpenAI and Anthropic converged to identical coverage profiles in Q1 2026 — both at 23% competitor, 21% financial — despite being founded on opposite principles.
Regime breaks. TEXXR's regime detection identifies when an entity's drift exceeds a threshold, signaling a structural change in how the entity is covered. These breaks often precede market events — Arm's regime break preceded its stock jump by days.
Drift vs. Momentum
Drift and momentum measure different things. Momentum tracks how much coverage an entity receives (volume, velocity, acceleration). Drift tracks what kind of coverage (meaning, associations, context). An entity can have high momentum (lots of articles) with low drift (same topic each time) — like Nvidia during an earnings season. Or low momentum with high drift — like a small company that pivots into a new market.
The most significant signals occur when both momentum and drift are high simultaneously: an entity is getting more coverage AND the coverage is about something fundamentally new.
Try It
TEXXR's Drift Map visualizes entity trajectories across semantic space. Enter any entity to see its drift over time. The Story Arc tool shows how coverage themes evolve quarter by quarter. Both are built on the same embedding infrastructure.
For programmatic access, the Pulse API exposes drift data via /api/drift/compute?entity=OpenAI and trajectory projections via /api/drift/trajectories.
Related: Vector Patterns and Triangulation explains how the same embeddings power semantic search. What Is a Knowledge Graph? covers how structured relationships complement drift measurement. What Is Entity Momentum? explains the volume-side counterpart to drift.