Step 1
Corpus Boundary
Define the institution, domain, time range, document classes, inclusion rules, and update cadence. A Semantic Terrain only has meaning inside a declared corpus boundary.
The Method Stack
How Semantic Terrains are computed
CAINC computes terrain before a user asks a question. The methodology turns a bounded corpus into a source-grounded terrain whose objects can be navigated, queried, compared, and tracked over time.
Step 2
Ingest
Collect and preserve the full document corpus: press releases, filings, opinions, regulatory actions, public communications, and other approved source types. Every document remains available as evidence.
Step 3
Normalize & Embed
Normalize text, preserve metadata, and map each document into semantic space. Similar documents move near each other; distinct themes separate. This creates the semantic substrate for the terrain.
Step 4
Terrain Structure
Compute the density landscape of the corpus. Identify peaks, basins, passes, contours, corridors, and the skeleton that connects them. Detect communities at multiple scales, from broad institutional themes to fine-grained topic neighborhoods. Use persistence analysis to distinguish durable structure from noise.
Step 5
Evidence & Entities
Attach source documents, entities, dates, jurisdictions, organizations, people, amounts, and other extracted signals back onto the terrain. Every region can be inspected through the documents and entities that define it.
Step 6
Search & Navigation
Build terrain-aware search so queries can be scoped to the whole corpus, a region, a peak, an entity, a corridor, or a route. Search becomes navigation through the terrain, not just retrieval from a list.
Step 7
Delivery & Cadence
Render the computed structure as a navigable Semantic Terrain with Terrain Map, Network View, Article View, and Entity View. Each terrain updates on its defined cadence as the live corpus grows, producing versioned snapshots that can be compared over time.
Instrumentation
Transparency baked in
Every terrain computation leaves a trail that both humans and automated agents can audit.
- →Versioned computation pipelines with full audit trails
- →LLM-ready Markdown mirrors for every research artifact
- →Source graph linking each terrain feature to the documents that define it
- →Health checks on embeddings and topological computation parameters
What the method does not claim
Constraints
- →Terrain is not truth; it is structured evidence
- →Shape is not intent
- →Movement is not certainty
- →Movement requires multiple terrain snapshots across cadence, not one static map
- →Evidence must be inspected
- →Every terrain reading is constrained by corpus boundaries and source quality
Related resources
For definitions of terrain features, see Terminology. For the full product overview, visit CAINC. To explore a live Semantic Terrain, visit the free public terrain.