The Problem: Integration Debt, Not Data Scarcity

Secure organisations accumulate data faster than they can make it usable. Device extractions, mailboxes, file shares, application databases, licensed feeds, sensor streams, geospatial files and legacy data lakes each arrive with their own schema, access path and provenance story.

The result is integration debt. Every new source triggers another bespoke importer, another mapping exercise and another attempt to reconstruct where the material came from. Analysts spend time learning source systems, and the chain from a finished assessment back to the original bytes weakens at each handover.

The Approach: Separate Information Handling From Intelligence Judgement

AIC’s response splits the problem along a deliberate boundary. NERVE handles data and information. CORTEX owns intelligence judgement.

NERVE owns the technical path from source to searchable information: acquire, preserve, inspect, decode, normalise, enrich, index and serve. CORTEX owns the lifecycle around it: direction, collection governance, analysis, product, dissemination and feedback. A NERVE record can state that a message contains a telephone number, and that two records share the same normalised number. Deciding whether that supports a judgement remains an analyst decision inside CORTEX.

Design Decisions That Carry the Weight

• Preserve before interpreting. Accepted original bytes are immutable and retained with their acquisition context before any parsing takes place.

• Transformations are immutable runs. Parsing, normalisation, enrichment, OCR, transcription, translation and embedding all execute as versioned runs, so reprocessing produces new outputs rather than rewriting prior evidence.

• A common envelope with typed content. Canonical fields stay small enough to remain coherent, while source-specific detail is retained rather than discarded to fit the model.

• Provenance as a first-class graph. Every object carries acquisition and transformation lineage, and lineage is transitive across derivatives.

• Authorise before you match. Search authority is applied before matching, counting and aggregation, so hidden records cannot be inferred from counts or facets.

• Idempotent, resumable pipelines. Every stage can be retried safely, backpressure is explicit, and workers treat all source content as hostile input inside bounded isolation.

How NERVE and CORTEX Interact

The integration is service-mediated by design. CORTEX never reads NERVE’s object store, metadata database, search index or message broker. It calls stable search and evidence contracts, which means NERVE can change storage engines, index layouts or worker topology without breaking the analyst experience.

Every analyst search carries two things: the CORTEX workload identity that proves which application is calling, and the end-user authority context that determines what that specific analyst may see. A trusted service cannot borrow its own broad privilege to widen an individual’s scope.

Results return stable evidence references rather than copies. CORTEX stores the reference as a citation, and resolving it later recovers the exact object revision, the locator for the matched passage and the lineage back to the original acquisition. Because references survive reprocessing, an assessment written against version one of a parser still resolves after version two is deployed.

The relationship is deliberately one-directional in terms of dependency. NERVE can ingest, preserve, transform and serve system-level search with no CORTEX requirement attached, which matters for standing datasets, licensed feeds and historical archives. When a requirement or authority reference does exist, it is recorded rather than invented.

The Chain Worth Proving

A meaningful evaluation is not a long connector list. It is one exact chain, demonstrated end to end: source, acquisition, immutable original, transformation run, canonical object, search hit, exact locator, CORTEX evidence reference, analytical judgement.

The second demonstration worth running is source convergence: semantically similar messages or locations arriving from two very different systems, found through a single governed CORTEX query, while each retains its distinct provenance. Add a denied search to prove that counts and facets do not leak, restart workers mid-job to prove durable recovery, then reprocess one source with a new transformation version and confirm the original evidence reference still resolves.

Deployment and Assurance Posture

Local and disconnected operation is a first-class architecture rather than a cut-down mode. A physically disconnected node can accept approved removable-media imports, process content entirely through local services, serve CORTEX locally and export only authorised packages. Core ingestion and search do not depend on public cloud AI, and optional enrichment providers remain separate, tracked derivative stages governed by the same handling policy.

Status

NERVE is published as an intended-state specification. This study describes the designed behaviour and the boundary with CORTEX rather than claiming delivered, accredited or fielded capability. Performance, scale and accreditation statements will follow measured acceptance evidence.

To discuss source landscapes, deployment constraints or an evaluation against your own data, talk to the AIC team.