NERVE is live across local, private and approved-cloud deployments. Book an evaluation

Platform

One governed path from approved source to searchable information.

The NERVE pipeline separates source preservation from interpretation, keeps processing history explicit and publishes records to search only when their required envelope is complete.

Shipped platform

Processing lifecycle

Preservation first. Interpretation after.

Objects all follow the same route. Metadata and content transformation run as parallel branches after extraction, and local vectorisation is an optional branch rather than a gate, so structured and full-text publication never waits on embeddings.

Ingest

Source, dataset, actor, authority references and scope

Raw commit

Complete accepted bytes, hash and acquisition envelope recoverable before ordinary extraction

Malware scanning, required policy checkpoint

Runs after original preservation and before ordinary parser dispatch. Unavailable scanning keeps affected content quarantined; it is never reported as clean.

Extract

Bounded, isolated processors identify containers and supported structures

Transform: metadata

Provenance, source context, distinct timestamps, handling, privacy, purpose and retention

Transform: content

Source-explicit content mapped to typed canonical records and exact locators; unmapped fields preserved in extensions

Govern / label

Inherited restrictions and policy applied. Ordinary processing can raise restrictions but cannot silently lower them

Persist canonical

Typed records, metadata and lineage committed durably

Index

Rebuildable full-text, structured, temporal and geospatial projections where applicable

Direct path

Structured and full-text publication proceeds without embeddings

Vectorise locally (optional)

Derived semantic representations from eligible text using an approved pinned local model

Publish to search

The object’s search-publication state is selected under policy. Storage alone is not publication

The processing lifecycle end to end. Stage names match the service contracts described below.

Stage by stage

Open a stage to read its purpose, durable result and failure behaviour.

Purpose

Source, dataset, actor, purpose and authority references and scope identify what is being acquired. Admission and original-content acceptance are distinct decisions.

Durable result

A recorded admission decision and an acquisition envelope that states where the material came from and under what authority.

Failure behaviour

A rejected admission produces no accepted original and no canonical record.

Publication state

Stored is not the same as searchable.

Each object carries an explicit publication state. Policy selects the state; persistence alone never implies ordinary discovery.

State

Meaning

NotSearchable

Ordinary discovery is unavailable.

Ordinary discovery is unavailable.

MetadataOnly

Policy permits a metadata-only view.

Policy permits a metadata-only view.

StructuredAndText

Eligible conventional indexes are available.

Eligible conventional indexes are available.

SemanticReady

Approved eligible semantic indexing is also available.

Approved eligible semantic indexing is also available.

Withdrawn

The object is no longer published to ordinary search.

The object is no longer published to ordinary search.

Three information layers

Original, canonical and derived stay distinguishable.

Keeping the layers separate is what allows an improved parser to produce better interpretation without disturbing the retained original or an existing citation.

Layer 01

Original

The exact accepted bytes or source payload, retained with acquisition context and cryptographic identity.

Layer 02

Canonical / normalised

Immutable source-derived records in a common, versioned envelope with typed content and source-specific extensions.

Layer 03

Derived

Subsequent outputs such as OCR text, transcripts, translations, thumbnails, extracted identifier sets, machine suggestions and embeddings, each with its own provenance.

Core acquisition, deterministic normalisation and conventional search carry no mandatory public-cloud AI dependency. Optional AI outputs stay visibly derived, and vector retrieval stays switched off until an organisation approves it.

Preservation and reprocessing

A better parser never rewrites the past.

Reprocessing creates a new immutable transformation run. The earlier output is retained and an existing citation stays resolvable for that retained version, subject to current authority and retention.

Original

Accepted bytes, cryptographic identity and acquisition context. Committed once and never rewritten.

Immutable

Transformation run v1, parser version 1

Output retained. An earlier evidence reference stays attached to this version and this passage.

Transformation run v2, parser version 2

A new immutable output from an improved parser. Nothing earlier is replaced or silently corrected.

How a reprocessing run adds a new output while earlier citations stay resolvable.

Operational truth

Processing failure remains visible.

Outcomes are distinguished rather than collapsed into success or error, so an operator can always tell the difference between work in progress, held content and a genuine failure.

Queued

Accepted and awaiting a worker.

Processing

An attempt is running within its bounded limits.

Partial

Some children or stages succeeded. A partial job is not a complete job.

Quarantined

Policy holds the content, including when mandatory scanning is unavailable.

Failed

The attempt did not produce its required output. The original remains retained.

Searchable

A publication state permits ordinary discovery under policy.

Disposed

An authorised disposition decision has been carried out.

Unprocessed

An unsupported format may be preserved without semantic interpretation when policy permits retention.

A failed optional enrichment need not block otherwise eligible canonical content. Independent accepted records continue while a corrupt archive child produces a visible partial job.

Durability

Built to be interrupted and resumed.

Resumable stage boundaries

Work picks up at the last completed stage instead of restarting an entire acquisition after an interruption.

Idempotent retries

Repeating a stage with the same inputs produces the same durable result, never duplicate records.

Explicit backpressure

When a queue or worker pool is saturated, admission slows visibly instead of work being dropped quietly.

Deduplication boundaries

Identical bytes may share physical storage within one authorised organisation, but separate acquisitions keep their own histories. Cross-organisation physical deduplication is prohibited.