ARCHITECTURAL SPECIFICATION

An analytical engine for institutional memory

Relay combines deterministic document extraction, multi-agent cross-examination, and direct cryptographic provenance to guarantee that no organizational transition occurs in the dark.

PILLAR 01 // EXTRACTION

Ingestion & Document Intelligence

Relay ingests organizational records without altering their underlying structure, establishing a permanent, searchable institutional knowledge repository.

Multi-Format Parsing

Directly extracts textual structures, markdown hierarchy, and tabular data from PDF, DOCX, and TXT archives.

Deterministic Entity Triples

Identifies explicit relationships (reporting lines, renewal triggers, spend authorities) as structured triples.

Cryptographic Anchors

Every extracted claim preserves character-level start and end coordinates into the original source file with SHA-256 integrity hashes.

PILLAR 02 // RIGOR

The Multi-Tier Evidence Taxonomy

Unlike traditional knowledge bases that treat all statements equally, Relay tags every assertion with an evidence classification.

Evidence-Backed

Ground-truth factual claims backed by verbatim character quotes from authoritative documents.

Inferred Context

Derived logically from adjacent context or patterns, but lacking an explicit, unequivocal primary citation.

Missing Record

Referenced in other records (e.g. “See Schedule B”) but absent from the institutional archive.

Conflicting Evidence

Two or more authoritative records make contradictory assertions (e.g. differing budget totals or conflicting sign-off owners).

Pending Switch

Surfaced during automated pipeline processing and awaiting human adjudication at the Switch review desk.

Experience Relay in action

Open the live Acme BioSystems transition workspace and inspect real cross-examination results.