Spoke 01 · Ingestion & Storage

SEIB Data Forge

Deterministic Multi-Modal Ingestion & Authoritative Source of Truth (ASOT) Construction

Data Forge is the deterministic ingestion engine responsible for parsing heterogeneous technical data into the immutable cited evidence store (asot/ingest/) without requirement hallucination.

Dual Ingestion Pathways

Legacy Recovery (Brownfield): Layout-aware OCR and bounding-box parsing for scanned PDF specifications, dense multi-page ICD/IDD pinout tables, repeating-chrome Cable Books, and spatial TDP drawings.
Digitally-Native (Greenfield): Direct ingestion of modern digital engineering baselines from Day 1, including OMG SysML v2 (.sysml / .kerml), DOORS/DNG ReqIF, Cameo XMI, and .reqifz archives.

SEIB Data Forge multimodal extraction graph

asot/ingest/ — Provenance Fabric

Coordinate Bounding Box & Ledger

Every extracted requirement, interface pin, and shall statement retains its exact (x, y, w, h) bounding box coordinates and page index.

                                [CITED_SPAN: p.14, box(120, 340, 480, 28)]

                                • shall_text: "System shall maintain telemetry at 100Hz"

                                • sha256_hash: a3f982...b71c04

                                • audit_status: VERIFIED_NO_MUTATION

Integrity & Fail-Closed Governance

Deterministic Edge Ledger: Maps relational edges (traces_to, refines, verifies) directly back to source spans, preserving relational integrity without injecting synthetic XML or hallucinated text.
Span-Guarding: Extracted candidate text must match physical source bytes verbatim; ghost characters or unverified hallucinated expansions trigger instant validation failure.
Operator HITL Staging: Mandatory human-in-the-loop quarantine where engineers inspect and accept extractions before baseline promotion.
Configuration Management: Every artifact is mathematically bound to its source content hash, parser version, and immutable audit timestamp.
Reversibility: Log-driven, batch-scoped rollback mechanics allow teams to excise contaminated ingest batches safely without graph damage.

Next in the SEIB Workbench Pipeline

Explore SEIB Studio & Synthesis →