FR-1 / INTAKE / COMPONENTS
1 unit
Email Ingestion Service DETERMINISTIC
KEY DECISION
Raw email + attachments are written once to blob storage. Nothing downstream can mutate the evidence trail.
L0 · Architecture explorer
DETERMINISTIC AI-SCOPED
An evidence-first pipeline,
decomposed into nine functional requirements.
Four stage groups. Click any [+] to open the components inside — their ports, their stack, and whether a human, a rule, or a model is on duty.
FR-2 · 3 · 4 / DOCUMENT PROCESSING / COMPONENTS
3 units
PDF Splitter DETERMINISTIC
OCR & Layout Extractor AI-SCOPED
Canonical Normalizer DETERMINISTIC
KEY DECISION
Every extracted value carries an ExtractionCitation pointing back to {artifact, page, snippet}.
FR-5 · 6 · 7 / EVALUATION / COMPONENTS
3 units
Material Classifier DET · AI-READY
Standards Resolver DETERMINISTIC
Evaluation Engine DETERMINISTIC
KEY DECISION
No guessed units. Missing data routes to needs_review — never silent confidence.
FR-8 · 9 / ACTION / COMPONENTS
2 units
Notification Service DETERMINISTIC
Reviewer Access & Export DETERMINISTIC
KEY DECISION
Corrections never overwrite. The audit log is a hash chain — replay reproduces current state exactly.
Raw evidence is never mutated. Derived artifacts trace back to evidence. Canonical records are the sole input to evaluation.
01
Raw Evidence
[ immutable ]
- Original .eml + headers
- Original PDF attachments
02
Derived Artifacts
[ traceable ]
- Split cert instances
- OCR text + layout tables
- Extraction bundle w/ citations
03
Canonical Records
[ structured ]
- Material codes (grade · form · spec)
- Measured values + units
- Cert metadata · vendor · lot
04
Decisions
[ auditable ]
- Mapping decision
- Standard version selected
- Requirement results