Forensic Platform Architecture Map Professional Review Context

Codebase-grounded schematic for evidence-linked financial review, deterministic analysis, and containerized deployment options. Use of the platform does not create or guarantee privilege.

Environment: Docker / Gunicorn / Celery
Core Framework: Django 5.2.9 (current Django 5.2 line)
Database: PostgreSQL (ACID Compliant)

Infrastructure & Persistence Layer

🗄️

PostgreSQL Relational Store

Primary financial source of truth for evidence relationships, reviewed ledger records, coverage, reviewed transfers, tracing runs, and finalized analytical output.

Redis Cache / Broker

Production Compose uses Redis for cache-backed throttling and optional background-task transport. This is not represented as complete DDoS mitigation or durable queueing.

⚙️

Celery Worker

The current worker supports selected background operations such as optional SIEM forwarding. Financial imports and LIBR task adapters execute synchronously and require capacity testing.

⚙️ core

SYSTEM INFRASTRUCTURE

🔐 accounts

IAM & IDENTITY

⚖️ cases

MATTER BOUNDARY

🧮 finance

FORENSIC DATA MODELS

📁 evidence

CRYPTOGRAPHY

💬 communication

GUARDRAILS

🧊 audit

AUDIT LOGGING

💳 subscriptions & payments

DISABLED / CONTROLLED SCOPE
ISOLATED SYSTEM: CORE FORENSIC ENGINE

👁️ ocr_pipeline

SPATIAL EXTRACTION

🏎️ libr_calculation

DETERMINISTIC MATH

Trust Boundaries & Current LIBR Workpaper Flow

The first schematic shows platform trust zones. The second intentionally documents the current LIBR workpaper and verification path; Money Flow remains a read-only projection and Value Lineage remains authenticated, single-account analytical state.

Visual Module 01

Trust boundary map

Distinct zones separate educational public pages from authenticated ingestion, deterministic report generation, and byte-exact verification.

flowchart TB
    subgraph Public["Public zone"]
        PP[Conversion pages and /verify/]
        SP[Synthetic sample workpapers]
    end

    subgraph Auth["Authenticated app"]
        UI[Cases accounts and claims]
        ING[Ingestion and normalization]
    end

    subgraph ClioZone["Clio integration"]
        OAuth[OAuth tokens encrypted at rest]
        SEL[User-selected matter and document]
        REV[Local extraction review job]
    end

    subgraph Gen["Report generation"]
        LIBR[Deterministic LIBR engine]
        RPT[ForensicReport and PDF render]
        HASH[file hash and snapshot digest]
    end

    subgraph Vault["Evidence Vault"]
        ED[EvidenceDocument storage]
    end

    UI --> ING --> LIBR
    OAuth --> SEL
    SEL -->|read-only download| ING
    LIBR --> RPT --> HASH
    RPT --> ED
    SEL --> REV
    HASH --> PP
    SP -.->|demo bytes may not verify| PP

What it provesPublic copy stays bounded while financial review happens in authenticated zones. Clio currently reads one selected source into a local review job. Verification is a separate public gate on exact file bytes.

What can breakEndorsement language crossing from Clio into public pages. Snapshot-only verification messaging. Silent bulk ingestion. Missing file hashes. LIBR or ledger inconsistency undermines every downstream zone.

Visual Module 02

V1 LIBR workpaper data flow

Selected financial records become a single-claim attorney-reviewable workpaper with provenance and integrity surfaces.

flowchart LR
    A[Selected account] --> T[Normalized transactions]
    T --> C[SeparatePropertyClaim]
    C --> S[Disclosed LIBR settings]
    S --> L[LIBR calculation]
    L --> SN[Snapshot SHA-256]
    L --> FR[ForensicReport]
    FR --> PDF[Attorney-reviewable PDF]
    PDF --> EV[Evidence Vault copy]
    PDF --> V["/verify/ exact file hash"]

    SN -.->|printed on workpaper| PDF

What it provesV1 is one inspectable pipeline: calculation-setting disclosure, deterministic replay, stored file hash, and public verification anchored to PDF bytes — not narrative alone.

What can breakTransaction ordering errors. Calculation settings absent from the PDF face. Ledger disagreeing with summary. Vault copy diverging from ForensicReport bytes. Multi-claim exports through the default V1 path.

Selected Financial Review Pipeline

01

Ingestion & Isolation

User uploads a bank PDF. RateLimitMiddleware applies configured cache-backed limits. CaseContextMiddleware provides case context while each view must still enforce object-level permissions.

02

Cryptographic Lock

The file routes to EvidenceDocument. The system records a SHA-256 integrity reference and applies application-layer authenticated encryption at rest. The hash can detect byte changes; it does not authenticate origin or establish legal chain of custody by itself.

03

Reviewed Extraction Candidates

Native PDF extraction runs first where possible. Configured local or cloud OCR may produce candidate rows for degraded pages, with explicit consent where external processing is used. Reconciliation and human review are required before import.

04

Reviewer Decisions

Reviewers approve financial rows, qualify coverage, and decide whether supported transfer proposals become ReviewedTransfers. The software does not infer ownership or legal characterization.

05

Bounded Deterministic Analysis

The LIBR workpaper applies selected LIBR settings to reviewed rows and an explicit source. Separately, EXPLICIT_ALLOCATION_V1 conserves one modeled source inside one account and stops at a transfer boundary.

06

Reviewable Surfaces

Money Flow and Value Lineage remain authenticated review surfaces. When LIBR is selected, a ForensicReport PDF may be generated with disclosed settings, selected rows, provenance references, and a final-file SHA-256 integrity reference.