← → Navigate · B Static · ESC Index
Tari · Compliance Verification
SCQA · Product Narrative · 01 / 20
SITUATION · COMPLICATION · QUESTION · ANSWER

Tari
Compliance
Verification

Use SCQA to explain the import compliance context, the pain in traditional workflows, and how AI agents turn messy files into an auditable evidence chain.
Based on business-process narrative + AI agent design
IKB Swiss Style
S · Situation
02 / 20
THE BUSINESS STORY

Imports
Need
Proof

From the CBP perspective, every shipment must prove who is responsible, what the goods are, how they are classified, what they are worth, where they originated, how duties are calculated, and where the evidence lives.

Entry-level evidence
Documents
7501

Entry Summary

Entry number, IOR, HTS, origin, entered value, duty.

Invoice

Commercial facts

Goods, price, buyer, seller, quantity, Incoterm.

HBL / B/L

Shipment facts

Route, container number, shipper, consignee, bill of lading number.

ISF / IOR docs

Party facts

Manufacturer, buyer, seller, EIN, account, authorization.

Situation: Every entry is a claim that needs proof.
Situation · Roles
03 / 20
Who participates in one import entry

One import entry, six roles

01

Purchaser

Owns product, supplier, price, and purchase facts.

02

IOR

Importer of Record, accountable to CBP for the filing.

03

Broker

Files on behalf of the IOR and owns entry filing facts.

04

Logistics

Forwarders, carriers, and NVOCCs generate shipping evidence.

05

CBP / ACE

Official system for review, liquidation, refunds, and audits.

06

Tari

Connects roles, documents, fields, and risks into an evidence system.

C · Complication
04 / 20
The trap in product framing

The real blocker is an unclear evidence chain

OLD Narrow view
Extract fields
Export CSV

If the need is framed as OCR, file classification, or CAPE entry-number export, the product becomes a one-off utility.

  • Sees only fields inside a single file
  • Only answers which entries may get refunds
  • Cannot explain why a field is trustworthy
NEW Compliance view
Build evidence
Run verification

Mike’s new direction is a compliance verification database: every entry needs sources, relationships, check results, and next steps.

  • Documents must corroborate one another
  • Risks must become a worklist
  • Audits need a binder ready to show
Q · Question
05 / 20
THE PRODUCT QUESTION

How do we help importers, IORs, brokers, and counsel trust that
this Entry filing is evidenced, explainable, and traceable?

AI should stay focused on evidence organization and risk signaling: evidence supports / evidence missing / needs review / potential issue.

A · Answer
06 / 20
Product positioning

Tari is an evidence-backed compliance database

LAYER 01

Messy trade records

PDFs, Excel files, screenshots, archives, duplicates, corrupt files, and inconsistent naming.

Intake + Classification
LAYER 02

Entry evidence graph

Connect Document, Party, Product, Shipment, Entry, EntryLine, and each field source.

Evidence Graph
LAYER 03

Verification package

Outputs worklists, readiness status, review tables, CAPE lists, and audit binders.

Human Review + Export
Architecture · Evidence System
07 / 20
Core data model

The database must carry evidence relationships across files

Every field needs a source page, confidence, supporting relationships, and review status. That is what makes agent findings handoff-ready.

Objects

Batch / Document / Party / Product / Shipment / Entry / EntryLine

Evidence

ExtractedField / EvidenceLink / source document / page / confidence

Action

VerificationCheck / Finding / QuestionnaireItem / AuditBinder

Entry
Document
7501 / Invoice / ISF
Party
IOR / seller / broker
Product
SKU / HTS / origin
Finding
risk / owner / next action
Agent Workflow
08 / 20
From batch upload to business action

The agent workflow is an evidence production line

01IntakeInventory, unpack, dedupe, corrupt-file detection
02Classify7501, invoice, HBL, ISF, account files
03ExtractEntry, IOR, HTS, value, origin
04Graphentry / shipment / party / product links
05VerifyDomain agents run consistency and gap checks
06ReviewGenerate worklists, tables, audit binder
The system never jumps from file to legal conclusion.
Process Detail · Phase 1
09 / 20
Intake and file normalization

First normalize the client document package

Input

User uploads a batch of files

Files may be PDFs, Excel workbooks, screenshots, or archives. The uploader may be an importer, purchaser, IOR, broker, or compliance team.

Batch

System creates a batch manifest

Generate a batch ID and upload manifest to record exactly what the client submitted.

File Queue

Each file enters the processing queue

The system unpacks, deduplicates, validates, and detects corrupt files; scanned files proceed to OCR.

First Output

Produce a file inventory

Each file gets a type assessment, readability status, and processing log for downstream extraction.

01Upload
02Manifest
03Queue
04Inventory
Process Detail · Phase 2
10 / 20
Per-file extraction stream

Extract business facts by file type

7501

entry、IOR、HTSUS、value、duty

Invoice / Packing

Goods, buyer, seller, quantity, invoice amount

B/L

HBL / MBL, container number, carrier, ports

ISF

manufacturer、seller、buyer、ship-to

IOR Docs

EIN, registration details, entity status

Account / Banking

ACE、5106、TAO、ACH、Notify Party

Support Docs

POs, payments, contracts, certificates of origin, specifications

Facts Store

Field + source page + confidence

02

This phase does not judge risk. It reliably extracts key facts from each file type and records which document and page each fact came from.

Process Detail · Phase 3
11 / 20
Batch close and evidence graph build

Connect scattered files into one Entry evidence chain

After the batch completes, the system links the same entry, shipment, parties, products, and supporting documents.

Live Draft

Build a live draft while processing

As files finish, the system creates provisional links so business users can see progress.

Batch Barrier

Merge after the batch closes

After processing finishes or the user closes the batch, merge aliases and match key fields.

Evidence Graph

Create a versioned evidence graph

Every relationship has a source and version so later review is traceable.

Entry · Shipment · Party · Product · Code · Document
Process Detail · Phase 4
12 / 20
Deterministic checks before AI-heavy review

Clean the facts first with explicit rules

CHECK 01

Required document check

Check whether the entry has 7501, invoice, packing, B/L, ISF, and IOR-related materials.

required docs
CHECK 02

Code and entity normalization

Normalize HTSUS, Chapter 99, IOR, EIN, port, manufacturer, and related naming.

normalization
CHECK 03

Shape the fact layer

Keep clean facts, conflicts, and missing evidence for later compliance agents.

verified fact layer
Process Detail · Phase 5
13 / 20
Compliance agent fan-out

Then specialized agents check in parallel

Verified fact layer

Distribute by entry / product / IOR scope

parallel fan-out ↓
Account / Payment

ACE、5106、TAO、ACH、Notify Party

Origin

Country-of-origin and manufacturer evidence

HTS Review

Classification consistency and supporting documents

Valuation

Invoice, declared value, payment, and discounts

Exemption Docs

Special tariff and exemption support documents

dependency checks ↓
CAPE readiness

ready / candidate / blocked

Tariff class

Chapter 99, IEEPA, MPF, HMF, AD/CVD

Assists

Buyer-provided tooling, design, materials, and related signals

First sale

Multi-tier transaction and first-sale document chain

Findings queue

All check results flow into a unified findings queue

Process Detail · Phase 6
14 / 20
Findings, human review, and re-run loop

Turn findings into follow-up tasks

Findings

Conflicts, gaps, low-confidence items, and blockers from Phase 5

Normalize

Normalize status, severity, entry scope, and suggested owner

Human review

Assign to importer, purchaser, IOR, broker, or lawyer

Questionnaire

Ask only for facts the files cannot answer, and request supporting materials

review accepted ↓
missing facts ↓
Ready for package

Reviewed, supplemented, or accepted as an open reservation

Re-run affected checks

After supplementation, rerun only affected scope, then return to Normalize / Human review

User / Broker / Lawyer

Add facts, upload files, or accept the open issue

Process Detail · Phase 7
15 / 20
Final package generation

Generate a deliverable compliance work package

Reviewed state

Reviewed, supplemented, and reservations recorded

↓ package generation
Audit Binder Agent

Organize evidence package, manifest, source files, and exception notes by entry

Export Agent

Export check results into business-ready tables and lists

Compliance package

audit binder + evidence index + worklists

Business outputs

CAPE CSV, account checklist, code table, risk workbook

deliverable handoff ↓
Satisfied output

Importers, IORs, brokers, compliance managers, and counsel can keep working from it

Verification Scope
16 / 20
What the AI agents check

12 verification domains, all tied back to evidence

01

CAPE readiness

02

Account / payment

03

Import codes

04

HTS classification

05

Tariff classes

06

Valuation basis

07

Country of origin

08

Assists signals

09

First sale packet

10

Exemption docs

11

Audit binder

12

Human review

12
Verification domains

Each verification domain leaves an evidence status, risk finding, owner, and next step.

AI Boundary
17 / 20
What AI should and should not do

The system organizes evidence first; professionals make the judgment

— 01 / CLASSIFY

Organize facts

Classify files, extract fields, and preserve source and confidence.

— 02 / COMPARE

Find conflicts

Compare invoice vs 7501, HBL vs ISF, IOR vs EIN.

— 03 / ASK

Generate questions

When files are insufficient, the system generates follow-up questions.

— 04 / REVIEW

Hand off for confirmation

Broker, compliance manager, or lawyer confirms HTS, origin, valuation, and first sale.

MVP Roadmap
18 / 20
Build from the smallest valuable loop

MVP starts with one Entry evidence package

01
Phase 1

Evidence chain + CAPE readiness + account/payment readiness + audit binder.

02
Phase 2

HTS, tariff, origin, valuation, party / supply-chain verification.

03
Phase 3

Assists, first sale, exemption documentation, product master history.

04
Phase 4

Continuous compliance database: continuously import broker / ERP / ACE data and monitor historical changes.

Product Requirements
19 / 20
What the first product must prove

The final deliverable must be usable by the team

Import / Export
Compliance
Verification
Package

A deliverable that leadership, compliance teams, brokers, and counsel can continue working from.

Artifacts
10

Output structures such as summary, index, worklist, review tables, and binder.

Core objects
14

From Batch and Document through Finding, Questionnaire, and AuditBinder.

Agent modules
15

Classification, extraction, graphing, domain checks, audit package, and human review.

Ready /
Candidate /
Blocked

The core state machine revolves around evidence: what it supports, what is missing, who reviews it, and what comes next.

20 / 20
CLOSING
ANSWER
Build
evidence.
Let people
decide.
Tari · Compliance Verification
END
TAKEAWAYS
03 RULES
  • 01

    Start with an entry-level evidence package

    The MVP does not try to cover every legal issue; first make one entry evidence chain work end to end.

  • 02

    Put CAPE inside a broader verification framework

    Refund is the business entry point; the long-term value is a continuously compounding compliance database.

  • 03

    Every AI conclusion must be traceable

    A finding with no source, no owner, and no next step should not enter the main product workflow.

SCQA · Situation / Complication / Question / Answer