Main site

Retail AI agent · Import evidence

Forced-Labour Import Evidence AI Agent

Map each product to its raw inputs before the order is placed, register each document with where it came from, and hold the pack for the named person who signs the importer statement.

Typical delivery
4–6 weeks
Deployment
Your stack
Named signer
Pre-import
After launch
Agent Care

What this agent does

Builds the pack, never signs the statement

In

  1. 01

    An order is raised, and the agent opens a chain map for that product from the finished good to the raw input.

  2. 02

    A tier answers, and the named party, the site and the source of that answer are carried forward.

Reason

  1. 03

    A tier goes unanswered, and the gap is written into the pack as a named absence against that supplier.

  2. 04

    A document arrives, and its issuing entity, translation status and redaction status are recorded with it.

  3. 05

    A shipment is booked, and the map re-screens against the UFLPA Entity List as published, and the screen is dated.

Decide

  1. 06

    A document comes from inside the region, and the pack marks that on its face instead of netting it away.

  2. 07

    An article is imported, and the three-month proof clock in 19 CFR 12.43 runs on the file from that date.

Out

  1. 08

    An exclusion advice is delivered, and the sixty-day abandonment clock in 19 CFR 12.44 opens against it.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The upstream chain assembles evidence; the retailer is the downstream party that certifies, and a named person signs, not the agent.

Example workflow

One product, order to pack

  1. 1Order raisedPurchase order, vendor record, bill of materials or import file
  2. 2Chain mappedA named party at every step from the finished good to the raw input, or an explicit gap
  3. 3Documents registeredPresent, absent, translated, redacted, and where each one came from
  4. 4Controls appliedEntity List re-screen, provenance marking, translation and redaction checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is filed at any of them — the agent is assembling, and the human lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the trade compliance lead.

Low confidence

Adds a customs counsel read first.

Trade compliance review

The pack is held with its named gaps, its provenance marks and the confidence.

Approve · Amend · Send to counsel

Approved — released to the signer

  1. 6Import file updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedTier coverage, named gaps, provenance marks and post-entry corrections

Amendments

Reviewer amendments are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Opining that evidence is clear and convincing.
    • Scoring a pack against a statutory evidentiary standard.
    • Treating a redacted or unverifiable document as evidence.
    • Promising that a detained shipment will be released.

  • Automation boundary

    Agent acts unaided

    • ✓Map each product to its raw inputs, with a named party per step.
    • ✓Register what is present and absent, per supplier and per tier.
    • ✓Mark where each document came from, on the face of the pack.
    • ✓Run the clocks, and retain the pack under 19 U.S.C. 1508.

    Completeness is a fact about the file. Sufficiency is the Commissioner's, the signature a person's.

    • Signing the certificate of origin or the consignee statement.
    • Submitting, responding, protesting or disclosing to CBP.
    • Choosing between an applicability and an exception position.
  • Changes to the checklist, screening or provenance rules.

Example output

One tier in the pack, annotated

What the agent registers is attached to the document it came from.

Import-evidence output · single tier

Illustrative example
Tier
Tier 4 · spinning
What the pack shows
Spinner named and dated; the cotton behind it is not yet evidenced
Input
Combed cotton
Document on file
Production record
Confidence
88%
Provenance
Issued inside the region

As received

Taken from the supplier response and the document register — nothing on this side is asserted by the agent.

Evidence registered

  • Production record
  • Purchase-order trail
  • Transport document

Why this is marked

The document exists and is legible; where it was issued is what the signer weighs, and the pack shows it.

What the score decides

Below the configured threshold the pack picks up a counsel read before it reaches the signer.

Action

ApproveAmendSend to counsel

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every product line

From the order record

03Assembly

Build the pack from the order

Draw on the documents on file and the configured checklist. The catalogue agent left the origin field; this is its evidence.

01Approved path

Completeness is not credibility

A complete tier is a count against a document list, not a finding about the world. The pack shows both.

02Human review

Send the read to the marked tiers

The Xinjiang Supply Chain Business Advisory of 13 July 2021 says audits in the region may not be a credible source; those tiers stay marked.

04

Build an evidence trail

The tier, the document behind it and the person who signed stay on the pack.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Sourcing and purchase orders

    • SAP Ariba
    • Coupa
    • Oracle
    • NetSuite POs

  • Supply-chain traceability

    • TrusTrace
    • Sourcemap
    • Retraced
    • Textile Genesis

  • Supplier documents

    • Sedex
    • Higg
    • EcoVadis
    • Vendor portals
    • SFTP

Agent

Forced-labour import evidence

  • Maps the chain
  • Registers documents
  • Holds for the signer

  • Customs and brokerage

    • Descartes
    • e2open
    • Broker platforms
    • entry data

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the pack

Six screens tightening inward. What survives them all is named in the map beneath this.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull the agent back to register-only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, checklist and screening-configuration changes.Track
L4TraceabilityRecord the source, the document, the provenance mark, the gap and what the signer was shown.Record
L3Reviewer approvalHold packs for a named reviewer; it governs release, not whether the documents are true.Gate
L2Policy guardrailsTest the pack against the checklist, provenance and translation rules; a failure returns it.Restrict
L1Confidence thresholdsRoute low-confidence packs to a counsel read before the reviewer sees them.Require review
Model corePack assembled — tier map, document register, provenance marks and confidence
L1 – L2
Test whether a pack may stand
L3
Puts the release in a reviewer's hands
L4 – L5
Keep the tier and the document behind it
L6
Narrows to pack assembly when signals degrade

How Nestack evaluates it

Evaluate the whole assembly — not only the finished pack.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the pack the reviewer reads

  1. E1Final-output evaluationDid each registered document match the source it is filed against?
  2. E2Step-level evaluationDid the agent use the right checklist, list publication and supplier set?
  3. E3Tool evaluationDid it read and write the correct product and the correct tier?
  4. E4Confidence calibrationDo low-confidence packs actually attract more reviewer amendments?
  5. E5Slice evaluationHow does performance change across specific input and tier types?
  6. E6Business outcomeHow many packs needed an amendment or a correction after entry?

Floor — the outcome the importer answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the lifecycle stage where each begins.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • GT-03

      Stale list publication

      The map is screened against a superseded Entity List.

    Stage gathers

    Orders, supplier responses, documents and the list

  2. 02 · Reasoning

    2 modes

    • GT-04

      Provenance netted away

      A marked document is counted as present without its mark.

    • GT-06

      Sector checklist gap

      The template omits a document the sector guidance names.

    Stage proposes

    Tier map, document register, marks and confidence

  3. 03 · Tool / write

    2 modes

    • GT-02

      Unverified clock quoted

      A relayed response window is shown as a settled rule.

    • GT-05

      Two arguments merged

      Applicability and exception evidence land in one pack.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • GT-01

      Tier presented as complete

      An unanswered sub-tier is not shown as a named gap.

    Stage returns

    The pack the reviewer reads and the signer signs

  5. 05 · Change / Version

    1 mode

    • GT-07

      Silent checklist regression

      A model or rule change loosens what counts as present.

    Stage tracks

    Model, prompt, checklist and screening config

  • Sev-1 · agent acts outside the boundary
  • Sev-2 · a wrong tier reaches the signer
  • Sev-3 · source degrades, pack routes to review
See the complete Retail & E-commerce failure-mode catalogue

Affected slices

A healthy total can hide one raw input

A programme-level tier-coverage figure can read well while a few raw inputs carry nearly all of the unevidenced tiers. Nestack reports coverage by input, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Cotton and yarn tiers9.4%3.7×Review
Polysilicon and cell tiers6.8%2.7×Review
Aluminium and casting tiers4.6%1.8×Watch
Single-tier finished goods2.2%0.9×Normal

Bar: unevidenced-tier lift vs. single-tier baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

No cycle closes without a new case

A cycle is done when the unevidenced tier has become a case the next release must pass. That suite is what the next pack built is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    The unevidenced-tier rate rises for one raw input.

  2. 02

    Diagnose

    The mill two steps past the supplier who answered is read back until the cause narrows to one.

  3. 03

    Improve

    Number the change; the shipments that drove it are filed under it.

  4. 04

    Verify

    Each touched tier case is run once more, and one red holds it back.

  5. 05

    Learn

    It stays as a standing test, and the provenance rules move with it.

  6. Learn → Detect

    The return edge. Every later detection meets a suite that grew by one case.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, pack assembly, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Import workflow discovery and boundary definition.Week 1
  2. 02Sourcing, PO and traceability assessment.Week 1, Week 2
  3. 03Sector document checklist and provenance mapping.Week 1, Week 2
  4. 04Order and supplier-response ingestion.Week 2
  5. 05Chain mapping and document registration.Week 2, Week 3
  6. 06Confidence scoring and gap routing.Week 3
  7. 07Reviewer approval workflow.Week 3, Week 4
  8. 08Purchase-order and broker integration.Week 5
  9. 09Tier and provenance cases.Week 4, Week 5
  10. 10Guardrails and submission controls.Week 4
  11. 11Pack-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne product lineProductionProduction import systemsAdvancedMultiple categories / entities
Introduced at Pilot
Mapping to your orders and tiers✓✓✓
Reviewer approval✓✓✓
Tier-coverage baseline✓✓✓
Introduced at Production
Reporting by input—✓✓
Approval workflow in your systems—✓✓
Approved write-back—✓✓
Purchase-order integration—✓✓
Introduced at Advanced
Multi-sector checklist rules——✓
Multi-stage trade approvals——✓
High import volume——✓
Multi-tier evidence controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your purchase-order and vendor record structure

    Order ingestion and supplier mapping

    Week 1

  2. 02Representative supplier responses and documents

    Chain-mapping baseline, document registration and provenance marking

    Week 2

  3. 03Your sector document checklist and approved sources

    Sector document checklist and provenance-rule mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Sourcing, PO and traceability assessment, then integration setup

    Week 2

  5. 05Packs you would not want detained

    Document-provenance cases and failure testing

    Week 4

  6. 06What no pack may pass off as proof

    Confidence scoring, gap routing, guardrails and approval controls

    Week 3

  7. 07A named trade compliance lead to review packs

    Reviewer approval workflow, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The chart is laid out in worked weeks, not in slide inches, and that is why the fifth one doubles.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Import workflow discovery, checklist mapping and the automation boundary

W2Source integration and the chain-mapping baseline

W3Pack assembly, confidence logic and approval controls

W4Evaluation suite, provenance checks and failure-mode testing

W5Broker integration, pilot packs and targeted corrections

W6One import season run under the trade compliance lead, then Agent Care handover

Reading the band
Each bar sits only on the weeks its work is named in. The week 5 overlap is real, not padding.
At the end of W6
When the pack validates, Agent Care takes the agent on.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Retail AI agent

Build an import-evidence agent around your sourcing calendar.

Show us your purchase orders, your tier map and who signs the consignee statement. The pack has to survive an officer reading it cold, which is why it is built before the order ships — this is pre-import prevention, not detention rescue.

Typical delivery
4–6 weeks
Deployment
Your PO and vendor data
Named signer
Pre-import
After launch
Agent Care available

Nestack Agents · Forced-labour import evidence

AGT-RT-18 · Agent Care available after launch

Build an import-evidence agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your imports

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.