Main site

Transportation AI agent · Yard and gate

Yard & Gate Computer-Vision AI Agent

Read container codes, plates and visible damage at the gate, check what arithmetic can check, and leave the identity, the interchange receipt and the movement to a gate officer.

Typical delivery
4–6 weeks
Deployment
Your stack
Officer decides
Read, not ID
After launch
Agent Care

What this agent does

Reads the characters, never the person

In

  1. 01

    Where the lanes are configured, take frames from supported OCR portals and yard cameras.

  2. 02

    When a container code is read, recompute the ISO 6346 check digit before the read is carried anywhere.

Reason

  1. 03

    Where the check digit passes, say only what it proved — that the characters are internally consistent.

  2. 04

    When the misread is a letter collision, such as A for K or U, the check digit is blind to it, so the read is marked.

  3. 05

    Where a licence plate is read, carry it as the weaker signal it is, because a plate has no check character at all.

Decide

  1. 06

    Where the model would become face-shaped it stops: codes, plates and seals are read; faces, gait and emotion are not.

  2. 07

    When damage is visible, capture it as evidence towards the interchange receipt, never as a grading or a charge.

Out

  1. 08

    When a read falls below the threshold, route the movement to the gate officer with the frame.

  2. 09

    Where write actions run, they run only inside the approval boundaries agreed during implementation.

Product statement

The agent reads and checks; a gate officer decides what the equipment is, what the receipt says and whether the truck moves.

Example workflow

One gate movement, frame to release

  1. 1Frame capturedOCR portal, gate kiosk camera, yard camera or handheld capture
  2. 2Read attemptedContainer code, chassis and plate characters, seal number and visible condition, each with its frame
  3. 3Checks runRead, check-digit result and confidence
  4. 4Controls appliedCheck-digit and structural checks, appointment cross-reference, plate-assurance marking and the confidence threshold

No human action required

Stages 1 to 4 run unaided and open no gate — the read is prepared, and the officer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the gate officer to accept.

Low confidence

Adds a second frame and a lane read.

Gate officer decides

The read is held with its frame, its check result and the confidence.

Accept · Correct · Send to the lane

Accepted — movement recorded

  1. 6Yard systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedRead corrections, exception volume, damage disputes and movements reworked afterwards

Corrections

Every officer correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Deciding that a read is a particular container.
    • Chocking, restraining or securing any vehicle.
    • Releasing a truck through the gate on a read alone.
    • Signing or closing an equipment interchange receipt.

  • Automation boundary

    Agent acts unaided

    • ✓Read container, chassis and plate characters from the frame.
    • ✓Recompute the ISO 6346 check digit and report what it proved.
    • ✓Cross-reference a read against the appointment and.
    • ✓Capture visible damage as evidence and hold it for.

    Any write happens inside the boundaries agreed at implementation, never ahead of the officer.

    • Grading damage or deciding who is charged for it.
    • Matching a face, a body or a gait to a person.
    • Setting or lifting a retention or litigation hold.
  • Changes to gate rules, thresholds or access controls.

Example output

One gate read, annotated

Everything the agent reads is attached to the frame it was read from.

Gate read output · single movement

Illustrative example
Movement
Inbound container
Characters read
Owner code, category identifier and serial, as read off the box
Check digit
Recomputed, consistent
Frame reference
Lane 3 camera, frame 2
Confidence
93%
Status
Held for the gate officer

As received

Taken from the frame as captured and the arithmetic run over it — nothing on this side is an identification.

What the check proves

  • The characters are consistent
  • Position four is valid
  • Serial substitutions covered

Why it is not an identity

Letter collisions such as A, K and U clear the arithmetic untouched.

What the score decides

Below the configured threshold the read picks up a second frame and a lane read before.

Action

AcceptCorrectSend to the lane

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 movement

From the gate lane or the yard camera

03Reading

Read it, check the arithmetic

Draw on the frame, the ISO 6346 check digit, the appointment on file and the plate and retention rules configured for the site.

01Approved path

A read is not an identity

Routine movements come back read, with the check result attached.

02Human review

Send the rest to the officer

Failed checks, plate reads and low-confidence characters are marked — and a camera that sees a hazard reports it, it does not remove it.

04

Build an evidence trail

The read, the frame it came from and the gate officer who accepted it stay on the movement record.

Integrations

Typical integrations

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

  • Gate and OCR hardware

    • OCR portals
    • gate kiosks
    • Fixed cameras
    • edge devices

  • Terminal and yard systems

    • Navis N4
    • Tideworks
    • Yard management
    • appointments

  • Equipment and interchange

    • Interchange receipt records
    • Chassis and equipment registries

Agent

Yard and gate computer vision

  • Reads the frame
  • Checks the code
  • Holds for the officer

  • Access and evidence

    • Gate access controls
    • Video and evidence archives

  • 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 gate

Each control encloses the last. Whatever slips past all of them is named below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeCut the agent back to manual gate entry when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, lane-configuration and retention-rule changes.Track
L4TraceabilityRecord the frame, the read, the check result, the flags and the officer's decision.Record
L3Officer approvalHold reads for the gate officer; it governs the movement, not whether the characters were right.Gate
L2Policy guardrailsTest reads against the check digit, the structural rules and the appointment; a failure returns the read.Restrict
L1Confidence thresholdsRoute low-confidence reads to a second frame and a lane read before the officer sees them.Require review
Model coreRead produced — characters, check result, damage evidence and confidence
L1 – L2
Test whether a read may stand
L3
Puts the movement in an officer's hands
L4 – L5
Keep the read and the frame it came from
L6
Cuts back to manual gate entry when signals degrade

How Nestack evaluates it

Evaluate the reading workflow — not only the characters returned.

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

Surface — the read the yard acts on

  1. E1Final-output evaluationDid the characters match the frame they were read from?
  2. E2Step-level evaluationDid the agent run the check digit and the structural rules it should have?
  3. E3Tool evaluationDid it read and write the correct movement and the correct equipment?
  4. E4Confidence calibrationDo low-confidence reads actually attract more officer corrections?
  5. E5Slice evaluationHow does performance change across specific lanes and conditions?
  6. E6Business outcomeHow many reads were corrected, or disputed after the movement closed?

Floor — the outcome the terminal answers for

Failure modes

Where each failure originates in the agent

Seven ways a gate movement goes wrong, by stage.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • BE-03

      Fouled or occluded surface

      Rust, snow or lashing hides part of the number.

    Stage gathers

    Frames, container codes, plates, seals and appointments

  2. 02 · Reasoning

    2 modes

    • BE-04

      Letter collision passed

      A substitution inside a residue class clears the check.

    • BE-06

      Read taken as identity

      A character string is treated as the equipment itself.

    Stage proposes

    Read, check-digit result and confidence

  3. 03 · Tool / write

    2 modes

    • BE-02

      Plate read over-trusted

      A plate is acted on as though it could be checked.

    • BE-05

      Damage missed on capture

      A visible defect never reaches the interchange receipt.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • BE-01

      Frame reference lost

      A read arrives without the frame behind it.

    Stage returns

    The read the officer accepts and the yard acts on

  5. 05 · Change / Version

    1 mode

    • BE-07

      Silent lane regression

      A model or lane change widens what the agent asserts from a frame.

    Stage tracks

    Model, prompt, lane config and retention rules

  • Sev-1 · a movement clears outside the boundary
  • Sev-2 · a wrong read reaches the yard record
  • Sev-3 · conditions degrade, read routes to review
See the complete Transportation failure-mode catalogue

Affected slices

One lane can carry most of the corrections

In aggregate the correction rate reads low. By slice it does not — night and low-sun lanes run at a multiple of the daylight rate on the same cameras. Nestack reports corrections by lane and condition, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Night and low-sun lanes8.2%3.9×Review
Rusted or overpainted boxes5.0%2.4×Review
Licence-plate reads3.2%1.5×Watch
Daylight portal lanes1.7%0.8×Normal

Bar: correction-rate lift vs. daylight-portal baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The loop ends in a case, not a post-mortem

The cycle does not end in a post-mortem. It ends in a case the next release has to clear, and that suite is what the next gate movement is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Officer corrections rise in one lane.

  2. 02

    Diagnose

    The frame, the read and the movement record go back on the desk together until the cause narrows to one.

  3. 03

    Improve

    Each change carries a version and the movements that forced it.

  4. 04

    Verify

    Release waits on the affected cases going green again.

  5. 05

    Learn

    The case is kept for good, and the gate rules are rewritten.

  6. Learn → Detect

    The return edge. The next read meets a suite that just got longer.

Typical build scope

Twelve workstreams across six weeks

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

Workstream

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

  1. 01Gate and yard workflow discovery and boundary definition.Week 1
  2. 02Camera, lane and terminal system review.Week 1, Week 2
  3. 03Check-digit, structural and appointment rule mapping.Week 1, Week 2
  4. 04Frame ingestion and read normalisation.Week 2
  5. 05Character extraction and check logic.Week 2, Week 3
  6. 06Confidence scoring and exception routing.Week 3
  7. 07Gate officer workflow.Week 3, Week 4
  8. 08Yard and terminal-system integration.Week 5
  9. 09Misread and retention cases.Week 4, Week 5
  10. 10Guardrails and access controls.Week 4
  11. 11Movement-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 gate, one laneProductionProduction yard systemsAdvancedMultiple sites / gates
Introduced at Pilot
Reading to your gate rules✓✓✓
Gate officer approval✓✓✓
Read-accuracy baseline✓✓✓
Introduced at Production
Reporting by gate and lane—✓✓
Officer workflow in your systems—✓✓
Approved movement write-back—✓✓
Yard-system integration—✓✓
Introduced at Advanced
Multi-site retention and access rules——✓
Multi-stage gate approvals——✓
High gate volume——✓
Multi-site gate 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 gate layout, lanes and camera positions

    Camera, lane and read-condition mapping

    Week 1

  2. 02Representative frames, including the bad ones

    Read baseline, character extraction and check logic

    Week 2

  3. 03Your retention schedule and who may view footage

    Retention-rule mapping and automation-boundary definition

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Camera, lane and terminal assessment, then integration setup

    Week 2

  5. 05Reads you would not want acted on

    Misread cases and failure-mode testing

    Week 4

  6. 06What a read may never authorise on its own

    Confidence scoring, exception routing, guardrails, retention and access controls

    Week 3

  7. 07Named gate officers to accept reads

    Gate officer 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 bands sit on the weeks the work occupies, which is why the fifth carries two kinds at once.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Gate and yard workflow discovery, lane rules and the automation boundary

W2Camera and terminal integration and the read baseline

W3Read workflow, confidence logic and officer controls

W4Evaluation suite, misread cases and failure-mode testing

W5Yard-system integration, pilot lanes and targeted corrections

W6One gate cycle run under the yard team, then Agent Care handover

Reading the band
Every bar covers the weeks its own work runs and no others. The fifth week genuinely doubles up.
At the end of W6
The cycle closes the validation and Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Transportation AI agent

Build a gate vision agent around who decides at the lane.

A wrong read releases the wrong box, and a missed defect writes a clean interchange receipt over real damage. Show us your gates and we'll name what stays with the officer.

Typical delivery
4–6 weeks
Deployment
Your gates and yard systems
Officer decides
Read, not ID
After launch
Agent Care available

Nestack Agents · Yard and gate computer vision

AGT-TR-16 · Agent Care available after launch

Build a gate vision agent

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

About you
Your gates

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