Kasynets AI Solutions

Study · synthetic paired comparison

How warehouse backlog forms — and how aisle congestion affects picking

One paired comparison from the existing CoupledSimulation model: identical synthetic work, identical pickers, one difference. In the second run the front of sorter row 39 is blocked for the whole configured hour. We stop both runs at 1800 seconds and read the counters.

Author: Kenan Muratović · Reviewed 2026-11-17 · Evidence: picking-study.json

Scope. Synthetic, deterministic, one paired run. Not Monte Carlo, not client data, not a return-on-investment claim. The half-hour observation window is not a whole shift or a day.

The setup

60 original jobs were released, containing 4,025 location lines, worked by 30 pickers on a one-hour configured shift. The scenario is clustered-strict, with zero interrupts. The only change between runs is blockedSeconds: 0 in the open run, 3600 in the blocked run, applied to the front of sorter row 39.

A location line is one source/SKU confirmation: a picker confirms quantity at one location for one SKU. It is not a job and not a piece.

Key inputs

Inputs as recorded in the evidence JSON
ParameterValue
jobs60
interruptCount0
wmsHours1
blockedZonesorter
blockedRow39
blockedSectionfront

Model defaults

The comparison changes only aisle blockage. Every other resolved default is shared, including requested vehicle speed, safety clearance, STOP dwell, pick timing, mixed packaging and finite handover resources.

Inspect the complete resolved baseline configuration
Full baseline configuration reproduced by the existing model
ParameterValue
wmsHours1
jobs60
controlStudyfalse
controllerModelegacy
twinCongestionSeconds90
twinFairnessSeconds900
twinSlaSeconds7,200
allocationStudyfalse
allocationPolicyrestricted
allocationScopewarehouse
priorityPickerIds["PK01","PK02","PK03","PK04","PK05","PK06","PK07","PK21","PK22","PK23"]
allocationWaveCount8
allocationJobsPerWave90
allocationGoodPerWave30
allocationCardsPerWave30
allocationWaveGapSeconds3,600
binCaseProfilestandard
binCaseConsolidatefalse
binCaseCartons25
binOpenSeconds3
binCutSeconds12
interruptCount0
policymanual
voiceVerify3
scannerVerify3
voiceRetryRate0
scannerRetryRate0
retryDelay5
quantityConfirm2
pickDwell8
pickPerUnit1
outerPackSize12
catalogModemixed
swapChannelsfalse
clearance3.5
stopDwell1.5
requestedSpeed6
burstSorter12
burstClassic8
reachSorter6
reachClassic4
reachJobs3
liftSeconds40
jobChangeSeconds30
recoveryDelaySeconds120
recoveryHandoverSeconds60
blockedZonesorter
blockedRow39
blockedSectionfront
blockedSeconds0
deliveryRows[1,2,3,4,5,6,7,8]
deliveryRowsProvisionaltrue
laneCapacity4
sorterPlaceSeconds30
sorterClearSeconds300
sorterCarriersPerTask1
binDeliveryToSortertrue
binHandoverCapacity4
outerMaxLocations25
binMaxLocations100
classicMaxLocations100
classicMaxPieces200
wrapperBufferCapacity4
wrapperMachines1
wrapperHandoverSeconds30
wrapperFeedSeconds15
wrapSeconds90
wrapperTransferSeconds20
loadingBufferCapacity4
loadingBays1
truckLoadingSeconds45
truckAvailableAt0
truckCapacityCarriers20
truckChangeSeconds300

What we observed at 1800 seconds

Reproduced synthetic results at 1800 seconds elapsed
MetricOpen aisleBlocked aisle
blockedSeconds03,600
releasedJobs6060
releasedLines4,0254,025
confirmedLines794654
backlogLines3,2313,371
confirmedBaseUnits9,8908,977
pickerWaitingSeconds32,840.3635,412.8
completedOriginalJobs77
loadedBaseUnits00
teamBaseUnitUPH19,78017,954

In the open run, 794 lines were confirmed and 3,231 remained in backlog, with 9,890 confirmed base units and 32,840.36 summed picker waiting seconds. With the aisle blocked: 654 confirmed, 3,371 backlog, 8,977 base units and 35,412.8 waiting seconds.

Both runs completed 7 original jobs and loaded 0 base units. Those are deliberately reported: half an hour in, almost nothing is finished or shipped in either case, so the confirmed-unit difference must not be read as shipped output.

Definitions and formulas

  • Backlog lines = released lines − confirmed lines. Open: 4,025 − 794 = 3,231. Blocked: 4,025 − 654 = 3,371.
  • Base-unit UPH per team = confirmed base units ÷ 0.5 h. Open: 9,890 ÷ 0.5 = 19,780. Blocked: 8,977 ÷ 0.5 = 17,954. This is a whole-team rate over the elapsed half-hour. It is not per picker, not jobs per hour and not a daily projection.
  • Summed waiting is person-seconds added across all 30 pickers, not wall-clock seconds.
  • Base units versus packages. The existing catalog uses a mixed pack mapping, so package counts are not interchangeable with pieces. Source quantities are preserved as recorded.

A separate, high-level day model

The Warehouse Digital Twin also contains a coarse demand-and-capacity queue for a day. Its identity is: opening backlog 14,200 + demand 450,000 − throughput 300,000 = ending backlog 164,200, a growth of 150,000. Capacity coverage is 66.7% while utilization is 100%: the team is fully busy and still cannot keep up.

That model counts a different thing on a different time scale. Do not equate its backlog with the line backlog in the detailed study above.

How the model works

The workload is generated deterministically from synthetic inputs; there are no independent stochastic repetitions. The discrete-event solver decides when each picker moves, waits and confirms. The 3D view in the 3D Picking Lab visualizes solver state. It does not determine throughput.

Reproduce it

  1. Open the detailed lab at 3D Picking Lab.
  2. Download the evidence JSON. Each resolved entry contains the complete configuration; scenarioDefinition identifies the dispatch and rules. This evidence file is not an importable lab preset.
  3. Using the supplied v21 model module, construct new CoupledSimulation(configuration, scenarioDefinition), run advanceTo(1800), and read summary(). Repeat with the second resolved configuration. The repository verification command is node artifacts/kasynets-ai-solutions/scripts/verify-picking-study.mjs.

Download the full configuration (JSON)

Limitations

  • Synthetic inputs only; no live WMS connection.
  • No client results, no ROI estimate and no certification.
  • Any operational decision requires pilot validation on real data.
  • Observed over an elapsed half-hour, not a whole shift or day.

Related: the warehouse simulation film and the warehouse automation overview.