Solution design · all figures are example or default values

Parcel Hub WMS + WCS + Digital Twin: Solution Design

A full design for a platform that plugs into every hub's existing sorters, systems and cameras with minimal load, mirrors each floor as a live digital twin, tracks every parcel, bag, person and machine from gate-in to dispatch, and lets you configure WMS and WCS logic safely. All numbers in this document are example or default values, configurable per site, not real data.

1. Executive overview and goals

The platform sits on top of what each hub already runs. It reads events from sorters, scanners, DWS, handhelds, cameras and existing WMS and TMS without loading them, builds one live model of the floor, and turns it into control: who should be where, which chute serves which destination, and what each parcel category costs in time and money.

Goal	What it means	Example target
Visibility	Every vehicle, bag, parcel, person and machine located and timed at every stage	Scan-event coverage of 99.5% or more of parcels at every mandatory stage
Throughput	Raise effective site TPT without new hardware by removing bottlenecks	Effective PPH up 10 to 20% versus baseline
Turnaround	Cut TOT inbound-to-dispatch per category, hit every connection cutoff	Connection adherence 98% or more
Workforce	Right person at the right station at the right hour	People utilisation 80 to 88%, idle under 10%
Quality	Fewer missorts, no-reads, damages and shortages	Sort accuracy 99.9% or more, no-read under 1%
Security	Camera-based detection of pilferage, intrusion and rough handling	Every HV-cage incident with a video clip in under 60 s
Cost	Lower cost per parcel by stage and by site	Cost per parcel down 8 to 15%
Control	One studio to configure WMS process logic and WCS device logic, tested on the twin first	Zero untested config pushed to a live sorter
Low intrusion	Integrate with existing sorters, systems and cameras without loading them	Under 2% added CPU or IO on any source system

Design principle. Observe first, then advise, then control. Every site starts read-only; WCS control is switched on per site only after shadow-mode certification.

2. System architecture

Layer	Responsibilities	Typical technology	Note
L0 Site systems	Existing sorter PLC and WCS, scan tunnels, DWS, handhelds, legacy WMS and TMS, HRMS, badge readers, CCTV	Customer-owned	Untouched; read-first
L1 Edge and integration	Protocol adapters, CDC agents, file watchers, edge AI for video, local store-and-forward, time sync, device registry	Edge gateway plus GPU edge box on site	Survives 72 h WAN outage
L2 Event backbone	Durable ordered log, schema registry, replay, per-tenant isolation	Kafka, Redpanda or Pulsar	Exactly-once into state
L3 Stream processing	Normalise, dedupe, join scan to shipment, fuse vision with scans, sessionise people, rolling KPIs	Flink, Kafka Streams	p95 under 2 s end to end
L4 Twin engine	Live state of every object, layout graph, history, replay, discrete-event simulation	In-memory state store plus time-series DB	Floor refresh 1 s
L5 WMS core	Shipment lifecycle, bag and manifest, exceptions, RTO, cage, DG, workforce	Domain microservices, Postgres	Multi-tenant
L6 WCS orchestration	Sort-plan compiler, chute mapping, device command API, manual fallback	Deterministic service plus edge executor	Command ack under 100 ms at edge
L7 Analytics and simulation	Lakehouse, forecasting, optimiser, what-if, ML for ETA and anomaly	Iceberg or Delta, ClickHouse, Python	Hourly retrain where needed
L8 Config studio	No-code editors, version control, sandbox on twin, approvals, staged rollout	Web app plus config service	Git-like history
L9 Experience	Control tower, site and shift dashboards, floor tablet, handheld app, client portal, public APIs	Web, Android, REST and GraphQL, webhooks	Role-based

3. Integration layer

Connector catalogue

Source	Protocols	Low-load method	Data obtained
Sorter PLC	OPC-UA subscribe, Siemens S7, Modbus TCP, EtherNet/IP	Subscribe to tags with deadband; no polling of control loops	Read only by default; write only via WCS executor with interlock
Sorter WCS / SCADA	Vendor API, TCP socket telegrams, DB read replica, log tail	Mirror telegrams from a span or secondary port	Sort result, divert confirm, alarms, states
Scan tunnel and DWS	TCP or serial telegram, FTP image drop, vendor SDK	Tap the existing feed; no second trigger	Barcode, L x W x H, weight, image ref, read status
Handhelds	Our Android app, or intent and broadcast capture on existing app	Batch upload every 5 s, offline queue	Scans, user, station, GPS where outdoors
Existing WMS / OMS	Log-based CDC (Debezium) on read replica, outbox table, webhooks	Never query primary; replica or binlog only	Shipments, pre-alerts, manifests, routes
TMS / linehaul	REST API, webhooks, EDI, SFTP CSV	Rate-limited poll with ETag and since-cursor	Trips, vehicles, ETA, schedule, seals
Client / carrier pre-alerts	API, SFTP, email-to-parse	Pull windows aligned to cutoffs	Expected AWBs, weights, COD, DG flags
HRMS and attendance	API, biometric device export, CSV	Shift-start pull plus 15 min delta	Employee, contract type, attendance, cost rate
Badge and BLE / UWB tags (optional)	MQTT	Edge aggregates to 5 s presence per zone	Person location by zone
CCTV / IP cameras	RTSP, ONVIF Profile S and T, NVR re-stream	Pull the NVR sub-stream or a second stream; inference on site; only events and short clips leave	Headcount, activity, parcel sightings, security events (section 10)
IoT sensors	MQTT, LoRaWAN	Report by exception	Chute full, door open, temperature, power
ERP and billing	Batch API, nightly file	Off-peak only	Cost centres, rates, invoices

Load-minimisation strategies

Strategy	How	Effect
Read from replicas and logs	CDC from binlog or WAL, read replicas, never the primary OLTP	Near-zero query load on source
Subscribe, do not poll	OPC-UA monitored items, MQTT, webhooks	Push only on change
Deadband and exception reporting	Analog tags report only on change beyond threshold	Cuts tag traffic 80 to 95%
Adaptive rate limits	Token bucket per source; back off when source latency rises	Source SLA protected
Passive taps	Mirror telegram ports, log tailing, span ports	No added transactions
Video on the edge	Inference next to the cameras; events of a few hundred bytes go to the cloud, not video	WAN use per camera drops from Mbps to kbps
Edge pre-aggregation	Compress, batch, dedupe at site before WAN	WAN bandwidth down
Off-peak backfill	Historical loads scheduled outside sort windows	No peak impact
Circuit breakers	Connector pauses itself if source health degrades	Fail-safe
Shadow mode first	All WCS logic runs in observe-only until certified	No control risk

Latency targets

Path	Target	Applies to
Device event to edge	under 50 ms	Sorter, scanner
Camera frame to vision event	under 500 ms	Edge AI box
Edge to cloud bus (p95)	under 1 s	All events
Event to twin state (p95)	under 2 s	Floor view
Event to KPI tile (p95)	under 5 s	Rolling 15-min KPIs
Sort decision lookup at edge	under 30 ms	Scan to divert window
WCS command to device ack	under 100 ms	Edge executor
Pre-alert to sort-ready	under 60 s	Rules and destination assigned
Alert raise after breach	under 30 s	Push and app

Data contract

Aspect	Rule
Schema	Avro or Protobuf, versioned in registry, backward compatible
Envelope	`event_id`, `event_type`, `schema_version`, `tenant_id`, `site_id`, `occurred_at`, `received_at`, `source`, `idempotency_key`
Identity	Canonical IDs: AWB or parcel ID, bag ID, vehicle and trip ID, user ID, device ID; mapping table for source IDs
Ordering	Partition key by parcel or bag ID; per-device sequence number
Quality checks	Mandatory fields, valid enums, timestamp sanity, referential lookups; failure goes to quarantine
SLA	Per connector: freshness, completeness, error rate, monitored on an integration health board

Failure handling and buffering

Scenario	Behaviour
WAN down	Edge stores events (72 h, disk-backed), keeps local sort-decision cache running, replays in order on reconnect
Edge gateway down	Active-passive pair; sorter keeps running on its own WCS since we are non-blocking
Camera or edge AI box down	Zone marked vision-blind in the twin; workforce and parcel logic fall back to scans and logins only
Source schema change	Schema registry rejects incompatible change, routes to quarantine topic, alerts integration team
Duplicate or late events	Idempotency key per event, event-time processing with 10-min allowed lateness, watermarks
Clock drift	NTP or PTP at edge, cameras included; both device time and receive time stored
Poison message	Dead-letter topic with replay tool
Our cloud down	Site runs independently; edge keeps sort cache and handheld offline mode

4. Site and floor master data

Site

Field	Type / unit	Source	Notes
`site_id`, `tenant_id`, `name`, `code`	string	Setup	Unique per tenant
`site_type`	enum: gateway, hub, mother hub, DC, spoke, RTO centre	Setup	Drives template
`address`, `lat`, `long`, `timezone`	text, deg, IANA	Setup	Default Asia/Kolkata
`operating_hours`, `shift_pattern`	time ranges	Setup	e.g. 3 x 8 h or 2 x 12 h
`built_area`, `floors`, `mezzanine`	sq ft, count	Layout	Per floor
`rated_capacity_pph`, `rated_capacity_daily`	parcels/h, parcels/day	Engineering	Design value
`clients_served`, `service_types`	list	Commercial	Express, surface, B2B, RTO
`connection_schedule`	list of route, cutoff time, vehicle	TMS	Drives priorities
`power_backup_min`, `dock_count_in`, `dock_count_out`	min, count	Facilities
`cost_rates`	INR per hour by role, per sq ft, per kWh	Finance	For cost per parcel

Zone

Field	Type / unit	Source	Notes
`zone_id`, `site_id`, `floor`	string	Layout
`zone_type`	enum: yard, inbound dock, de-bag, inscan, induction, sorter, manual sort, chute and bagging, staging, outbound dock, cage, DG, RTO, exception, damage repair, MHE parking	Layout
`polygon`	list of x,y (m)	CAD import	For twin, heatmaps and camera zones
`area_m2`, `capacity`, `capacity_unit`	m², number, parcels/bags/pallets	Engineering
`access_level`, `owner_team`	role, team	Admin	Secure zones need roles
`max_dwell_min`, `max_headcount`	min, people	Config	Dwell and crowding alerts
`camera_ids`	list	Camera registry	Cameras covering the zone

Dock

Field	Type / unit	Source	Notes
`dock_id`, `door_no`, `direction`	string, enum in/out/both	Layout
`dock_type`	enum: leveller, ground, telescopic conveyor, boom	Facilities	Affects unload PPH
`vehicle_types_allowed`	list: 2W, 3W, LCV, 14 ft, 17 ft, 20 ft, 22 ft, 32 ft SXL, 32 ft MXL	Config
`dock_height_mm`, `has_conveyor`, `conveyor_rate`	mm, bool, parcels/h	Facilities
`assigned_routes`, `priority`	list, rank	Config	Outbound docks
`state`	enum: free, reserved, occupied, unloading, loading, blocked	Live	Twin, confirmed by camera
`x`, `y`, `heading`	m, deg	CAD

Station

Field	Type / unit	Source	Notes
`station_id`, `zone_id`, `station_type`	enum: de-bag table, inscan, DWS, induct, manual sort wall, put wall, bagging, exception desk, label reprint, repack, cage desk	Layout
`positions_max`, `positions_min`	people	Engineering	Headcount bounds
`rated_pph_per_position`	parcels/h	Time study	Standard rate
`skills_required`	list	Config	Matches skills matrix
`devices_bound`	list of device IDs	Device registry
`state`	enum: open, closed, starved, blocked, down	Live

Sorter

Field	Type / unit	Source	Notes
`sorter_id`, `oem`, `model`, `install_year`	string	Asset
`sorter_type`	enum: cross-belt, tilt-tray, sliding shoe, bomb-bay, narrow-belt, pop-up wheel, pusher, AMR robotic, linear	Asset
`loop_length_m`, `speed_mps`, `carrier_count`, `carrier_pitch_mm`	m, m/s, count, mm	OEM
`design_throughput_iph`	items/h	OEM	Ideal rate for OEE
`min_dims`, `max_dims`, `min_wt`, `max_wt`	mm, kg	OEM	Defines conveyable
`induct_count`, `chute_count`, `reject_chute_ids`	count, list	OEM
`max_recirc`	count	Config	Then reject
`scan_tunnel`	sides, cameras, spec read rate	OEM
`plc_protocol`, `plc_endpoint`, `tag_map`	enum, URI, map	Integration
`planned_maint_windows`	time ranges	Maintenance	Excluded from OEE availability

Induct

Field	Type / unit	Source
`induct_id`, `sorter_id`, `induct_type`	enum: manual, semi-auto, auto, robotic	Asset
`rated_rate_iph`, `positions`	items/h, people	OEM
`merge_priority`, `gap_mm`	rank, mm	WCS
`scanner_id`, `dws_id`	IDs	Registry
`state`	enum: running, starved, blocked, stopped, faulted	PLC

Chute

Field	Type / unit	Source	Notes
`chute_id`, `sorter_id`, `chute_no`, `side`	string, int, L/R	OEM
`chute_type`	enum: gravity, roller, spiral, bag hanger, gaylord, cage, tote	Asset
`capacity_parcels`, `full_sensor`	count, bool	Asset	Camera fill estimate where no sensor
`destination_map`	pincode set, route, hub, client, service	Sort plan	Versioned
`bag_rule`	max parcels, max kg, max fill %	Config
`ptl_light_id`	ID	Registry	Put-to-light
`state`	enum: open, full, near-full, disabled, jammed, overflow	Live

Lane / staging

Field	Type / unit	Source
`lane_id`, `lane_type`	enum: route lane, cutoff lane, overflow, bulky staging	Layout
`capacity`, `capacity_unit`	bags, pallets, cages	Engineering
`linked_dock`, `linked_route`, `cutoff_time`	ID, ID, time	Config
`polygon`	x,y	CAD

Cage / secure

Field	Type / unit	Source	Notes
`cage_id`, `cage_type`	enum: high-value, DG, hold, legal hold, damage, lost and found	Layout
`capacity`, `access_roles`, `camera_ids`	count, roles, IDs	Security
`dual_control`	bool	Config	Two-person open and close
`seal_policy`	per bag or per cage	Config
`loiter_threshold_s`	s	Config	Vision alert

Device

Field	Type / unit	Source	Notes
`device_id`, `device_type`	enum: handheld, ring scanner, fixed scanner, scan tunnel, DWS, printer, scale, RFID, PTL, camera, edge AI box, PLC, badge reader, BLE tag	Registry
`make`, `model`, `firmware`, `serial`	string	Asset
`ip`, `mac`, `protocol`	string	Integration
`bound_station`, `bound_user`	ID	Live
`battery_pct`, `last_heartbeat`, `state`	%, timestamp, enum	Live	Alert when stale over 2 min

Camera

Field	Type / unit	Source	Notes
`camera_id`, `nvr_id`, `stream_url`	string, RTSP URI	CCTV registry	Credentials held in vault
`make`, `model`, `lens_mm`, `fov_deg`	string, mm, deg	Asset
`resolution`, `fps_native`, `fps_inference`	px, fps	Config	Inference FPS lower than native
`mount_height_m`, `tilt_deg`, `x`, `y`	m, deg	Survey	Placed in twin
`homography`	3x3 matrix	Calibration	Pixel to floor coordinates
`zones_covered`, `rois`, `tripwires`	polygons, lines	Config	Counting and intrusion lines
`models_enabled`	list	Config	Per camera
`privacy_masks`	polygons	Config	Toilets, rest areas, outside views
`state`	enum: online, offline, obstructed, low-light	Health	Tamper detection

Layout and MHE

Field	Type / unit	Source	Notes
`origin`, `unit`, `floor_plan_file`	x,y, m, DXF, DWG, SVG or IFC	CAD	Twin base
`object_x`, `object_y`, `object_z`, `rotation`, `footprint`	m, deg, polygon	CAD	Per object
`path_graph`	nodes and edges with length m and speed m/s	Engineering	Walking and conveyor travel time
`conveyor_segments`	length m, speed m/s, accumulation	OEM	Transit time
`mhe_assets`	type (pallet jack, BOPT, forklift, roll cage, trolley), count, charge state	Asset	Equipment utilisation

Every object carries x, y and a footprint so the twin can draw it, plus valid_from and valid_to so layout changes are versioned like any config.

5. Process-by-process design

Floor flow

5.1 Gate-in and yard

Pre-arrivalTrip ETA from TMS, expected bags and AWBs from pre-alert
Gate-inVehicle number (ANPR or manual), seal number and photo, driver ID, trip ID
Seal checkSeal intact and matches dispatch record
Yard slotVehicle placed in yard queue with priority score
Dock callDriver called to assigned dock

Scan and vision events

Event	Trigger	Key fields
VEHICLE_GATE_IN	Gate scan or ANPR camera	`vehicle_no`, `trip_id`, `seal_no`, `seal_ok`, `photo_ref`
VEHICLE_YARD_PARKED	Yard slot scan or camera	`yard_slot`
VEHICLE_DOCK_CALLED	System	`dock_id`, `called_at`

Parameters captured

Parameter	Unit	Source
Scheduled vs actual arrival	min	TMS, gate
Seal mismatch flag	bool	Gate
Yard wait (gate-in to dock-on)	min	Events
Vehicle type, fill %	enum, %	TMS
Priority score	0 to 100	Rule: cutoff proximity, volume, client SLA

Resources

1 security guard per gate per shift (example)
ANPR camera or handheld
Yard marshal for 15+ vehicle yards

Rules

Docking priority = earliest outbound connection at risk, then volume, then arrival order
Seal mismatch puts vehicle on hold and requires supervisor

Exceptions

Seal tampered or mismatch
Unscheduled vehicle
Vehicle late beyond threshold

KPIs

Arrival adherence %
Yard dwell min
Gate-in to dock-on min

5.2 Dock and unloading

Dock-onVehicle docked, dock state to occupied (camera confirms)
Seal breakSeal scanned and broken, photo
UnloadBags and loose parcels scanned on exit from vehicle
ReconcileLive count vs pre-alert: short, excess, damaged
CloseUnload complete, vehicle released or reassigned to outbound

Scan and vision events

Event	Trigger	Key fields
VEHICLE_DOCKED	Dock scan or camera	`dock_id`, `vehicle_no`
UNLOAD_START / UNLOAD_END	Handheld or camera activity	`dock_id`, `team_ids`
BAG_INSCAN	Bag tag scan	`bag_id`, `origin`, `weight`, `seal_ok`
LOOSE_PARCEL_INSCAN	AWB scan	`awb`
BAG_SHORT / BAG_EXCESS	Reconcile	`bag_id`, `manifest_id`

Parameters captured

Parameter	Unit	Source
Bags expected vs received	count	Pre-alert, scans
Unload duration	min	Events
Unload rate	bags or parcels per hour	Derived
Bag weight variance vs manifest	kg, %	Scale
Dock occupancy	%	Derived
Crew size	people	Check-in, camera headcount

Resources

Unload crew 2 to 4 per dock (example)
Telescopic conveyor, pallet jack, roll cages
Handheld or ring scanner per crew
Dock camera

Rules

Bulky and loose parcels go straight to manual sort or bulky staging
High-value bag goes to cage, DG bag to DG zone
Weight variance over 5% (example) flags tamper check

Exceptions

Short bag
Excess bag
Wet or torn bag
Seal broken bag
Wrong-hub bag (misroute in)
Rough handling seen on camera

KPIs

Unload PPH
Dock turn time
Shortage and excess rate
Dock utilisation

5.3 De-bagging

Bag open scanBag ID scanned at de-bag table, seal number recorded
Empty bagParcels tipped onto table or feed conveyor (camera detects de-bag action)
Parcel scanEach parcel scanned to bag (or captured by tunnel downstream)
Bag close-outCount vs bag manifest, bag marked empty and sent to bag pool

Scan and vision events

Event	Trigger	Key fields
BAG_OPENED	Handheld or fixed scanner	`bag_id`, `station_id`, `user_id`
PARCEL_DEBAGGED	Scan	`awb`, `bag_id`
VISION_DEBAG_DETECTED	Camera	`station_id`, `parcel_count_est`
BAG_RECONCILED	System	`expected`, `found`, `short_awbs`, `excess_awbs`

Parameters captured

Parameter	Unit	Source
Parcels per bag	count	Manifest
De-bag time per bag	s	Events
De-bag PPH per person	parcels/h	Derived
Bag queue at station	bags	Twin, camera
Short and excess parcels	count	Reconcile

Resources

1 to 2 people per de-bag table (example)
De-bag tables or tipping stations, feed conveyor
Fixed scanner or handheld
Overhead camera per 2 to 4 tables

Rules

Excess parcel gets auto-lookup; if not found it becomes no-data exception
Damaged parcel diverted to damage desk with photo
Camera count differs from bag count by more than 1: recount task

Exceptions

Short parcel in bag
Excess parcel
Damaged or leaking
Unlabelled parcel

KPIs

De-bag PPH
Bag queue time
Parcel-level shortage rate

5.4 Inscan and DWS

InscanParcel scanned into hub custody (handheld or tunnel); camera at profiler gives auto inscan status
DWS captureLength, width, height, actual weight, image
CategoriseCategory assigned from dims, weight, flags
Destination resolveSort rules give destination, chute and bag
LabelReprint if barcode unreadable or relabel needed

Scan and vision events

Event	Trigger	Key fields
PARCEL_INSCAN	Scan	`awb`, `station_id`, `user_id`
VISION_INSCAN	Profiler camera	`track_id`, `awb_ocr`, `confidence`
DWS_CAPTURED	DWS	`l_mm`, `w_mm`, `h_mm`, `wt_g`, `vol_wt_g`, `image_ref`
CATEGORY_ASSIGNED	System	`category`, `flags`
DESTINATION_RESOLVED	Rules engine	`dest`, `route`, `chute_id`

Parameters captured

Parameter	Unit	Source
Declared vs actual weight	g, %	Pre-alert, DWS
Volumetric weight	g	L x W x H / divisor (example 5000 cm³ per kg)
Chargeable weight	g	max(actual, volumetric)
Read result	enum: good, no-read, multi-read	Scanner
Data status	enum: found, no-data	Lookup

Resources

DWS units, label printers
Inscan operators where not automated
Label reprint station
Profiler camera

Rules

Weight discrepancy over threshold triggers billing adjustment event
Category drives path: conveyable to induct, others to manual sort

Exceptions

No-read
No-data
Weight discrepancy
Dimensions out of sorter range
Seen by camera but not scanned

KPIs

Inscan PPH
DWS read rate
Weight discrepancy rate

5.5 Machine sort

InductOperator places parcel label-up on induct belt, or auto-induct
ScanTunnel reads barcode, sends to WCS
DecideWCS returns chute from sort plan cache
DivertCarrier tilts or belt discharges to chute
ConfirmDivert confirm or recirculate or reject; camera at sorter exit marks primary sort done

Scan and vision events

Event	Trigger	Key fields
PARCEL_INDUCTED	PLC	`induct_id`, `carrier_id`
SORTER_SCAN	Tunnel	`read_status`, `codes`
SORT_DECISION	WCS	`chute_id`, `reason`
PARCEL_DIVERTED	PLC	`chute_id`, `confirm`
VISION_PRIMARY_SORT	Chute camera	`chute_id`, `track_id`
PARCEL_RECIRC / PARCEL_REJECT	PLC	`reason`: no-read, chute full, no-data, late decision

Parameters captured

Parameter	Unit	Source
Induction rate	items/h per induct	PLC
Carrier fill %	%	PLC
Recirculation count per parcel	count	PLC
Decision latency	ms	WCS
Sorter state time	running, idle, starved, blocked, fault	PLC
Chute fill level	%	Sensor or camera

Resources

1 to 2 inductors per induct (example)
Sorter, scan tunnel, chutes
Maintenance technician on shift
Chute cameras for jams

Rules

Max recirculation 2 (example) then reject chute
Chute full: hold divert, recirculate, or overflow chute per plan
Dynamic chute reassignment allowed only from approved plan

Exceptions

No-read
No-data
Chute full
Jam or pile-up
Double induct
Oversize detected

KPIs

Sorter TPT and OEE
Induct PPH per person
Recirculation %
No-read %
Sort accuracy

5.6 Secondary / manual sort

ReceiveBulky, non-conveyable, rejects, exceptions and secondary-sort parcels arrive (camera marks arrival as secondary sort)
ScanHandheld scan shows destination and location
PlacePut to wall, cage, pallet or floor grid location
ConfirmLocation scan confirms placement

Scan and vision events

Event	Trigger	Key fields
MANUAL_SORT	Handheld	`awb`, `location_id`, `user_id`
VISION_SECONDARY_SORT	Area camera	`zone_id`, `track_id`
LOCATION_FULL	Handheld	`location_id`

Parameters captured

Parameter	Unit	Source
Manual sort PPH	parcels/h per person	Derived
Walk distance per parcel	m	Path graph, vision tracks
Wrong-location scans	count	Validation

Resources

Manual sorters (example 1 per 120 to 200 parcels/h)
Put walls with PTL, pallets, cages
Pallet jacks for bulky

Rules

Destination locations laid out by volume, closest for highest volume
Scan-to-confirm mandatory

Exceptions

Wrong location
Location full
Parcel too heavy for one person (2-person lift rule above 25 kg example)

KPIs

Manual sort PPH
Manual sort accuracy
Bulky TOT

5.7 Sort exceptions

CaptureException raised with type, photo or camera clip, location
RouteTo exception desk queue by type and priority
ResolveRelabel, data lookup, repack, hold, RTO, or reinduct
CloseOutcome and time recorded

Scan and vision events

Event	Trigger	Key fields
EXCEPTION_RAISED	Any	`type`, `awb`, `severity`, `photo_ref`
EXCEPTION_ASSIGNED	System	`owner_id`
EXCEPTION_RESOLVED	Desk	`resolution`, `reinduct`

Parameters captured

Parameter	Unit	Source
Open exceptions by type and age	count, min	Events
Resolution time	min	Events
Repeat exception per AWB	count	Events

Resources

Exception desk 1 to 3 people per shift (example)
Label printer, scale, repack materials

Rules

Types: no-read, no-data, misroute, hold (client or legal), damage, short, excess, weight dispute, address issue, DG undeclared, seen-not-scanned
Age SLA per type, e.g. no-read 15 min, no-data 60 min (example)

Exceptions

Unresolved past cutoff becomes a missed connection

KPIs

Exception rate per 1,000 parcels
Mean time to resolve
Exception backlog

5.8 Bagging and consolidation

Open bagBag ID assigned to chute, destination printed on tag
FillParcels scanned into bag (or auto-associated by divert); camera detects bagging action
CloseOn count, weight, fill or cutoff rule
SealSeal number scanned, bag weighed
ManifestBag added to outbound manifest for the connection

Scan and vision events

Event	Trigger	Key fields
BAG_CREATED	Handheld or PTL	`bag_id`, `chute_id`, `dest`
PARCEL_BAGGED	Scan or divert	`awb`, `bag_id`
VISION_BAGGING_DETECTED	Camera	`chute_id`, `count_est`
BAG_SEALED	Scan	`seal_no`, `wt_g`, `count`
MANIFEST_UPDATED	System	`manifest_id`, `bag_id`

Parameters captured

Parameter	Unit	Source
Parcels per bag	count	Events
Bag weight	kg	Scale
Bag fill time	min	Events
Chute dwell	min	Events
Bagging PPH per person	parcels/h	Derived

Resources

1 bagger per 4 to 8 chutes (example)
Bags, tags, seals, scale, printer
Roll cages to staging

Rules

Close bag at max 25 kg or 60 parcels or 2 h fill (example defaults)
Forced close at cutoff minus 30 min (example)
Mother bag for low-volume destinations

Exceptions

Parcel scanned into wrong bag
Bag overweight
Seal not scanned

KPIs

Bagging PPH
Bag fill rate
Chute dwell
Missort caught at bagging

5.9 Staging and loading

StageSealed bags moved to route lane
Assign vehicleTrip and dock assigned
Load scanEvery bag scanned into vehicle, sequence for multi-drop
ReconcileManifest vs loaded, camera bag count at dock
Seal vehicleSeal number and photo

Scan and vision events

Event	Trigger	Key fields
BAG_STAGED	Scan	`lane_id`
LOAD_START / LOAD_END	Handheld or camera activity	`dock_id`, `trip_id`
BAG_LOADED	Scan	`bag_id`, `trip_id`
VEHICLE_SEALED	Scan	`seal_no`, `photo_ref`

Parameters captured

Parameter	Unit	Source
Staging dwell	min	Events
Load duration and rate	min, bags/h	Events
Vehicle fill (volume and weight)	%	Bag dims and weight vs vehicle capacity
Left-behind bags	count	Reconcile

Resources

Loading crew 2 to 3 per dock (example)
Telescopic conveyor, roll cages
Handheld
Dock camera

Rules

Block vehicle seal if manifest bags are missing (override needs supervisor)
Load by drop sequence for milk runs

Exceptions

Bag missing at load
Wrong-vehicle load
Vehicle late or not placed
Loaded on camera but not scanned

KPIs

Loading PPH
Vehicle fill %
Left-behind rate
Dock turn time

5.10 Dispatch and handover

Gate-outVehicle, seal and trip checked at gate
DepartDeparture time vs schedule
HandoverLinehaul or last-mile partner e-POD
NotifyPre-alert to destination hub and client webhook

Scan and vision events

Event	Trigger	Key fields
VEHICLE_DEPARTED	Gate	`trip_id`, `departed_at`
HANDOVER_CONFIRMED	App	`partner_id`, `pod_ref`
PREALERT_SENT	System	`dest_site`, `bag_ids`

Parameters captured

Parameter	Unit	Source
Departure vs scheduled	min	Events
Connection made	bool	Cutoff vs bag load time
Handover count match	bool	Partner scan

Resources

Security at gate
Dispatch supervisor

Rules

Late departure reason code mandatory
Auto pre-alert on gate-out

Exceptions

Missed connection
Partner count mismatch

KPIs

On-time departure %
Connection adherence %
Inbound-to-dispatch TOT

5.11 RTO and returns

IdentifyRTO flag from client or failed delivery
InscanSeparate RTO lane or sort destination
ConsolidateBy client or seller return address
DispatchBack to origin or client warehouse

Scan and vision events

Event	Trigger	Key fields
RTO_MARKED	System	`awb`, `reason`
RTO_SORTED	Scan	`dest`

Parameters captured

Parameter	Unit	Source
RTO volume and share	count, %	Events
RTO dwell	h	Events

Resources

RTO desk or chute group
Bags for client consolidation

Rules

RTO parcels never re-enter forward sort
Ageing alert over 24 h (example)

Exceptions

RTO with damage
Client address unknown

KPIs

RTO TOT
RTO ageing

5.12 High-value cage

DetectFlag from declared value over threshold (example INR 10,000)
MoveEscorted to cage, dual scan in
StoreCage location scan, camera watches cage
ReleaseDual scan out to sealed HV bag

Scan and vision events

Event	Trigger	Key fields
CAGE_IN / CAGE_OUT	Scan	`cage_id`, `user_ids`
VISION_LOITER	Cage camera	`cage_id`, `dwell_s`, `clip_ref`

Parameters captured

Parameter	Unit	Source
Items in cage, value in cage	count, INR	Events
Cage dwell	min	Events
Access events	count	Badge, camera

Resources

Cage custodian per shift
Lockable cage, cameras with 2 angles

Rules

Two-person rule for in and out (camera checks two people present)
No HV parcel on open chute

Exceptions

Unmatched cage count
Unauthorised access
Loitering

KPIs

HV loss rate
Cage dwell

5.13 Dangerous goods

DetectDG flag, UN number, keyword screening, or DG label seen by camera
SegregateDG zone, separate bag with DG label
RouteSurface only where air is restricted

Scan and vision events

Event	Trigger	Key fields
DG_FLAGGED	System, scan or camera	`un_no`, `class`
DG_SEGREGATED	Scan	`zone_id`

Parameters captured

Parameter	Unit	Source
DG count by class	count	Events
Undeclared DG found	count	Exceptions

Resources

DG-trained staff
Spill kit, DG cage

Rules

Only DG-certified users can handle; enforced at login
Lithium battery rules by mode

Exceptions

Leak
Undeclared DG

KPIs

DG compliance %

6. Parcel and bag model, categories

Parcel entity

Field	Type / unit	Source	Notes
`awb`, `parcel_id`, `client_id`, `order_ref`	string	Pre-alert	Canonical ID plus source IDs
`service_type`	enum: express, standard, surface, air, same-day, B2B	Pre-alert	Drives priority
`origin_site`, `dest_site`, `dest_pincode`, `route_code`	string	Pre-alert, rules
`declared_wt_g`, `actual_wt_g`, `vol_wt_g`, `chargeable_wt_g`	g	Pre-alert, DWS
`l_mm`, `w_mm`, `h_mm`, `volume_cm3`	mm, cm³	DWS
`category`, `handling_flags`	enum, list	Rules	See category table
`payment_type`, `cod_amount`, `declared_value`	enum, INR	Pre-alert	COD and HV rules
`dg_flag`, `un_no`, `dg_class`	bool, string	Pre-alert, screening
`fragile`, `orientation`, `stackable`	bool	Pre-alert, label
`promised_by`, `connection_id`, `cutoff_at`	timestamp, ID	SLA engine
`state`, `location_id`, `bag_id`, `last_event_at`	enum, IDs, timestamp	Twin	Live state
`stage_times`	map of stage to in and out timestamps	Events	Per-stage TOT
`vision_status`	map: inscan, primary sort, secondary sort, bagged, each with source scan, vision or both	Fusion	Scan vs camera reconciliation
`exception_ids`, `rto_flag`, `hold_flag`	list, bool	Exceptions

Bag entity

Field	Type / unit	Source	Notes
`bag_id`, `bag_type`	enum: standard sack, mother bag, HV bag, DG bag, tote, roll cage, pallet	Bag master	Reusable or single-use
`origin_site`, `dest_site`, `route_code`, `chute_id`	string	Bagging
`seal_no`, `seal_state`	string, enum	Scan
`parcel_count`, `wt_g`, `fill_pct`	count, g, %	Scans, scale, camera
`parent_bag_id`	string	Consolidation	Mother and child
`manifest_id`, `trip_id`	IDs	Manifest
`state`	enum: open, closed, sealed, staged, loaded, in transit, received, opened, empty	Twin

Parcel categories and their effect on TOT, TPT and PPH

Category	Default threshold (configurable)	Handling path	Effect on PPH and TPT	Effect on TOT (example)
Flyer / polybag	Up to 0.5 kg, flat, max side 400 mm	Machine sort, bagged	Highest PPH; recirculation risk on some sorters	Low
Small box	Up to 2 kg, max side 450 mm	Machine sort, bagged	High PPH	Low
Standard	2 to 10 kg, within sorter max dims	Machine sort, bagged	Baseline	Baseline
Large	10 to 25 kg or a side 600 to 900 mm	Machine sort if within OEM limits, else manual	Slower induct, fewer per carrier	+15 to 30 min
Bulky	25 to 50 kg or a side over 900 mm	Manual sort, loose load	Manual PPH 3 to 5 times lower	+30 to 90 min
Non-conveyable	Irregular, rolls, tyres, liquids, over 50 kg, outside sorter limits	Manual, MHE needed	Lowest PPH; 2-person or MHE	+60 to 180 min
Fragile	Flag from client or label	Manual or low-drop chute	Lower PPH, damage risk	+10 to 30 min
High-value	Declared value over INR 10,000	Cage, dual control	Extra scans	+15 to 45 min
COD	Payment type COD	Normal path, COD tracking	Neutral	Neutral
Dangerous goods	DG flag or UN number	DG zone, segregated bag	Manual, certified staff	+30 to 120 min
Express / priority	Service type	Priority induct and early bag close	Neutral	Shorter, by rule
RTO	RTO flag	RTO lane or chute	Neutral	Separate SLA

Thresholds and time effects are example defaults; each site sets them from its sorter's OEM limits and its own time studies. A parcel can hold one size category plus several flags (fragile, HV, COD, DG, express, RTO).

How mix changes site throughput

Effective site PPH = 1Σ ( share of category ÷ PPH of category )

Because the slow paths sit in the denominator, a small rise in bulky or non-conveyable share pulls site PPH down sharply.

7. Workforce module

Workforce parameters

Field	Type / values	Source	Notes
`employee_id`, `name`, `employment_type`	enum: permanent, contract, agency, gig	HRMS
`vendor_id`	string	HRMS	Contract labour vendor
`team_id`, `role`	enum: unload, de-bag, inscan, induct, manual sort, bagging, load, exception, cage, MHE, supervisor	Roster
`skills`	list of skill, level 1 to 4, certified until	Skills matrix	DG, MHE licence, induct, cage
`shift_id`, `planned_start`, `planned_end`	time	Roster
`attendance_in`, `attendance_out`	timestamp	Biometric
`station_id`, `checkin_at`, `checkout_at`	timestamp	Scanner login or badge	Deployment log
`state`	enum: active, idle, starved, blocked, travel, break, training, off-floor	Derived from scans plus camera activity	Every minute
`zone_presence`	zone, time in zone	Camera or BLE	Anonymous unless linked to badge
`scans`, `parcels_handled`	count	Events
`std_pph`	parcels/h	Time study by station and category	Engineered standard
`cost_per_hour`	INR/h	Finance	Includes overtime multiplier
`overtime_min`, `break_min`	min	Attendance
`errors`	count: missort, wrong bag, damage	Exceptions	Quality per person

Utilisation states

State	Rule (example)	Meaning
Active	Scan or task event in the last 60 s (example), or camera sees a work activity at own station	Productive time
Idle	Logged in, no event over 60 s, camera sees idle pose, work available at station	Coachable
Starved	No event, station queue empty	Planning issue: redeploy
Blocked	No event, downstream full or machine fault	Machine or flow issue
Travel	Moving between stations (camera track crosses zones)	Layout or task issue
Break	Scheduled break window or break check-in	Planned
Off-floor	Attended but not checked in and not seen in any work zone	Leakage

Formulas

Required heads = Peak hourly load at stationStd PPH per person × Efficiency

Utilisation = Active minutesAttended minutes

PPH per person = Parcels handledDeployed hours

Labour cost per parcel = Σ (hours × cost rate)Parcels processed

Peak hourly load = daily volume ÷ operating hours × peak factor × stage share of volume. Daily heads = heads per shift × number of shifts, plus an absenteeism buffer (example 8 to 12%).

Demand-based redeployment logic

Forecast loadPer station, per 15 min, from pre-alerts, yard queue, trips ETA, and in-floor WIP
Compute needRequired heads per station from load, std PPH, efficiency and category mix
CompareGap = required minus deployed (camera headcount cross-checks logins); flag stations over plus or minus 1 head for 2 intervals
MatchPick donors from starved or over-staffed stations with the required skill, shortest walk
GuardrailsMinimum heads per station, max 2 moves per person per shift, no move within 20 min of last move
RecommendShift lead gets a card: move 3 from bagging to de-bag for 40 min, expected TOT gain
LearnTrack accepted moves and actual effect to tune the model

Where people are most utilised

Utilisation heatmap: station by 15-min interval, coloured by active share
Marginal value per head: TOT minutes saved by adding one person at each station, ranked
Skills coverage: stations where fewer than 2 certified people are on shift
Contract vs permanent: PPH, error rate, attrition and cost per parcel by employment type and vendor
Person scorecard: PPH vs standard, accuracy, attendance, stations worked

8. Machine and WCS module

Sorter parameters

Parameter	Unit	Source	Notes
Design throughput	items/h	OEM	Ideal rate
Actual throughput	items/h	PLC divert count	Rolling 15 min
Induct rate per induct	items/h	PLC
Carrier utilisation	%	Occupied carriers / available
Read rate	%	Tunnel	Good reads / scans
No-read, multi-read	%	Tunnel
No-data	%	WCS	Read but no destination
Recirculation	%	PLC	Recirc events / inducted
Reject	%	PLC	To reject chute
Divert confirm failure	%	PLC	Possible missort
Chute full time	min per chute	Sensor or camera
State durations	min: running, idle, starved, blocked, fault, planned stop	PLC	For OEE
Alarms	code, severity, duration	PLC
Energy	kWh	Meter	Cost per parcel

OEE for sorters

OEE = Availability × Performance × Quality

Availability = Running timePlanned production time

Performance = Items inductedRunning time × Design throughput

Quality = Correctly sorted itemsItems inducted

Planned production time excludes planned maintenance and scheduled breaks; starved time is reported separately so a sorter waiting for parcels is not blamed on the machine.

WCS configuration and sort plans

Item	Description
Sort plan	Destination to chute mapping per wave or time window; versioned; compiled to edge cache
Sort keys	Pincode, route, hub, client, service, category, priority
Chute groups	Several chutes per destination with fill-balancing
Overflow rules	Chute full: recirculate N times, then overflow chute, then reject
Dynamic reassignment	Swap low-volume destinations in and out of chutes by hourly volume, only from approved options
Induction control	Induct priority, gap, speed limits, stop and start by schedule
Device config	Scanner symbologies, DWS thresholds, printer templates, PTL colours, camera ROIs and models
Manual fallback	On sorter fault: switch to handheld sort to floor grid using the same plan

Alarm catalogue (examples)

Alarm	Source	Severity	Owner
Jam at chute or induct	PLC alarm or camera pile-up	High	Technician
E-stop	PLC	Critical	Shift lead and maintenance
Scan tunnel read rate under 97% (example)	KPI	High	Maintenance
Chute full over 5 min (example)	Sensor or camera	Medium	Bagger
Recirculation over 5% (example)	KPI	Medium	Shift lead
Sorter starved over 10 min while yard has vehicles	Twin	High	Shift lead
Comms loss with PLC	Edge	Critical	Integration on-call

Manual fallback

DetectSorter state fault over 3 min or availability below threshold
AnnounceAlert to shift lead, twin shows affected chutes
SwitchHandhelds switch to manual sort mode with destination-to-floor-location plan
RedeployInductors moved to manual sort per staffing model
RecoverSorter back: drain manual WIP to bagging, resume machine sort

9. Digital twin and simulation

Twin components

Component	Content	Fed by
Static model	Layout, zones, objects, path graph, conveyor segments, capacities, camera positions and coverage from master data	CAD import, master data
Live state	Every vehicle, bag, parcel, person, device, chute and dock with state and location	Scan, PLC and vision events
People layer	Headcount and activity per zone, anonymous tracks, utilisation	Vision, logins, BLE
Flow state	Queues and WIP per stage, arrival and service rates, utilisation	Stream KPIs
History	All events with event time, enabling replay	Event log, time-series DB
Simulation	Discrete-event model calibrated from history	Simulation engine
Forecast	Next 1 to 24 h load per stage	ML on pre-alerts, trips, seasonality

Replay

Scrub any past shift on the floor view at 1× to 60× speed
See queues, people states, camera headcounts and machine states at any minute
Trace one parcel or bag through every scan and vision sighting with stage times
Compare two days side by side

What-if levers

Lever	Inputs	Outputs
Volume	Daily volume, hourly curve, peak factor	TPT, TOT, backlog, cutoff misses
Mix	Category shares (flyer, standard, bulky, non-conveyable, HV, DG)	Manual sort load, TOT by category
Headcount	Heads per team, shift times, skill mix, contract share	PPH, utilisation, cost per parcel
Machine	Downtime window, induct closed, chutes disabled, speed change	TPT drop, fallback load
Sort plan	New chute map, destination merges	Chute balance, walk distance, missort risk
Arrivals	Vehicle delays, schedule change	Dock queue, connection adherence
Layout	Add de-bag tables, move staging lanes	Travel time, congestion (from vision tracks)

Bottleneck detection

MeasureArrival rate and capacity for every stage each 15 min
UtilisationStage with highest utilisation and growing queue is the active constraint
Root causeSplit capacity loss into people (idle, absent), machine (fault, starved, jam) and mix (bulky share)
QuantifyMinutes of TOT added and parcels at risk of missing cutoff
ActRecommend redeploy, chute swap, induct priority, or dock change

10. Computer vision and camera layer

Uses the cameras each hub already has. Edge AI boxes on site read the NVR streams, run the models locally and send only small events (and short clips for alerts) to the cloud, so the WAN, the NVR and the cameras carry almost no extra load.

People: headcount, activity, time in zone

Capability	How it works	Event fields	Frequency
Live headcount per zone	Person detection plus tracking inside zone polygons; count is unique tracks present in the last 10 s (example)	`zone_id`, `count`, `confidence`	Every 5 s
Activity recognition	Pose and action model per track: unloading, de-bagging, inducting, scanning, manual sorting, bagging, loading, carrying, pushing cage, MHE driving, idle standing, phone use, walking	`track_id`, `activity`, `confidence`, `duration_s`	On change
Time in zone	Entry and exit of each track per zone; dwell per visit and cumulative	`track_id`, `zone_id`, `enter_at`, `exit_at`	On exit
Anonymised tracking	Random track ID per visit, re-identification across cameras inside the site only, no face stored; IDs rotate every shift	`track_id`, `x`, `y`	1 s
Badge linking (optional)	Fuse track with BLE tag, scanner login at station or badge tap to attach `employee_id`	`track_id`, `employee_id`, `link_method`	On link
Face ID (optional, opt-in)	Only with explicit consent, only at entry points, templates encrypted on edge; off by default	`employee_id`, `match_score`	On match
Crowding and flow	Density per m² and walking paths heatmap	`zone_id`, `density`	1 min

Live headcount by zone (example view)

Zone	Seen on camera	Planned	Logged in	Gap
Inbound docks	2	3	3	1 short
De-bag	6	6	5	1 present without login
Induction	4	4	4	OK
Secondary sort	5	4	4	1 extra
Bagging	3	4	4	1 short
Outbound docks	3	3	2	1 present without login
Exception desk	1	2	2	1 short

Illustrative numbers only. The gap column is what drives redeploy cards and login-gap alerts.

Parcel tracking by vision

Status or detection	Vision trigger	Event	Effect in WMS
Auto inscan	Parcel passes profiler or DWS; camera track plus label OCR or barcode read	VISION_INSCAN	Status inscan (vision); confirmed when scan matches
Primary sort done	Parcel seen leaving sorter into chute	VISION_PRIMARY_SORT	Status primary sort done for that chute's destination
Secondary sort	Parcel arrives in the secondary sort area or put wall	VISION_SECONDARY_SORT	Status secondary sort in progress
De-bagging	Bag opened and emptied at station; parcel count estimate	VISION_DEBAG_DETECTED	Compare with bag manifest count
Bagging	Parcels dropped into bag at chute; bag closed and lifted	VISION_BAGGING_DETECTED	Compare with scans into bag
Load and unload count	Bags crossing dock tripwire in or out	VISION_DOCK_COUNT	Compare with load or unload scans
Chute jam or pile-up	Fill level estimate over 90% or no movement for 60 s while divert continues	VISION_CHUTE_JAM	Alert bagger and maintenance
Conveyor jam or gap	Parcels stationary on running belt, or long empty gaps at induct	VISION_CONVEYOR_JAM	Alert and induct coaching
Floor parcel	Parcel lying outside any valid zone over 5 min	VISION_STRAY_PARCEL	Pick-up task
Damage	Crushed, torn or leaking parcel classifier	VISION_DAMAGE	Damage desk with image

Camera vs scan reconciliation

Case	Rule (example)	Outcome
Seen but not scanned	Vision event with no matching scan within 120 s (example) at the same stage	Missed scan task to the station; counts in scan compliance
Scanned but not seen	Scan with no vision event at a covered stage	Check camera coverage or ghost scan (scan without physical handling)
Count mismatch	Vision count for bag, dock or chute differs from scan count by more than tolerance (example 1 or 2%)	Recount or audit task
Location mismatch	Scan says chute A; vision sees parcel at chute B	Possible missort; hold bag before seal
Matching keys	Time window, location, OCR text, size and colour signature, sorter carrier ID	Fusion confidence score 0 to 1

Security and suspicious activity

Behaviour	Detection	Zones	Default confidence	Severity
Pilferage gestures	Action model: concealment into clothing or bag, opening a parcel, hand-to-pocket after handling	All work zones	0.80	High
Loitering near HV cage	Track dwell inside cage buffer over 60 s (example) without a cage scan	HV cage	0.70	High
Two-person rule breach	Cage open with fewer than 2 people	HV cage	0.75	Critical
Non-dock exit	Person or parcel crossing a tripwire at a non-dock door or fence	Perimeter, side doors	0.70	Critical
Restricted zone intrusion	Person in DG, cage or server room zone without access role (badge link) or after hours	Secure zones	0.70	High
Rough handling	Throw, drop from height, kicking, stepping on parcels; parcel velocity and pose	Docks, sort, bagging	0.75	Medium
Missing PPE	No safety shoes, high-vis vest or helmet where required	Docks, MHE areas	0.70	Medium
Unsafe MHE	Forklift speed or person too close to moving MHE	MHE areas	0.75	High
Camera tamper	Blocked, moved or defocused view	All cameras	0.90	High
Smoke or fire	Smoke and flame classifier	All	0.80	Critical

Every security alert carries a clip (10 s before, 20 s after, example), a still, the zone and the confidence. A human reviews it before any action against a person.

Models

Model	Task	Approach	Default threshold	Latency per frame
Person detector	Detection	YOLO-class detector, fine-tuned on hub footage	0.50	under 20 ms
Multi-object tracker	Tracking	ByteTrack or BoT-SORT	n/a	under 5 ms
Cross-camera re-ID	Anonymous re-identification	Appearance embedding, site-local, shift-scoped	0.60 similarity	under 10 ms
Activity and pose	Action recognition	Pose keypoints plus temporal classifier (2 to 4 s window)	0.65	under 80 ms per clip
Parcel and bag detector	Detection and segmentation	Detector with parcel, bag, roll cage, pallet classes	0.55	under 20 ms
Label OCR and barcode	Recognition	Text detector plus recogniser; barcode decoder	0.80	under 40 ms
Chute fill estimator	Segmentation	Fill % from segmented parcel area	n/a (threshold 90% fill)	under 30 ms
Damage classifier	Classification	Crushed, torn, wet, leaking	0.75	under 15 ms
PPE detector	Detection	Vest, helmet, shoes	0.70	under 20 ms
Suspicious action	Action recognition	Concealment, throwing, kicking	0.75 to 0.80	under 100 ms per clip
Smoke and fire	Classification	Smoke, flame	0.80	under 20 ms

Camera placement per zone

Zone	Camera type	Mount	Count rule	Resolution	FPS	Models
Gate and yard	Bullet, ANPR at gate	4 to 6 m	1 per gate lane plus 1 per 40 m of yard	1080p	5 to 10	Vehicle, ANPR, person
Inbound and outbound docks	Dome or bullet facing dock door	4 to 5 m	1 per dock (1 per 2 docks minimum)	1080p	10	Person, activity, bag count, PPE
De-bag stations	Overhead or angled dome	3.5 to 5 m	1 per 2 to 4 tables	1080p	10 to 15	Person, activity, parcel count
Profiler / DWS	Fixed box camera near tunnel exit	2 to 3 m	1 per DWS	1080p to 4 MP	15 to 25	Parcel track, OCR
Induction	Angled dome	3.5 to 4 m	1 per 2 inducts	1080p	10 to 15	Person, activity, belt gaps
Sorter chutes	Overhead along chute bank	4 to 6 m	1 per 6 to 10 chutes	4 MP	10	Fill level, jam, bagging
Secondary sort and put walls	Dome	4 to 6 m	1 per 150 to 250 m²	1080p	10	Person, activity, parcel
HV cage	2 angles, one overhead	3 to 4 m	2 per cage	4 MP	15	Person, loiter, gesture
DG zone and restricted rooms	Dome	3 to 4 m	1 to 2 per zone	1080p	10	Intrusion, PPE, smoke
Perimeter and side doors	Bullet	3 to 5 m	1 per door	1080p	10	Tripwire, intrusion

Runtime parameters

Parameter	Range or default	Note
`fps_inference`	5 to 25 fps	Per use case: headcount 5, activity 10, parcel at DWS 15 to 25 (example)
`resolution`	1080p default, 4 MP for chutes and cage	Sub-stream where enough
`conf_threshold`	0.50 to 0.90	Per model and per alert; tuned on site data
`min_duration_s`	2 to 60 s	Event must persist before firing (debounce)
`zone_polygons`, `tripwires`	Floor coordinates	Drawn in config studio on camera view
`privacy_masks`	Polygons	Blur or skip areas
`clip_pre_s`, `clip_post_s`	10 s and 20 s (example)	Alert clip window
`event_latency_target`	under 500 ms frame to event; under 3 s to alert
`model_version`	Semantic version	Rolled out like any config, shadow first
`camera_health`	online, fps achieved, blur score, tamper	Every 30 s

Edge hardware sizing (example)

Site size	Edge hardware	Capacity	Note
Small hub (up to 16 cameras)	1 edge box with an embedded GPU module (Jetson Orin NX class)	8 to 16 streams at 1080p, 5 to 10 fps	Example sizing
Medium hub (16 to 64 cameras)	2 to 3 edge servers, each with 1 inference GPU (L4 or T4 class)	24 to 32 streams per GPU at 1080p, 10 fps	Example sizing
Large hub (64 to 200 cameras)	Edge rack with 4 to 8 GPUs, N+1 redundancy	Scale by streams × fps × models	Example sizing
Storage on edge	4 to 16 TB for clips and buffers	Raw video stays on existing NVR
Network	Cameras on a separate VLAN; uplink 1 to 10 GbE	To cloud: kbps per camera for events, clips on demand
Power	UPS covering edge box and switches

Retention

Data	Retention (example)	Controls
Raw video	Existing NVR policy (often 7 to 30 days)	Not copied by the platform
Alert clips	30 to 90 days (example), longer if under investigation	Encrypted, watermarked, access-logged
Vision events (no images)	13 months hot, then aggregated	Pseudonymous track IDs
Track-to-employee links	Shift duration plus 30 days (example)	Only where badge linking is on
Face templates	Only if face ID is on; deleted on consent withdrawal or exit	Never leaves edge
Training data	Curated, blurred faces, opt-in sites only	Separate approval

Thresholds, stream counts and hardware are example starting points; each site is tuned during the vision pilot on its own footage and lighting.

Privacy and compliance (India)

Area	Design
Lawful basis	Use for safety, security and operations under the Digital Personal Data Protection Act, 2023 and its Rules: give notice, and rely on consent or on the employment-related legitimate uses allowed by the Act where they apply. Confirm the basis with legal counsel per client.
Notice	Clear signs at entry and in each covered zone in English, Hindi and the local language; written notice in the joining kit explaining purposes, data kept, retention, rights and grievance contact.
Consent	Separate, explicit, withdrawable consent for any optional feature such as face ID; a non-biometric alternative (badge) always offered; no penalty for refusal.
Purpose limitation	Each model tagged with its purpose; no use outside the listed purposes without a new notice.
Data minimisation	Events not video; anonymous tracks by default; privacy masks on rest and toilet areas; no audio.
Retention and erasure	Per the retention table; automatic deletion jobs with logs.
Rights of data principals	Access, correction, erasure and grievance handling through a named grievance officer, with response timelines tracked.
Security and breach	Encryption, role-based clip access, audit trail; breach notice to the Data Protection Board and affected people as the Act requires.
Fairness	Productivity scores never from vision alone; no automated discipline; human review of every security alert before action.
Worker engagement	Brief worker representatives and contractors before go-live; publish what is and is not monitored.
Assessment	Privacy impact assessment per site before enabling vision; repeat on new models.

Legal review. This is a design baseline, not legal advice. Have counsel confirm the lawful basis, notices, consent wording and retention under the DPDP Act and Rules, and any state labour rules, before switching on people analytics at a site.

How vision feeds the rest of the platform

Module	Integration
Digital twin	Live people dots and counts per zone, parcel sightings, chute fill, security events on the floor map; replay includes vision layers
Workforce	Activity time feeds active, idle, travel and off-floor states; camera headcount vs logins vs roster per zone each 5 min; time-in-zone by team
Parcel tracking	Vision statuses fill gaps between scans and raise seen-not-scanned tasks; dock and bag counts reconcile against scans
WCS	Chute jam and fill events feed overflow logic and maintenance; induct gap analysis
KPIs	Headcount accuracy, vision-scan match rate, seen-not-scanned rate, activity time shares, rough-handling per 10,000, PPE compliance, camera uptime
Alerts	Security, safety and flow alerts with a clip link, routed by severity (section 13)
Simulation	Real walking paths and activity times calibrate the twin's people model

Staffing and TOT calculator

Simplified planning estimate using example defaults; the twin's simulation gives the precise answer.

Daily volume (parcels)Operating hours per dayPeak factor (peak hour ÷ average)Shift length (h)Efficiency %Staffing level % of requiredSorter rated rate (items/h)Sorter availability %Flyer and small %Standard %Bulky %Non-conveyable %Exception share %

PPH per person by station (editable)

UnloadingDe-bagInductionManual sortBaggingLoadingException desk

Result

Station	Peak load/h	PPH/person	Heads/shift	Heads/day	Utilisation

Method: heads = ⌈peak load ÷ (PPH × efficiency)⌉; stage time = base handling time + 5 min × ρ ÷ (1 − ρ), where ρ is utilisation at the staffing level entered. Base times (example): unload 15, de-bag 10, machine sort 8, manual sort 20, bagging 25, staging wait 30, loading 15 minutes.

11. KPI dictionary

Throughput

KPI	Definition	Formula	Granularity	Example target
TPT (site)	Parcels dispatched in a period	Σ parcels with VEHICLE_DEPARTED in period	Site, hour, shift, day	At or above planned hourly
Effective PPH (site)	Dispatched parcels per operating hour	parcels dispatched / operating hours	Site, shift	Plan value
Stage PPH	Parcels processed by a stage per hour	stage out-events / hours	Stage, hour	At or above rated
PPH per person	Productivity of a person	parcels handled / deployed hours	Person, team, station	Example 300 to 600 for induct
Team PPH	Productivity of a team	team parcels / team deployed hours	Team, shift	Std PPH
Category PPH	PPH for one category	category parcels / hours on that path	Category, stage	Per category std
Sorter TPT	Items diverted per hour	divert confirms / running hours	Sorter, hour	Over 85% of design
Induct rate	Items inducted per hour per induct	inducts / hour	Induct	Rated
Capacity headroom	Spare capacity	1 − load / capacity	Stage, hour	10 to 20%

Time

KPI	Definition	Formula	Granularity	Example target
TOT (site)	Hub turnaround, first inscan to dispatch	departed_at − first_inbound_scan_at	Parcel, category, client, site	Example under 4 h express
Stage TOT	Time in a stage	stage_out − stage_in	Stage, category	Per stage target
Dwell time	Time a parcel or bag sits idle	Σ waiting gaps between stages	Stage, zone	Under 30 min
Yard dwell	Gate-in to dock-on	docked_at − gate_in_at	Vehicle	Under 20 min
Dock turn time	Dock-on to dock-off	undocked_at − docked_at	Dock, vehicle type	Under 45 min
Bag fill time	Bag open to seal	sealed_at − created_at	Chute, destination	By rule
Exception MTTR	Mean time to resolve	avg(resolved_at − raised_at)	Type	Under 30 min
Cutoff buffer	Time between bag load and cutoff	cutoff_at − loaded_at	Connection	Over 15 min

Quality

KPI	Definition	Formula	Granularity	Example target
Sort accuracy	Parcels sorted to correct destination	1 − missorts / sorted	Sorter, person, site	99.9% or more
Missort rate	Wrong chute or bag	missorts / sorted	Sorter, chute	Under 0.1%
No-read rate	Tunnel failed to read	no-reads / scans	Sorter	Under 1%
No-data rate	Read but no shipment data	no-data / reads	Client	Under 0.3%
Recirculation rate	Parcels going round again	recirc events / inducted	Sorter	Under 3%
Reject rate	Parcels to reject chute	rejects / inducted	Sorter	Under 1%
Damage rate	Damaged parcels per 10,000	damages / parcels × 10,000	Stage, site	Under 5
Shortage / excess rate	Bags or parcels short or excess vs manifest	short / expected	Origin, lane	Under 0.05%
Scan compliance	Parcels with every mandatory scan	parcels with full scan chain / parcels	Stage	99.5% or more

Utilisation

KPI	Definition	Formula	Granularity	Example target
People utilisation	Share of attended time that is productive	active min / attended min	Person, team	80 to 88%
Idle %	Logged in without work done	idle min / attended min	Person, team	Under 10%
Starved / blocked %	Waiting on upstream or downstream	starved or blocked min / attended min	Station	Under 5%
Deployment adherence	Heads at station vs plan	actual heads / planned heads	Station, interval	95 to 105%
Attendance %	Showed up vs rostered	present / rostered	Team, vendor	Over 92%
Sorter OEE	Availability × performance × quality	A × P × Q	Sorter, shift	Over 75%
Dock utilisation	Share of time docks occupied	occupied min / available min	Dock	60 to 80%
Chute utilisation	Chute active and not full	volume share and full-time %	Chute	Balanced
MHE utilisation	Equipment in use	in-use min / available min	Asset	Over 60%

Vision

KPI	Definition	Formula	Granularity	Example target
Headcount accuracy	Camera count vs audited count	1 − \|vision − audit\| / audit	Zone	97% or more
Login gap	People seen in zone without a station login	vision headcount − logged-in heads	Zone, interval	0
Vision-scan match rate	Vision events matched to a scan	matched / vision events	Stage	99% or more
Seen-not-scanned rate	Parcels seen but not scanned	unmatched vision / vision events	Stage, station	Under 0.5%
Activity time share	Share of time per activity	activity seconds / attended seconds	Zone, team	Idle under 10%
Rough handling rate	Incidents per 10,000 parcels	incidents / parcels × 10,000	Zone, shift	Trend down
PPE compliance	People with required PPE	compliant detections / detections	Zone	98% or more
Security alert precision	Alerts confirmed true on review	true alerts / reviewed alerts	Model, site	Over 80%
Camera uptime	Cameras streaming at target fps	healthy camera min / total camera min	Site	99% or more

SLA and cost

KPI	Definition	Formula	Granularity	Example target
SLA adherence	Parcels dispatched before promised cutoff	on-time parcels / parcels	Client, service	98% or more
Connection adherence	Bags loaded on planned connection	bags on planned trip / bags planned	Route	98% or more
On-time departure	Vehicles leaving on schedule	on-time departures / departures	Route	95% or more
Cost per parcel	Total hub cost per parcel	(labour + MHE + energy + space + overhead) / parcels	Site, stage, category	Down vs baseline
Labour cost per parcel	Labour only	Σ(hours × rate) / parcels	Team, stage	Trend down
Overtime %	Overtime of total hours	OT hours / total hours	Team	Under 8%
Contract share	Contract heads of total	contract heads / total heads	Site	Policy
Vehicle fill	Load factor	loaded volume / vehicle capacity	Trip	Over 85%

Every KPI is computed in streaming form (rolling 15 min, hour, shift) and batch form (day, week, month), and can be sliced by site, stage, machine, team, person, client, category and route.

12. Configuration studio

Configurable in WMS

Area	What can be set
Process flows	Stages, mandatory scans, order, skip rules per site template
Sort rules	Pincode, route, client, service, category to destination; effective dates
Category rules	Thresholds for weight, dims, value, DG; handling path per category
Bag rules	Max weight, count, fill, time; forced close before cutoff; mother bag rules
Manifest rules	Per connection, per client, e-way bill links
Exception types	Codes, SLA, owner role, required evidence, resolutions
SLA and cutoffs	Per client, service, route; priority scoring
Workforce	Shifts, std PPH by station and category, efficiency, min heads, redeploy guardrails, cost rates
Vision	Zones, tripwires, models per camera, thresholds, reconciliation windows, privacy masks, retention
KPI targets and alert thresholds	Per site, shift, stage
Scan validations	Allowed scans per station, hard and soft stops, override roles
Labels and documents	Templates for bag tags, manifests, relabels

Configurable in WCS

Area	What can be set
Sort plan	Destination to chute mapping per wave or time window; versioned; compiled to edge cache
Sort keys	Pincode, route, hub, client, service, category, priority
Chute groups	Several chutes per destination with fill-balancing
Overflow rules	Chute full: recirculate N times, then overflow chute, then reject
Dynamic reassignment	Swap low-volume destinations in and out of chutes by hourly volume, only from approved options
Induction control	Induct priority, gap, speed limits, stop and start by schedule
Device config	Scanner symbologies, DWS thresholds, printer templates, PTL colours, camera ROIs and models
Manual fallback	On sorter fault: switch to handheld sort to floor grid using the same plan

Config lifecycle

DraftEdit in the studio; every change is a diff against the live version
ValidateStatic checks: every destination has a chute, no chute double-mapped, capacity rules
Sandbox on twinReplay a recent real day through the new config; compare TPT, TOT, chute balance, walk distance
ApproveMaker-checker: author cannot approve; WCS changes need engineering approval
SchedulePublish at a wave boundary or sorter idle window
ShadowRun new plan in parallel, decisions logged but not sent, for one wave (optional)
CanaryOne induct, one chute group, or one site first
LivePromote; edge cache updated atomically
RollbackOne click to the last good version, under 60 s at edge

Roles and access

Role	Can do	Scope	Note
Platform admin	Tenants, connectors, global templates	All	No floor operations
Tenant admin	Users, roles, sites, client setup	Tenant
Solution engineer	WCS configs, device and camera setup, sort-plan drafts	Assigned sites	Cannot approve own
Site manager	Approve WMS configs, targets, rosters	Own site
Shift lead	Redeploy, exception overrides, manual fallback switch	Own site, shift
Supervisor	Team view, overrides within limits	Own team
Security officer	View and review vision security alerts and clips	Own site	Clip access logged
Privacy officer	Retention, consent records, DPIA, erasure requests	Tenant
Operator	Handheld tasks only	Own station
Control tower analyst	Read all sites, raise actions	Region	No clip access by default
Client user	Own shipments, KPIs, exceptions	Own data	Row-level security
Auditor	Read config history and audit log	Tenant	Read only

Every change writes an immutable audit record: who, what diff, when, approved by, sandbox result, and the rollback point.

13. Alerts and exception management

Alert	Trigger (example threshold)	Severity	Owner	Default action
Cutoff at risk	Parcels for a connection not bagged at cutoff minus 45 min (example)	High	Shift lead	Priority induct, early bag close
Stage backlog	Queue over 2× 15-min capacity	High	Shift lead	Redeploy
Sorter down	Fault over 3 min	Critical	Maintenance, shift lead	Manual fallback
Read rate drop	Under 97% for 10 min	High	Maintenance	Clean or recalibrate tunnel
Chute full or jam	Over 5 min, or camera pile-up	Medium	Bagger	Close bag, clear jam
Shortage	Bag or parcel short vs manifest	High	Exception desk	Trace with origin
Seal mismatch	Gate or dock seal check fails	Critical	Security, site manager	Hold vehicle
Idle spike	Team idle over 15% for 30 min	Medium	Supervisor	Redeploy or release
Headcount gap	Camera headcount below plan by 2 or more for 10 min	Medium	Shift lead	Redeploy
Seen not scanned	Over 0.5% at a station in 15 min	Medium	Supervisor	Coach, recover scans
Security event	Any vision security rule, with clip	High or critical	Security officer	Review clip, act
Attendance gap	Present heads under 90% of plan at shift start	High	Site manager	Call vendor, re-plan
Exception ageing	Over SLA by type	Medium	Exception desk	Escalate
HV cage mismatch	Cage count ≠ system	Critical	Security	Freeze cage
Connector or camera stale	No data over 2 min from a source	High	Integration on-call	Check edge
Dock queue	Vehicles waiting over 30 min	Medium	Yard marshal	Open dock

Aspect	Design
Delivery	Floor app push, control tower, floor TV, WhatsApp or SMS for critical, email digest
Escalation	Unacknowledged in 5 min goes to the next level; critical goes to site manager in 10 min (example)
Noise control	Deduplication, suppression during known downtime, hysteresis on thresholds, minimum duration for vision events
Closure	Root cause code and action required to close; vision alerts marked true or false to retrain models

14. Data model and event schema

Key entities

Entity	Purpose	Key relationships
Tenant, Client	Logistics company and its shippers	Tenant 1–n Client
Site, Zone, Dock, Station, Lane, Cage	Physical model	Site 1–n Zone 1–n Station
Sorter, Induct, Chute, Device, Camera	Machines and devices	Sorter 1–n Induct, Chute; Zone n–n Camera
Route, Destination, Connection	Network and schedule	Route 1–n Connection
SortPlan, SortRule, CategoryRule, BagRule, VisionRule	Versioned configuration	ConfigVersion 1–n rules
Shipment / Parcel	Unit of work	Parcel n–1 Bag
Bag, Manifest	Consolidation units	Manifest 1–n Bag; Bag 1–n child Bag
Vehicle, Trip	Movement	Trip 1–n Manifest
Employee, Team, Shift, Skill, Attendance, Deployment, Consent	Workforce	Employee n–n Skill; Employee 1–n Consent
Track, VisionEvent, Clip	Camera layer	Track n–0..1 Employee; VisionEvent 0..1 Clip
Exception, Alert, Action	Problems and responses	Parcel 1–n Exception
Event	Immutable fact	References any entity
KPI snapshot	Pre-aggregated metrics	Per entity, interval
AuditLog, ConfigVersion	Governance	User 1–n AuditLog

Event envelope

Field	Type	Notes
`event_id`	UUIDv7	Unique and time-ordered
`event_type`	enum	From the catalogue below
`schema_version`	int
`tenant_id`, `site_id`	string	Partitioning and isolation
`occurred_at`, `received_at`	timestamp with zone	Device time and ingest time
`source_system`, `source_device_id`	string	Includes camera ID for vision
`actor_id`, `station_id`, `zone_id`	string	Who and where
`entity_type`, `entity_id`, `parent_id`	string	Parcel, bag, vehicle, track; parent bag or trip
`x`, `y`	m	Optional location
`confidence`, `model_version`, `clip_ref`	0 to 1, string, string	Vision events only
`payload`	typed object	Per event type
`correlation_id`, `idempotency_key`, `sequence_no`	string, int	Tracing, dedupe, order

Event catalogue

Domain	Event types
Yard	VEHICLE_GATE_IN, VEHICLE_YARD_PARKED, VEHICLE_DOCK_CALLED, VEHICLE_DOCKED, VEHICLE_UNDOCKED, VEHICLE_DEPARTED
Inbound	UNLOAD_START, UNLOAD_END, BAG_INSCAN, LOOSE_PARCEL_INSCAN, BAG_SHORT, BAG_EXCESS
De-bag and inscan	BAG_OPENED, PARCEL_DEBAGGED, BAG_RECONCILED, PARCEL_INSCAN, DWS_CAPTURED, CATEGORY_ASSIGNED, DESTINATION_RESOLVED
Sort	PARCEL_INDUCTED, SORTER_SCAN, SORT_DECISION, PARCEL_DIVERTED, PARCEL_RECIRC, PARCEL_REJECT, MANUAL_SORT
Bag and outbound	BAG_CREATED, PARCEL_BAGGED, BAG_SEALED, MANIFEST_UPDATED, BAG_STAGED, LOAD_START, BAG_LOADED, LOAD_END, VEHICLE_SEALED, HANDOVER_CONFIRMED, PREALERT_SENT
Exceptions and special	EXCEPTION_RAISED, EXCEPTION_ASSIGNED, EXCEPTION_RESOLVED, RTO_MARKED, CAGE_IN, CAGE_OUT, DG_FLAGGED
Machine	SORTER_STATE, SORTER_ALARM, INDUCT_STATE, CHUTE_STATE, DEVICE_HEARTBEAT
Vision: people	VISION_ZONE_COUNT, VISION_ACTIVITY, VISION_ZONE_ENTER, VISION_ZONE_EXIT, VISION_TRACK_LINKED
Vision: parcels	VISION_INSCAN, VISION_PRIMARY_SORT, VISION_SECONDARY_SORT, VISION_DEBAG_DETECTED, VISION_BAGGING_DETECTED, VISION_DOCK_COUNT, VISION_CHUTE_JAM, VISION_CONVEYOR_JAM, VISION_STRAY_PARCEL, VISION_DAMAGE, VISION_SCAN_MISMATCH
Vision: security and safety	VISION_PILFERAGE, VISION_LOITER, VISION_INTRUSION, VISION_NON_DOCK_EXIT, VISION_ROUGH_HANDLING, VISION_PPE_MISSING, VISION_MHE_UNSAFE, VISION_TAMPER, VISION_SMOKE
People	USER_LOGIN, USER_LOGOUT, STATION_CHECKIN, STATION_CHECKOUT, USER_STATE, ATTENDANCE_IN, ATTENDANCE_OUT, REDEPLOY_RECOMMENDED, REDEPLOY_ACCEPTED
Config	CONFIG_DRAFTED, CONFIG_APPROVED, CONFIG_PUBLISHED, CONFIG_ROLLED_BACK

15. Dashboards per role

Role	Main content	Time grain	Actions
Ops control tower	All sites live map, TPT vs plan, TOT, connection risk, sorter health, alerts by severity	Region, network	Drill into any site twin
Site manager	Shift plan vs actual, PPH by team, cost per parcel, OEE, SLA, attendance, top bottlenecks	Site, shift, day	Approve configs, rosters
Shift lead	Live floor twin, camera headcount per zone, queues per stage, redeploy cards, cutoff countdown, exceptions	Live, 15 min	Accept redeploys, switch fallback
Supervisor / team	Team heads, active, idle, PPH per person, errors, seen-not-scanned	Live, hour	Coach, move
Security	Vision security alerts with clips, cage activity, perimeter, review queue	Live, day	Confirm or dismiss, escalate
Maintenance	Sorter state timeline, alarms, read rate, chute faults and jams, camera health, MTBF, MTTR	Live, shift	Acknowledge alarms
Integration health	Connector and camera freshness, lag, error rate, quarantine queue	Live	Replay, pause
Client portal	Own volumes, TOT, SLA, exceptions, shortages, damage, PODs	Day, week	Download, webhooks
Finance	Cost per parcel by stage, labour cost, OT, contract vendor billing	Day, month	Export
Floor TV	Hourly target vs actual, cutoff board, safety reminders	Live	Display only

16. Non-functional requirements

Area	Requirement	Note
Scale	Up to 500 sites per tenant, 50 M events per site per day at peak, 200 cameras per site, 5,000 concurrent users	Example sizing target
Latency	Event to twin p95 under 2 s; sort decision at edge under 30 ms; frame to vision event under 500 ms	See latency table
Availability	Cloud 99.9%; edge 99.99% with HA pair; site never depends on cloud to keep sorting
Durability	No event loss: replicated log, edge buffer 72 h
Retention	Raw events 13 months hot, 7 years cold (configurable); vision per section 10	Client and legal needs
Multi-tenancy	Tenant ID on every row and topic, row-level security, per-tenant keys, optional dedicated deployment
Security	SSO (SAML, OIDC), MFA for admins, mTLS edge to cloud, encryption at rest and in transit, secrets vault for camera and PLC credentials
OT safety	Writes to PLC only through WCS executor, interlocks, read-only default, IEC 62443 aligned zoning; cameras on own VLAN
Privacy and compliance	DPDP Act 2023 and Rules: notice, consent where needed, minimisation, retention, rights, breach reporting; data residency in India regions; immutable audit logs
Observability	Tracing per event, connector and camera SLAs, model drift monitoring, synthetic tests
Offline	Handheld app offline queue, edge local cache, edge vision keeps running
Extensibility	Connector SDK, model registry, webhooks, public APIs, custom KPI builder
Localisation	English, Hindi and regional languages on floor apps and signage; INR; IST

17. Phased rollout roadmap

Phase	Duration	Scope	Exit criteria
0. Discovery	4 weeks (example)	Site surveys, system and camera inventory, data samples, baseline KPIs	Integration plan per site type
1. Observe	8 to 10 weeks	Edge gateway, read-only connectors, event bus, parcel and bag tracking, core KPIs, basic twin	One pilot site with 99% scan coverage
2. Workforce and twin	8 weeks	People states, staffing model, redeploy cards, full floor twin, replay	Utilisation and PPH visible per person
3. Vision pilot	8 weeks	Edge AI box, headcount and activity, parcel vision at DWS and chutes, scan reconciliation, HV cage security, privacy assessment and notices	Headcount accuracy 97%, privacy sign-off
4. Configure WMS	8 weeks	Config studio for WMS rules, versioning, approvals, exceptions module	Rules changed without code
5. Simulate	6 to 8 weeks	What-if simulation, forecasting, bottleneck engine	Simulated vs actual TOT within 10%
6. Control WCS	10 to 12 weeks	Sort-plan compiler, edge executor, shadow then canary, manual fallback	Live sort plan from studio at pilot
7. Scale	Ongoing	Connector library per sorter OEM, full vision security suite, multi-site rollout, client portal, AI optimisation	10+ sites live

Next step. Share one real site's numbers (daily volume, sorter type and OEM, headcount by team, shift pattern, camera count) to calibrate the calculator and build a worked example.