A pallet crosses the receiving portal and every case appears in inventory. Three cases from a parked pallet also appear, even though they never moved.
The reader worked. Its radio found valid tags. The product failed because it translated a nearby observation into a completed receipt.
RAIN RFID can remove scanning labor and expose movement that manual systems miss. It also creates a subtle product problem. Radio waves do not respect the tidy rectangles drawn on a warehouse map.
An RFID product needs a read-zone contract: explicit rules for where observations count, how they become transitions, what confidence means, and how people resolve ambiguity.
A tag read is an observation, not a location
A raw read usually identifies a tag, reader, antenna, and time. It may also carry signal strength, phase, channel, or a seen count. None of those fields alone proves that an item entered a business location.
The distinction matters because the same tag can be heard through a doorway, reflected from metal, or captured by two adjacent readers. Orientation and surrounding material can also make an item disappear for several cycles while it remains still.
GS1 EPCIS separates the physical observation from its business context. An event describes what happened, when, where, why, and sometimes how. That model is useful precisely because an identifier read is not yet a receiving, shipping, or stocking event.
Products create presence inflation when every successful read becomes durable evidence that an item is present. The dashboard looks complete, but certainty has been added by software rather than earned from evidence.
The safer default is to store the observation first. The product can then decide whether the evidence supports a location transition, a repeated confirmation, or an exception.
Define the read zone as a product contract
“At dock three” is not a sufficient zone definition. A deployable contract must connect physical design, reader configuration, and business meaning.
For each zone, document:
- The antennas and reader ports that belong to it.
- The physical boundary and acceptable leakage beyond it.
- The tag types, item materials, speeds, and orientations tested.
- The transmit power, session, search mode, filters, and reporting interval.
- The business action that a qualifying passage is allowed to trigger.
Impinj recommends tuning transmit power so a portal reliably reads the intended tags while minimizing stray reports. It also advises naming antennas so the source remains visible in the event stream. Those are product controls, not installation trivia.
The contract should include a test fixture. Put tagged cases at the zone boundary, behind likely reflectors, on adjacent lanes, and in the hardest expected orientations. Record both misses and strays.
Repeat that test after layout changes, antenna replacement, firmware updates, or a new tag inlay. A read zone is a measured operating envelope, not permanent geometry.
Turn observations into ordered transitions
Once the zone is bounded, the product still needs transition logic. A useful state machine might distinguish absent, approaching, inside, departing, and ambiguous.
Movement should require evidence across time and space. A portal can combine antenna sequence, first and last seen times, dwell, direction estimates, prior location, and an external trigger such as a photo eye.
Zebra exposes tag reads separately from inventory start, inventory stop, antenna, and general-purpose input events. That separation is a useful reminder. The application owns the interpretation that combines those signals.
Transition rules should be visible and versioned. For example, a receipt might require the expected tag to appear on the outer antenna, then the inner antenna, within four seconds while the dock sensor is active.
Keep the evidence behind each conclusion. An operator investigating a disputed move needs the reader, antenna, timestamps, rule version, and competing observations. A bare “received” status cannot explain itself.
This is another case where data products break when meaning changes. If “seen” quietly becomes “present,” every downstream metric inherits the semantic shortcut.
Reconciliation belongs in the first release
No tuned portal eliminates ambiguity. The first release therefore needs an exception path, not only a happy-path animation.
Create a review queue when a tag appears in competing zones, arrives without the expected sequence, remains visible after departure, or conflicts with the prior location. Group related tags so a pallet problem does not become forty separate tasks.
The reviewer should see the proposed transition, supporting observations, conflicting evidence, and the operational consequence. Available actions might include confirm, reject, assign location, rescan, or mark the tag for inspection.
Corrections must be additive and attributable. Preserve the original observations, the machine conclusion, the human decision, and the resulting inventory change. Silent overwrites destroy the evidence needed to improve both the installation and the rules.
Measure the system at the transition level. Useful metrics include confirmed transitions, false transitions, missed passages, ambiguous cases, time to resolution, and stray-read rate by antenna.
Tag read rate still matters. It measures radio performance. It does not measure whether the product told the truth about movement.
RFID earns trust when the interface makes that difference plain. Bound the zone, define the transition, retain the evidence, and provide a way back from uncertainty.
Comments
Comments are powered by GitHub Discussions through Giscus.