For the loss-prevention seat in general merchandise and apparel, where item-level RFID has been mature for years.
Your investigators do not lack data. That is the part vendors get wrong when they call on you.
You tagged at item level years ago. You read at the DC door, at the store back room, on the sales floor cycle count, at the point of sale. A single unit generates more observations in its life than most companies generate for an entire pallet. When something goes missing, your team pulls the read history in minutes.
And the investigation ends the same way it always ends: the item was seen here, then not.
That sentence is the whole problem, and it is not a data-volume problem. You cannot fix it by reading more often, tagging more densely, or adding another portal. You have already run that experiment. The reads answer what, when, and where with a precision most industries would envy. They do not answer who, and no quantity of them ever will — because the record they write has no field for it.
One handoff, read by read
Take the handoff where your money actually leaks: a vendor delivery into a DC.
A trailer arrives at 06:12. The ASN says 50 units of a style. Your dock portal reads 48 as the cases cross. Somebody hand-stacks two cases onto a different pallet mid-unload because the pallet wrap failed. At 06:41 a putaway read picks up 48. The cycle count nine days later finds 48. Your vendor-compliance team raises a chargeback for the 2-unit short. The vendor's WMS says 50 shipped, with their own scan history to prove it.
Now assemble what you actually have:
| Question | Your reads answer it? |
|---|---|
| What was it? | Yes — to the item, by EPC |
| When did it cross the dock? | Yes — 06:12, to the second |
| Where was it read? | Yes — portal, dock door, putaway location |
| How many were read? | Yes — 48 |
| Was it read again later? | Yes — putaway 06:41, cycle count day nine |
| Who received it? | No |
| Who moved the two cases? | No |
| Who says the count was 48? | No |
The first five rows are why you invested in RFID, and they delivered. The last three are why the dispute takes six weeks and settles on relationship rather than fact.
Notice what the vendor has: the mirror image. Their scans, their timestamps, their locations, their count. Both parties hold rich, internally consistent, mutually contradictory data. Neither holds anything the other is obliged to accept. This is not an accuracy problem between two systems. It is an evidence problem between two companies — and it is the ordinary condition of every chargeback your team runs.
Why the remaining error is handoff error
When item-level accuracy plateaus — and in a mature RFID estate it does — look at where the residual error actually lives. It is rarely in the read. Modern portals are very good. It is in the moments between reads, where a human made a decision the record did not capture: a case re-stacked, a pallet split, a unit pulled for a damage check and never re-scanned, a ship-from-store pick reversed at the bag.
Every one of those is a handoff. Every handoff has a person in it. And the record grammar you are using — EPCIS, the GS1 standard your readers almost certainly emit — defines a supply-chain event in five dimensions: what, when, where, why, and how (§7.2.2). There is no performer among them. The party fields it does have are organization-grain: a source or destination company, a GLN (§7.3.6.4; CBV 2.0 §8.7.1).
Which means the standard can record a company received goods at a location. It cannot record this associate, at this door, at 06:12. There is no Who field sitting empty in your data waiting to be populated. There is no field.
That is worth being precise about, because it changes what kind of purchase this is. You are not shopping for a better reader, a cleaner data lake, or another analytics layer over observations you already have. Those all improve rows one through five of that table. The rows that decide your disputes are structurally absent from the record, and no product that writes conformant events can add them without extending the standard — which the standard's own extension mechanism permits, and which is the entire thing we build.
What performer attribution would change in that chargeback
Return to the 06:12 delivery with two facts added to each event: who touched it — the person or agent that observed the goods — and who says so — the account standing behind the record.
The dispute stops being a contest between two internally consistent datasets. Your record now says a named receiver, on your account, observed 48 at that door at that time. The two re-stacked cases carry the move and the person who made it. When your vendor-compliance director sends the claim, they are not sending a count; they are sending an attributed observation that the vendor can verify without joining your network, because the event's identity is a hash of its own normalized fields (CBV 2.0 §8.9) — either side recomputes it, neither side asks an intermediary.
That is the difference between data and evidence. You have had the first in abundance for years. The second has never been available to you, in any product, because the record had nowhere to put it.
For your omni-channel seat the same fact reads differently: ship-from-store reliability is a handoff-integrity number, and handoff integrity is a who question wearing an inventory-accuracy costume. For your LP seat it reads most directly of all — attribution is not a nicer report, it is the thing every investigation currently ends one step short of.
What does not apply to you, said plainly
Two clocks get pitched at retail this year. Neither is yours.
Sunrise 2027 is a food-and-grocery-weighted point-of-sale programme. It does not bind apparel or general merchandise in any material way. If a vendor puts a Sunrise countdown in front of your general-merchandise business, one question ends the meeting — ask which of your categories it actually reaches.
FSMA 204 is a food rule. It has nothing to say to you.
So there is no regulatory clock here, and we are not going to manufacture one. The argument for this segment is structural, not scheduled: you have already exhausted what counting reads can tell you, and the only material information left in your supply chain is who handled the goods. That argument is either persuasive on its own or it should not persuade you at all.
We will also be straight about our own evidence. The public artifacts on traceability ROI are food-sector artifacts, and we will not stretch them over your business to make a slide work — a vendor who does that is one question from being caught, and you should catch them. We carry no customer references in your segment, and no conformance attestation has ever been issued to us. What we will do is show you the record structure, and let you check the standards clauses above yourself; they are public.
Your investigations end at seen here, then not because the record was built to count reads, and it counts them beautifully. The question your seat actually gets paid to answer is the next one — and it has been unanswerable, for everyone, because the standard never had a place to write it down.
Who touched it?