Purge sign-off rests on evidence, not on memory

The problem

It is irreversible, and it does not show at the outlet.

And it is irreversible. It is not visible on the way out. It becomes visible when the lab measures, hours later, or when the customer complains, weeks later. By then the material has been bagged and possibly shipped. Today the decision rests on somebody looking and saying it is clean now, with no objective evidence and under pressure to claw back changeover time. There is a structural conflict there: every bit of productivity pressure pushes in the opposite direction from purging properly, and the person resolving it is under pressure, every time, alone.

• Cross-contamination from the previous grade is irreversible and does not show at the outlet.
• It shows when the lab measures, hours later, or when the customer complains, weeks later. By then the material is bagged and possibly shipped.
• Today the decision rests on somebody looking and saying "it is clean", with no objective evidence and under pressure to recover changeover time.
• There is a structural conflict there: all the productivity pressure pushes against purging properly.

How it fits the IRIS system

A reference state per critical point, and JIDOKA AI that will not let it start.

With Edge, fixed cameras at each critical point compare the current state against the reference state of purged and ready for the next grade. Hopper and feeder, screw throat, screen changer, cutter and screen deck, first metres of belt. The model is trained to tell purged apart from carrying residue of the previous grade. And then JIDOKA AI does what it does: the line does not start until every camera reports clear and the quality manager signs. But the signature now rests on visual evidence point by point, not on memory. Accountability stops being an act of faith and becomes a documented act.

The quality lead still signs. But they sign on visual evidence point by point, not on memory and haste. That changes who carries the conflict.

See the full IRIS architecture →

Before and after

The grade change, before and after

Aspect	Today	With iLEAN Edge
Purge verification	Somebody looks and says it is done	A camera at every critical point
Points covered	Whichever get looked at	Hopper, screw, screen changer, cutter, screen, belt
Starting with residue	Possible, with pressure in favour	The line does not start
What the signature rests on	Memory and haste	Visual evidence point by point
When contamination is caught	Hours or weeks later	Before start-up
Changeover time	Depends who runs it	SMED AI orders the sequence

Impact estimate

Impact estimate — to validate against your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Less rework from cross-contamination; payback depends on grade changes per week.
• So the requirement does not penalise productivity, SMED AI runs alongside and orders the changeover sequence.
• The changeover evidence is stored on its own, and it is what the process audit asks for.
• And the conflict between purging properly and starting fast stops being settled by one person under pressure.

Less rework from cross-contamination · payback scales with changeovers per week. So that the requirement does not cost productivity, SMED AI walks the changeover sequence and orders it. The changeover does not get longer: it gets organised. The impact is a direct cut in rework and grade downgrading caused by cross-contamination, which is one of the biggest margin leaks a recycler has. Payback scales with the number of changeovers per week. An estimate to validate. Conservative ranges, an estimate to validate against the plant's real data.

And the fair question from the production manager

"Can a camera certify a purge?" — it is not asked to certify. It compares the current state against the reference at each specific point of this plant, which is an anchored comparison, and hands that evidence to whoever does certify. The signature stays human; what changes is what it rests on.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.