Why we built the twin before the product
No processor lets an unproven agent near a blade. Simulation was the only way to earn the first trial.
Blog
Notes from the cut floor.
Engineering write-ups, measurement methodology and honest industry analysis. No thought leadership, no hype cycles.
Featured
Latest writing.
The real cost of weight giveaway
A gram per pack sounds trivial until you multiply it across a year of fixed-weight production.
Detection versus prevention
Threshold detectors force a trade between escapes and false rejects. Fused perception dissolves it.
Calibrated uncertainty at line speed
A confidence number that is not calibrated is worse than no confidence number at all.
The cut-floor labour crisis is structural
Turnover and injury rates are not a cyclical staffing problem. They are a design problem.
Fail-safe is a hardware decision
If your stop path runs through a model server, you do not have a stop path.
Topics
What we write about.
Perception
Sensor fusion, calibration, uncertainty, drift and the unglamorous reality of condensation on a lens.
Cutting & yield
Seam-following, cut-response physics, yield measurement and how to avoid fooling yourself with averages.
Food safety
HACCP alignment, evidence design, recall readiness and the ethics of autonomous decisions in food.
Plant-edge systems
Deterministic latency, graceful degradation, OTA safety and running Kubernetes in a wet room.
“If your stop path runs through a model server, you do not have a stop path. You have a wish.”
Engineering series
The plant-edge series.
Sensor fusion when the sensors disagree
X-ray says bone, vision says shadow, hyperspectral is undecided. What should the agent do?
Latency budgets as a safety control
Treating a missed deadline as a safety event rather than a performance regression.
Rolling models onto a running line
Canary cells, staged promotion and rollbacks measured in seconds, not maintenance windows.
Behind the posts
The numbers we write about.
4.1%
Average saleable-yield uplift on pilot deboning lines [PLACEHOLDER]
61%
Reduction in weight giveaway per fixed-weight pack
38 ms
Median foreign-material reject decision at the plant edge
99.9%
Plant-edge runtime uptime target with fail-safe line stop
Industry analysis
Market and category.
Why point tools cannot close the loop
A camera that measures and a machine that cuts, with no path between them, is a data-collection exercise.
What a $25B category is actually made of
Cutting, grading, inspection, portioning and MES — and where autonomy captures the value.
The compounding data advantage
Why the cut-and-yield corpus is the asset, and why it cannot be bought.
Deep dive
From the archive.
Measuring yield without fooling yourself
Yield numbers are easy to inflate: shift selection, carcass mix, reconciliation windows and definitional drift all conspire. This post sets out the measurement protocol we use in every pilot, and why we reconcile against the customer’s MES rather than our own telemetry.
- Baseline construction and carcass-mix matching
- Reconciliation windows and lag handling
- Statistical significance before promotion
- Why the customer’s number always wins
Responsible AI
What we will not publish.
SOC 2 Type II (in progress)HACCP-alignedUSDA-FSISEU 853/2004GDPRISO 27001 (planned)
Per-tenant isolation
Every plant gets isolated data, models and vector stores. Cut recipes and species models never cross a tenant boundary — federated learning shares patterns, never your product IP.
Assurance-grade audit trail
Immutable logs link sensor frames, model version, autonomy level, approvals, overrides and outcomes for every agent action — built for recall defence and model governance.
Fail-safe by design
Blade, robot and line stops are deterministic and independent of the cloud. Graceful degradation returns the cell to a safe state if perception confidence drops.
Reader feedback
Who reads this.
“The line does not care that every bird is different — Slicium does. We stopped programming a machine and started supervising an operator that adapts to each carcass.”
“Giveaway was the quiet leak nobody could close. Watching the portioner hold target within a couple of grams, shift after shift, changed the economics of the whole pack line.”
“What sold my team was the audit trail. Every reject links back to the frame, the model version and who approved the autonomy level. That is what a recall investigation actually needs.”
Design-partner quotes are composite and pending publication approval. [PLACEHOLDER]
More
Resources and guides.
Designing a protein autonomy pilot
Picking the wedge line, defining the metric and structuring the gates.
Foreign material: detection versus prevention
A recorded session with quality leaders from three processors.
Agent API reference
The five calls every Slicium agent implements.
Get the next post.
Occasional, technical, no marketing filler. Unsubscribe in one click.