A number on the cell before you buy it.
Checked after.
We write the throughput and the labor math for a palletizing or packing cell down before the spend, with every assumption labeled, and measure the installed line after it. The integrator's number is theirs. This one is yours.
Three plants, three numbers that were not in the public record
Each analysis starts from the plant's own published facts, states every assumption so one line from the plant replaces it, and ends with the number the decision turns on.
The palletizing arithmetic a job ad implies
A new plant sized at 225 million cans a year advertised production roles that stack pallets by hand. We worked out what that is per person, per shift, and what it costs at design capacity.
Read the analysisWhat a new 40 to 50 million pound line has to move
The company's filing and its chief executive's later statement put the new line at four to five million cases a year. We turned that into a per-minute rate, a pallet count, and the labor behind it.
Read the analysisWhat a $13.3 million line has to earn per hour
A packager won a state incentive award for an automated line expected to create nine jobs. The record shows jobs created, not jobs removed, so on that reading the payback has to come from rate. We sized the rate.
Read the analysisCompany names are withheld. Each analysis is built from the plant's own public record. Every figure carries its label:
A public source, with the date of the event it reports and the date we last verified it.
HHA's own sourced work, not the plant's record. Sources itemized on request.
An assumption stated so that one line from the plant replaces it. Never presented as a measurement.
Arithmetic on the lines above it. Nothing new enters here.
Production lines are where a prediction can be checked within a year
HHA's research subject is verified prediction: models whose claims are written down before the result and measured after it. Most of that work lives in benchmarks and simulation environments, where a check takes months and the ground truth is contested. A packing cell is the opposite. The prediction is a throughput and a headcount, the measurement is the installed line, and the answer arrives inside a year. Every case here is published with its method so the arithmetic can be checked by anyone.
Most plants have no filing and no announcement. The same arithmetic runs from a case format, a target rate, a shift plan and current staffing.
Haedar Hadi, Hass Dhia, Ahmed Dhia
Haedar Hadi, Lead Principal Investigator. MS Computer Science (Boston University). Leads the simulation and measurement side: the cell models, the benchmark runs, and the red-team of every measurement protocol before it is used.
Hass Dhia, Co-Principal Investigator and Treasurer. MS Biomedical Sciences. Leads experimental design and analysis.
Ahmed Dhia, Director of Manufacturing. Design for manufacturability, production scaling and quality systems (ISO 13485): the plant-floor side of the work.
Published work you can open: the ColludeBench multi-agent benchmark, and the hemosim and woundsim reinforcement-learning environment suites, all open source with their test suites. The institute is an independent California 501(c)(3); its research, team and governance are on the institute page.
Simulation that was checked against the installed line
A peer-reviewed study at a food producer modeled a robotic tray-loading cell, not a palletizing cell, predicted its throughput, installed it, and measured. In the large-scale production case the line delivered exactly what the model said. In the order-based case it came in three packs short, and the authors reported that too. Their work, not ours. We cite it as the standard for the method, not as a result of ours.
Predicted before the spend, measured after it
Robot integrators simulate the cell they are selling. Their model is competent and it is theirs. The number that rarely exists in that transaction is one produced by a party not selling the equipment, and checked afterwards.
Start from the plant's own facts
Announced capacity, filed headcount, the job ads, the case size on the spec sheet, or, for most plants, the case format, target rate, shift plan and current staffing the plant gives us. Every input is labeled by where it came from, and the estimates carry their assumptions so one line from the plant replaces them.
Write the prediction down first
Throughput, staffing and cost for the cell as specified, in writing, with the tolerance agreed before anything is bought.
Measure the installed line the same way
The same quantity, on the cell as built, over a stated window. A hit is reported as a hit and a miss as a miss, the way the published record does it.
The two-page analysis carries no obligation. The full predicted-then-measured study on a cell a plant is buying is a paid engagement with a fixed scope, a stated duration and a stated list of what we need from the plant, agreed before it starts. Its method and its predicted-versus-measured result enter the institute's published work on verified prediction; the plant's own figures never do.
Send us the line you are deciding on.
A short note with what the cell is meant to do is enough. We reply within two business days with what we would measure, what it would cost to be wrong, and whether the question is worth thirty minutes on a call. If it is not, we say so.