Building Deep-Tech Products That Can Actually Be Manufactured

Separate the phenomenon from the repetition. Fund the first until the effect is undeniable on the bench you can describe. Fund the second as a different project: fixtures, incoming variation, the order of operations, what is adjusted by feel, and what happens when that person is absent. Hardware Product Development Process Explained is the stage structure around that repetition. An EV power stage is one example of an effect that still has to become an assembly someone else can build; that hardware is treated as a product in EV product design. The step rates below are invented for the arithmetic. They are not a measured yield, not a factory result and not a BrahmWorks figure.

Write the effect as an output you can check: a thickness band, a waveform into a stated load, a pass or fail you defined. If the only check is that an expert likes the look, the expert is the process. Keep them on the plan, visibly, until a check exists that someone else can run.

List the inputs they control without saying so. Material lot, humidity, a wait they never skip, a reject they pull before the count. Each one is a future difference between sites. A special-purpose machine is justified when an input cannot be held more simply. It is not a monument. Build the cheapest safe rig that holds the input before you design the machine around a process that still changes daily.

If the output moves when an input you cannot buy or measure moves, automation makes scrap faster. Keep the step in the lab with a named operator, or remove it from the product. Shipping a technician as a hidden part misleads partners.

The instruction has to include the judgements. "Align until it looks right" is not a step unless the failing cases are shown. Where a judgement remains, name who may make it. That is the constraint a slide calling the process scalable leaves out.

Fixtures should remove hand adjustments. If they need the same hands, they are clamps. Record the settings. A photograph without settings cannot go to a supplier. Incoming material needs a check about your effect, not only a generic certificate. Two lots can meet a datasheet and still diverge in your step.

Have a second person build a few units from the instruction with the inventor out of the room. Count every intervention. The product is not manufacturable while those interventions are required. If you state a rate, label its inputs, as the sum below does.

Manufacturing will not supply a theory you skipped, a measurement you cannot make, or a tolerance stack you refused to close. It will reveal them at a volume that is expensive and public. Do the revealing on a short run whose only job is repetition. Do not combine that run with a customer promise, a new enclosure and a new material in the same week. You will not know which change broke the effect.

Compliance, safety and, where relevant, a regulated claim stay in their own scopes. A repeatable lab effect is not an approval to sell. An EV assembly that repeats on a bench is not a vehicle interface. State the boundary in the same paragraph as the result. This article states no BrahmWorks price, timeline, yield or test outcome. The teaching rates are not a substitute for the short run you have not done.

The teaching process coats a small part. It works when one scientist watches a meniscus and lifts at a moment they cannot yet describe. It is not a BrahmWorks process. The manufacturing question is whether a second person can hit an agreed thickness band on five parts, using a written lift cue, with the scientist absent.

In the rig: the bath, a fixture for the angle, a visible cue, and a thickness check someone else can run. Out of the rig: a machine, a factory, a shipment and any yield promise. If the second person succeeds only when the scientist corrects the cue, the cue is not written. The effect may be real. The product is not repeatable. The next spend is a better cue, or an explicit decision to keep the scientist in the process and to show that cost.

A machine that automates an unwritten cue automates the scatter. Do not conclude a yield or a cost per part from this story.

These three rates are assumptions for a teaching product of probabilities. They are not a measured yield and not a BrahmWorks result. Assume the steps are independent, which is itself an assumption and often false.

Assume step A finishes inside limits on 97 of 100 tries, step B on 91 of 100, step C on 98 of 100. The combined fraction is 0.97 times 0.91 times 0.98. 0.97 times 0.91 is 0.8827. 0.8827 times 0.98 is 0.865. About 87 of 100 hypothetical units would finish if those rates were true.

Now assume step B is 91 of 100 only while the scientist is present, and 70 of 100 when they are not. The combined fraction becomes 0.97 times 0.70 times 0.98. 0.97 times 0.70 is 0.679. 0.679 times 0.98 is 0.665. About 67 of 100, under these new assumptions. The lab was reporting the 91 case and calling it the process. The manufacturing question is the 70 case. The arithmetic does not estimate your process. It shows how much of the story sits in who is in the room. Replace every rate with a count from a run the expert did not touch, or do not quote a rate at all.

• The effect has an output a second person can check.
• Unspoken inputs are listed, including rejects pulled out before the count.
• Judgements in the instruction are shown, not only named.
• Fixtures replace hand adjustments, or the hands are an explicit cost.
• A second person has built units with the inventor out of the room.
• Interventions were counted.
• The short run is not also a launch.
• Any rate in a pitch is either counted on that run or labelled as an assumption. No BrahmWorks yield is stated.

It contains judgements, material choices and rejected tries that were never written down. The inventor's presence is a process input. Scaling copies it, including the failures, until you turn it into settings and checks.

When a written input cannot be held by a simpler fixture, and the effect is stable enough that the machine will not freeze last month's experiment. Describe the cue before you build the frame.

Those that still track an input you cannot measure or buy, and those that still need a named person after a real attempt to write the cue. Keeping them in the lab is a decision. Hiding them in a production plan is not.

A phenomenon happens. A product happens again, inside limits, without the original person, with the inputs recorded. Until then, say so in the sentence, including when the effect is an EV assembly someone else must still build.

Bring the output check, the list of quiet inputs, and what the second person did when left alone. BrahmWorks can say whether the next piece of engineering is a fixture, a rewritten step, or a recommendation to keep the process in the lab until the effect sits still.

Review the next stage with BrahmWorks