Industrial Simulation Strategy

Path A · November 2026

← Back to the starting plan

  1. Stage 1Capital-Infused SimulationsStage real production runs in one partner shop, and record everything.
  2. Stage 2Scaled Manufacturing SimulationsContract outside shops to run and record. Traine owns the data.

Prove the simulation. Scale it across shops, and own every run.

The cost of a run is paid once. The data it produces can be licensed again and again.

Stage 1

Capital-Infused Simulations

Use in-house CNC capacity to stage real production runs and record everything. We create the scarce data instead of searching for it.

The scarcest industrial data is scarce because it is expensive or destructive to make. Equipment gets repaired before it fails, so real failure records rarely exist.1 A shop that runs tools to failure on purpose can produce them.

What “simulation” means here. These are real machining runs, staged deliberately on real machines. They are physical simulations of production, not computer-generated data.

The NIST foundation

The U.S. standards lab publishes part designs and a model for what a complete dataset contains.

11NIST CAD test cases with full tolerancing: 5 combined and 6 fully-toleranced, each modeled in four CAD systems2
4 layersin NIST's published test-bed packages: design, machining program, machine execution data, and inspection results34
20instances of the same part recorded in NIST's box-assembly dataset, machined from aluminum3
$0licence cost. NIST states files its federal staff develop are not subject to copyright5

NIST's own data is free, so it is our baseline, not our product. Ours must add what NIST's lacks: tools run to the end of their life, real failures, more materials, and far more volume.

What every run captures

NIST's four layers, plus the streams that make the data rare. All on one clock.

LayerWhat is recordedFormat
DesignThe part as designed, with tolerancesNative CAD and STEP
PlanThe machining programCAM file and NC code (ISO 6983)
ExecutionWhat the machine did, moment by momentMTConnect
InspectionMeasured result for every partQIF
SensorsSpindle load, vibration, sound, temperatureTime-stamped streams
Tool lifeWear measurements, images, time to failureRecords and images
VideoThe cut and the work areaSynchronized video
Operator actionsTool changes, offsets, interventions (with consent)Event log
OutcomePass or fail, defect type, confirmed by sectioningVerified label

The first four layers mirror NIST's published structure,3 which keeps the data in open formats a buyer can use immediately.

Scaling the model

A planning illustration for one CNC mill and one lathe, including small-diameter tubular bar stock.

24part designs
×
3materials
×
3tool and setting combinations
×
20repeats, matching NIST
=
= 4,320recorded part runs, each a complete package

Where 24 designs comes from: 14 public NIST designs as candidates (the 11 test cases plus the 3 box-assembly parts), and 10 turned and tubular designs drawn in-house for the lathe. NIST's published packages center on milled parts, so bar-stock work uses our own designs. Each NIST part still has to be checked for fit on our machines.

Run once, license many times

One programRun and paid for once. Owned outright, with clean rights.
Licence to buyer A
Licence to buyer B
Licence to buyer C
and each buyer after that
  • The cost is fixed. The revenue is not. Each non-exclusive licence sells the same recorded runs again, with almost no added cost.
  • The market already works this way. Aerometrex licensed survey data it had already captured to two AI companies.6
  • Exclusivity is priced as a premium. Founders report exclusive deals cost buyers roughly four to five times more than non-exclusive ones.7
  • Made-to-order data sells for more than off-the-shelf data in other categories,8 so a buyer-specified program is the goal.

What the pilot proves. The shop knows its real cost per run. A pilot on one machine, shown to buyers, measures the other side: what a licence is worth, and how many times it sells.

Stage 2

Scaled Manufacturing Simulations

Take the Stage 1 program to outside manufacturers. They run the parts on their machines with our capture equipment. Traine owns every run.

One shop proves the model. Many shops make the dataset. The same parts, cut on different machines by different people, give a buyer variety that no single manufacturer can supply.

Why more shops matter

Buyers are asking for variety and for results tied to process.

  • Every environment is different. micro1's head of robotics data put manufacturing first on his list of settings where physical AI needs data, because the movements differ from place to place.9
  • Buyers already describe this product. One open Mercor listing seeks “run records that tie each recipe to a measured result.”10
  • Manufacturing carries the highest listed range. Mercor's manufacturer listing tops out at $1.5M for one company's records.11
  • Stage 1 supplies the template. The run spec, the capture kit, and the cost per run are all proven in one shop before any outside contract is signed.

The offer to a manufacturer

Make these parts under supervised data collection. Get paid well for the production.

  • Nothing proprietary is at stake. The parts come from public NIST designs2 or designs Traine owns. No customer work and no shop product is involved.
  • The order is large enough to matter. A sizable block of guaranteed production justifies having capture equipment in the plant.
  • The margin is set to attract. It is agreed in the contract and paid whether or not the data ever sells.
  • Capture stays on the contracted job. Sensors and cameras cover the assigned machine and run only, and the shop can see what is recorded.
  • Traine takes everything the run produces: the data, the finished parts, and the scrap.

The capacity match

A manufacturer tells Traine what it can run. Traine answers with work it can quote.

  1. The shop submits its capacity on Traine.ai: machine make and model, work envelope, tooling, materials, and inspection equipment.
  2. Traine searches its part library for designs those machines can make.
  3. The shop receives the matches as standard request-for-quote packages.
  4. The shop quotes, and a contract run begins.
  • This is how Traine generates demand. Shops come to us with open capacity, and leave with paid work to quote.
  • The library starts with NIST's public designs.2
  • Growing it is a standing job at Traine: finding or buying designs whose licences allow open commercial use, with rights cleared on every file.
  • Each submission maps who can make what. That lets Traine plan runs across different machines, which is the variety buyers want.

How a contract run works

Traine buys machine time the way any customer does. The deliverable is the data.

  1. Traine issues a request for quote to a matched shop, with the part files, the run spec, and the data terms.
  2. The manufacturer quotes its cost and its margin, as it would for any job.
  3. Traine installs the capture kit. The same kit goes to every site.
  4. The shop runs the program. Finished parts and scrap belong to Traine.
  5. Traine removes the kit, processes the data, and licenses it. Parts are sold where there is a buyer.

The price floor

Every contract run has a known cost before any data is sold.

Manufacturing cost
÷1 − the shop's agreed margin
+Kit install and removal
+Data processing
−Part sales and scrap recovery
= Traine's net floorLicence revenue above this line is Traine's margin.

A worked example

LineExample
Manufacturing cost for the program$120,000
Contract price at a 15% shop margin$141,176
Kit install, removal, and data processing$15,000
Part sales and scrap recovery−$20,000
Traine's net floor$136,176

Every figure in this example is a placeholder. The margin in particular will be set against standard manufacturing margins, at a level that makes outside shops want the work.

  • The shop is paid like any customer job. Its margin is guaranteed, so it carries no sales risk.
  • Traine owns the parts. They are sold for whatever exceeds scrap value. What has not sold by a set date is scrapped, which caps storage cost.
  • The contract price is committed before a sale. That is the capital this stage needs, and what an investor would fund.
  • A bid can be judged at once. With the floor known, an exclusive offer or a non-exclusive licence is a simple margin decision.

How many runs we can issue

Two things set the pace: agreements with buyers, and investment capital.

Pre-sold runs

A buyer agrees to a spec and a price before the request for quote goes out. Capital is tied up only between paying the shop and being paid.

Speculative runs

Traine funds the run and finds buyers afterward. Capital carries the full floor, and the return comes from licensing the data more than once.

  1. Buyer agreements come first. A written spec or letter of intent from one platform shows investors that a buyer asked for the data.
  2. Capital then sets the pace. It decides how many speculative runs Traine can carry alongside the pre-sold ones.
  3. Stage 1 is the proof for both. One pilot gives buyers a sample to react to, and gives investors a real floor.

The working rule. The first outside runs are pre-sold, or at least buyer-specified. Speculative runs begin once one dataset has sold more than once.

Terms every contract carries

Signed before the first run, so ownership is never reopened.

Assignment

All data, parts, and scrap from the run belong to Traine from the start. The shop keeps no right to resell any of it.

Warranty

The shop confirms it has the right to assign, and that no customer designs or controlled data are involved.

Releases

Anyone recorded on camera signs an individual release. A company cannot sign that for its people.

  • Public or Traine-owned part designs only. No customer work enters a run.
  • One spec at every site, so runs from different shops can be compared and pooled.
  • Parts are sold as reference pieces, as-is and never as certified components. Defective parts are scrapped, not sold.

These terms are a starting outline. Counsel drafts the contract before any request for quote goes out.

How the stages connect

One shop proves the model. Other shops scale it.

Stage 1 builds

An asset owned outright, plus the run spec, the capture kit, and real costs.

Stage 2 scales

The same kit, spec, and contract in outside shops. More data without owning more machines.

Where this path begins. The starting step supplies the buyer contacts. Their answer to one question decides the first run: would they buy purpose-built run data, and to what spec?

What could go wrong

  • No buyer publishes a program for this. Their listings describe a company's operating records. Purpose-built run data has to be sold through direct conversations.
  • Run costs are paid before data sells. The floor is committed whether or not a licence follows.
  • Shops may resist recording in their plant. The size of the order and the limits on capture have to answer that.
  • A buyer may ask for exclusivity, which ends reuse of that dataset unless it is priced in.
  • The window moves. The most valuable data type shifts as the market does.

Next steps

  1. Ask the buyers' contacts whether they would buy purpose-built run data, and to what spec.
  2. Scope the Stage 1 pilot with the shop, using its real costs.
  3. Write the run spec and capture-kit list that Stage 2 will reuse.
  4. Have counsel draft the data-assignment terms for contract runs.
  5. List the capacity fields and the first part library for the Stage 2 match.

Sources

Public sources retrieved November 16 to 17, 2026.

  1. arXiv: Neuro-symbolic AI for predictive maintenance, review and recommendations
  2. NIST: MBE PMI validation and conformance testing (CAD test cases)
  3. NIST Journal of Research: Design, manufacturing, and inspection data for a box assembly
  4. Data.gov: NIST Smart Manufacturing Systems Test Bed technical data packages
  5. NIST: Additive manufacturing test artifact (copyright statement as quoted in its file listing)
  6. Grafa: Aerometrex secures $1.1M in AI data contracts
  7. AIData.Broker: How AI training data prices are negotiated
  8. AIData.Broker: AI training data prices
  9. Yahoo Finance (CNN): How filming your chores could train the android butlers of the future
  10. Mercor Data: open company listings
  11. Origin Data Partners: AI training data buyers by data type

For discussion only. This presentation describes concepts under evaluation. It is not an offer to sell or a solicitation to buy any security or investment, and nothing in it promises a return. Market figures come from the public sources listed, several of which are published by companies active in this market. The scale illustration in Stage 1 is a planning assumption, and the figures in the Stage 2 example are placeholders. Legal structures require review by counsel.