Using this OPD with an agent

opd.yaml is the whole product and its factory in one file. Give it to an agent and it can answer questions about the product, plan a line, or propose changes, with every answer traceable to a source.

Synthetic example: EC-1 is fictional. Use it to try the workflow.

What's in the one file

• card: what the product is, how far to trust this OPD, and its metrics.
• sources: every piece of evidence, each with a hash.
• parts, assemblies, bom, suppliers: what it's made of and what that costs.
• operations: the assembly graph. Each operation names its inputs, its output and its station, so the graph is the routing.
• stations, inspection_plans, tests: the line, how it checks parts, and the test limits.
• pfmea, issues: what goes wrong and how it was fixed.
• design_intent, variants: what the designer wanted, and alternatives scored on one scorecard.
• ext: category detail, here controller inputs and the battery cell.

Every value is a claim with source (measured, observed, public_filing, public_report, inferred, maker_verified), confidence (high, medium, low) and evidence ids. Tell the agent which claims to trust.

Prompts to try

• "Using only claims with source measured or observed and confidence high, list the stations you'd need to build this and the top three yield risks."
• "Rebalance the line to a 25 s takt. Which stations do you split or merge, and what equipment do you add? Use the operations graph and the cycle-time claims."
• "Which issues have a fix that's implemented, and what evidence proves it went in?"
• "We want a 0.10 mm shoulder gap. What does the variant study say it costs, and would you do it?"
• "Write the IQC plan for a new stick module supplier, using the existing sampling plans as the template."

Rules for agents

• Say which claims an answer rests on, and their confidence.
• Treat inferred and low as hypotheses, not facts.
• Never invent a value the file doesn't have. Say it's missing.
• card.synthetic: true means nothing here describes a real product.