Entry gate
Demand, not budget. A request enters intake only when someone real intends to use the result. Speculative units — built because capacity was idle — bounce here by definition; that is the pull rule doing its job.Standard work
The work order is OPORD-shaped and mission-command-disciplined: the five-paragraph structure — situation, mission, execution constraints, support, command and signal — with the Paragraph 2 test enforced. The mission paragraph states intent and end-state, never method. Method belongs in a skill, and a task body that reads like a script is a task that has failed before it started. Writing the work order is one of the line’s two takt-bound human touches (Thesis 6): this is where the operator spends intent. Everything downstream is machinery for honoring it. A conforming work order carries:- Why, present and linked to higher intent
- Acceptance criteria that are testable
- Unit family declared — skill, tool, or agent
- Target cost, assigned here at design time, where most of the margin is decided (price-minus: the ceiling comes first)
- Named principal
Exit gate
Compiled, per Thesis 4 — this is row 1 of the design’s compilation table: schema validation plus an intake linter; a nonconforming order cannot enter the queue. The inversion test applies: swap What and Why — if the order still reads sensibly, it is an order, not a mission, and it bounces.Escalation
Ambiguity is resolved here or the order does not pass — sharpened with the principal, never guessed at downstream. Ambiguity caught at intake costs minutes; caught at Verify it costs the whole run.Enforcement
workorders/validate-work-order.py — the compiled form of the work-order schema. It fails closed (unparseable is nonconforming) and rejects, among other defects: a missing or placeholder principal, an undeclared unit family, a non-testable end-state, a Situation with no Why linked to higher intent, and a Mission that states method — the Paragraph 2 test, as code. It gauges form deterministically; substance is judged again at Verify. A pass here is a shape check, not a quality verdict.
Self-audit
| Claim | Source |
|---|---|
| Work order as the human-intent/machine-execution interface; pull only; named principal | docs/factory-design.md:97 |
| OPORD shape; Paragraph 2 test; “a task body that reads like a script…“ | docs/factory-design.md:99 |
| Exit gate contents: Why linked, testable criteria, family, target cost (price-minus), principal; inversion test | docs/factory-design.md:101 |
| Compilation row 1: schema + intake linter; nonconforming orders rejected | docs/factory-design.md:194, 267 |
| Intake is one of the two takt-bound touches | docs/factory-design.md:71 |
| Linter behavior: fail closed, error classes, form-not-substance | workorders/validate-work-order.py (module docstring) |