Tier 4, authoring (a day)

The final tier: you stop reading models and write one, a small standard as a Primmel package, with the linker and the coverage gates as your feedback loop. The pattern is the OIML classroom’s (smart-r60, the reference instance that teaches R 60 exactly this way): author in Primmel, regenerate, let the gates grade you.

Learning outcomes. At the end you can:

• structure a package (manifest, quantity register, subject anatomy, requirements) from nothing;
• read the linter’s findings as a teacher (every rule names what it wants, with the doctrine’s section);
• use the coverage gates (congruence, the text-coverage budget) as the authoring feedback loop;
• say when a model is done: the gates green, the coverage honest.

The shape of a package

package {
  id my-standard
  kind rec
  version "2027"
  editions { 2027 }
  baseUrn "urn:example:my-standard:2027"
  uses { }
  requires { }
  status preview
  default_spelling eng-Latn
  description "My toy standard — the tier-4 artifact."
}

quantity_register si {
  kind mass { dimensions { M 1 } si_unit kg }
  unit kg { symbol "kg" kind mass factor 1 }
  kind dimensionless { si_unit "1" }
  unit dimensionless { label "dimensionless" kind dimensionless }
}

attribute_definition e_min { quantity_kind mass unit kg scope model }
attribute_definition d_min { quantity_kind mass unit kg origin measured scope sample }

subject MyInstrument {
  is {
    metadata { name "My instrument (fictional)" }
    design_parameters { e_min : mass }
  }
  has {
    attributes { d_min : mass test_dependent }
  }
}

requirement /req/metrological/measuring-range-min {
  name "Minimum load of the measuring range"
  statement "The smallest load applied during test shall not be less than E_min."
  binds_to { sample.test_context.d_min model.parameters.e_min }
  limit {
    expression "ocl{sample.test_context.d_min >= model.parameters.e_min}"
    uses { sample.test_context.d_min model.parameters.e_min }
  }
  acceptance_criteria {
    type: "threshold"
    description: "Minimum test load shall be at least E_min"
  }
  verification { method examination description "Verified in the type evaluation program." }
}

Every piece you met on the earlier tiers, authored now: the register (quantities, never bare numbers, INV-1), the subject (IS/HAS/DOES), the requirement (a constraint with a limit and a verification path).

Hands-on, author and be graded (a day)

1. Create primmel-packages/my-standard/ with the two files above (package.primmel, model.prl).

2. Grade it with the kernel:

   cd ../primmel/primmel-ts/packages/primmel
   npx tsx scripts/check.mts <path-to>/my-standard

   Zero errors, zero warnings is the bar. Read every finding as a teacher: the check id names the rule; the message names what it wants.

3. Break it on purpose, once: delete the attribute_definition d_min line and re-run. Two errors teach you the resolution discipline (binds_to and limit.uses resolve against definitions, not exhibited attributes). Restore it.

4. The classroom pattern. Look at the reference instance (~/src/oimlsmart/smart-r60/): the classroom syncs the data trees on load and runs its unit specs against them, your package is teachable the moment the gates pass.

Make it (the checkable artifact). Your package, lint-clean, plus one sentence per finding you fixed, that is the author’s log every real package in this program carries (each shipped package’s history is exactly such a log).

When a model is done

• primmel check is silent at default AND --strict --audit.
• Every reference resolves; every requirement has a verification path (the C5 honesty: no undeclared coverage gap).
• The text you didn’t model is a named gap with a reason, the coverage budget only burns down (C72), never quietly grows.
• The package regenerates byte-clean through the SSOT loop (npm run test:ssot on the program side).

Depth: the authoring volume on the OIML SMART site, and the the classroom’s own pattern (the smart-classroom-r60 repo, private during the pilot).

You have finished the curriculum: concept → hands-on → duality → chain → authoring. The volumes beside this section go deeper on every tier.