Skip to content
DRAFT⚠OIML SMART pilot programme · internal use only · all documents and specifications are drafts and may change without notice
OIML SMART

Tier 3, the chain (a day)

The supply chain of models: reference models, implementation models, and the mappings between them.

Tier 3, the chain (a day)

Tier 2 taught you the claim. This tier walks the claim’s whole enforcement path: requirements → tests → forms → evidence → verdicts → certificates. Every link is data, and every link can be inspected at runtime, you will drive the full path against a live twin in the Twin Lab and end with a test report that proves every link.

Learning outcomes. At the end you can:

  • trace one requirement from its clause to a verdict, naming every intermediate artifact;
  • perform a guided test run: environment legs, attested readings, the twin auto-capture, the compiled evidence form;
  • read a verdict’s basis, the resolved inputs, the freshness window, the override rule;
  • explain why requirements/tests/forms are secondary models, derived from the subject’s IS/HAS/DOES.
REQUIREMENT   a constraint on the subject's IS/HAS/DOES, with a limit
              (an OCL expression over declared quantities) and clause
              provenance
     ↓ tested by
TEST          a procedure (a process) that probes the subject —
              declared steps, preconditions, acceptance criteria; an
              INVALID run (a violated precondition) voids, never fails
     ↓ records into
FORM          the evidence interface — one form per test, fields
              bound to the entity graph, computed fields evaluating
     ↓ becomes
EVIDENCE      the captured readings: attested values (of record) and
              twin-served values (auto-captured, with the pair skew)
     ↓ judged to
VERDICT       the requirement's limit evaluated against the bound
              evidence — pass | fail | invalid | indeterminate, with
              the resolved inputs snapshotted (the re-judgment basis)
     ↓ aggregated to
CERTIFICATE   the determination's output — every applicable
              requirement determined, full coverage required

Two honesty invariants hold at every link: invalid voids, never fails (a wrong setup is not an instrument defect), and stale degrades to indeterminate (silence is not evidence, a value without a fresh timestamp judges nothing).

Hands-on, drive the whole chain in the Twin Lab (half a day)

Setup: the app (cd browser && npm run dev) and an SST from the two split checkouts (primmel/sst, the framework, next to oimlsmart/sst, the instrument library, cloned as sst-instruments): cd sst && npx tsx packages/runtime/sst-runtime/src/bin.ts run ../sst-instruments/packages/instances/acme-lc500 5290.

  1. Bind (/app/twin-lab): pick R 60, enter the SST’s /twin URL, Discover, read the proposed binding (the serves, the freshness windows, the command face). Bind.
  2. One cycle (the compliance monitor): the verdict stream fills , requirements reading at least one bound register judge, live.
  3. The guided run (the chain’s heart): pick the MDLO test. The interpreter reads the test’s declared steps verbatim:
    • Environment legs, “Set the chamber to 20 °C (click when done)”. Drive the SST’s world channel (the practice buttons , note the honesty banner: practice- ids, never certification evidence) or act on a physical instrument, then click.
    • Measurement rows, place the calibrated load, type the PHYSICAL display’s reading (the evidence of record); the twin’s serve auto-records with its timestamp and the pair skew.
    • Commands, issue the model’s invoke operations from the leg.
  4. Compile: the run lands in the Recommendation’s OWN evidence form (fields prefill, declared values never re-typed), the form’s computed fields evaluate, and the verdict chain judges the test’s target requirements. Read the report block.

Make it (the checkable artifact). Your compiled form instance + the target verdicts, printed by the page. The scripted equivalent to verify against:

cd browser && npx vitest run --config vitest.e2e.config.ts e2e/twin-lab.e2e.ts

, the whole workbench walked green: connect → compliance → guided run → compiled evidence → verdicts.

The why, in one paragraph

A certificate is not a judgment call, it is the aggregate of verdicts, each computed from evidence, each traceable to a clause. Any link you pull on unwinds: verdict → resolved inputs → the form’s computed values → the captured readings → the test’s declared steps → the requirement’s limit → the clause. That is why the platform can re-judge a window under a NEW limit later from the same stored snapshots (INV-5), evidence accrues, never rewrites.

Depth: the twin lab chapter and the operator guides on the OIML SMART site (the lab tester’s walkthrough).

Next: Tier 4, authoring.