G.7:4.5 - Worked mini‑examples (informative; post‑2015; row scopes + loss notes)

These worked rows use illustrative calibration values and source scopes. Actual calibration needs its stated evidence. Each receiving use still has a separate rule and loss tolerance.

1. Preference‑learning objective (Method; RowScope = “training‑objective‑intent”). Cells: RLHF@Context‑A ↔ DPO@Context‑B ↔ IPO@Context‑C RowCL_min: 2 (calibrated bounded support in this worked case) Loss notes: different inductive biases (reward model vs direct preference likelihood; sensitivity to preference noise model; implicit regularisation forms). Proposed use: a didactic comparison of objective intent. Its separate claim must limit the comparison to that intent and retain the listed differences; method eligibility and acceptance require their own rule.

2. Robustness evaluation (Measurement; RowScope = “metric‑family‑intent”). Cells: Accuracy@IID ↔ Robustness@ShiftBench (e.g., distribution‑shift benchmarks common in post‑2019 practice) RowCL_min: 2 Loss notes: shift taxonomy differs; comparability depends on pinned protocol editions and window selection; “robustness” is not a scalar substitute for accuracy.

3. Quality‑Diversity archive comparability (Measurement; RowScope = “DescriptorMap‑only”). Cells: MAP‑Elites grid indices ↔ CVT‑MAP‑Elites centroids ↔ CMA‑ME archive RowCL_min: 2 Loss notes: discretisation vs centroidal tessellation; archive pressure differs; drift occurs if DistanceDef or insertion policy changes. Proposed use: cross-reporting only the named descriptor-map relation under explicit edition pins and an affirmative bounded-use claim with passing reliance. Edition pins alone do not make the telemetry comparable.

4. Open‑ended transfer semantics (Method; RowScope = “transfer‑rule intent”). Cells: POET‑class transfer rule ↔ Enhanced‑POET‑class transfer rule ↔ “modern open‑ended transfer variants” RowCL_min: 2 Loss notes: environment validity region differs; transfer timing and selection pressures differ; pinning transfer rule editions is mandatory for audit.

Paired receiving case — Vehicle to TransportUnit. Use C.3.3 §9.1’s exact source and target kind declarations, pinned scheme editions, registryAPI v1.4 and selected time window. In row VehicleTransportOrder, record the obtaining KindBridge, preserved PassengerCar/Vehicle subkind order and collapsed EV distinction, with the reported CL^k=2 and battery-health loss. This is the kind channel; an F.9 sense Bridge is added only if the receiving claim separately relies on one.

For a G.5 shortlist of independently admitted Methods for a transport review, suppose the applicability criterion uses only the preserved transport/passenger order and explicitly ignores propulsion. The row can support that narrow applicability comparison after receiving admissibility, fresh target classification of the subject vehicles and the matching evidence-reliance result pass. The Methods’ other eligibility criteria remain applicable. For a battery-health review whose Method-selection rule needs EV/battery information, the same correspondence fails that use because the required distinction is lost. A favourable CL value or a waiver cannot supply the battery premise. The correspondence and calibration can stay unchanged while these two use conclusions differ. Use these two questions as a paired RegressionSet probe when this row is reused: recover the transport-order premise for the first and expose the missing battery premise for the second. Recheck the affected use after its criterion or the row’s preservation/loss basis changes.

In C.3.3 §9.3’s AdultPatient/AdultPerson_Y case, the age-boundary loss and CL^k=1 remain evidence about the kind correspondence. An authorized policy exception does not supply an unresolved date of birth; the receiving classification stays unknown.

Choose pins for the actual use. These cases apply the same conditions to a compact result and to the fuller kit: