Jump to text
On this page
Copy and source
Published source confirmed at last check

Source changed 2026-11-22 13:41:52 UTC · snapshot created 2026-11-22 13:42:49 UTC · last check 2026-11-22 14:10:20 UTC

A.19.UINDM:5 - Archetypal Grounding (informative)

A.19.UINDM:5.1 - Tell

Use UINDM to select the basis positions for a policy-bound projection:

  • Input: “the declared characteristic basis for this bearer, claim scope and selected slices, qualification window, evidence basis, and intended use, plus an explicit indicator choice policy”
  • Output: “the selected positions of this basis, each retaining its Characteristic, Scale and meaning for downstream use”

UINDM selects basis positions. A projection to those positions retains their coordinate values and Scale meanings.

A.19.UINDM:5.2 - Show (U.System) — cross‑unit engineering dashboard

A program manager maintains a DeclaredCharacteristicSpace for manufacturing sites, including ~30 characteristics (quality, safety, cost, throughput, sustainability).

  • The CN‑Spec’s indicator_policy for the “weekly executive dashboard” selects a subset: {DefectRate, IncidentRate, UnitCost, LeadTime, EnergyPerUnit, OnTimeDelivery}.
  • In this example each named Characteristic has one basis position. UINDM resolves the six names to those positions and returns S; each position retains its original Scale.
  • One site lacks reliable incident reporting for the last week. This example’s evidence-gated policy permits retaining the fixed six-position set under degrade, with that gap explicit. IndicatorizeEligibility therefore returns degrade, and the audit retains the effective MinimalEvidence and edition pins. The selected IncidentRate position supplies no missing value; a downstream comparison depending on that value follows its own insufficient-evidence rule.

Downstream mechanisms can now be held to the invariant: they may only score/compare/select using the declared indicator profile (or abstain/degrade). This avoids “dashboard drift” where different teams silently score on different subsets.

A.19.UINDM:5.3 - Show (U.Episteme) — robust evaluation across environments

A research lead wants indicators for model robustness under distribution shift (different hospitals, sensors, geographies).

  • The declared characteristic-space basis includes many candidate metrics (accuracy slices, calibration, subgroup error, OOD detection quality).
  • The indicator choice policy is “invariance-driven”: prefer indicators whose meanings remain applicable across the selected environments; deprioritize proxies with an established environment-sensitive interpretation. For example, a constructed policy retains subgroup-error positions only when the subgroup definition, target label and error criterion match the intended comparison in each environment.
  • UINDM returns the positions resolved by that policy. An action-changing unresolved match yields no selection. Once the positions are resolved, insufficient evidence follows the declared degrade/abstain rule. Keeping the same error definition does not establish stable model performance under distribution shift; the later evaluation needs the values and evidence for that claim.

A.19.UINDM:5.4 - One Characteristic at two Scale positions

Let the space declaration CS7 have ordered positions i1=(Temperature, Celsius) and i2=(Temperature, kelvin), followed by an unrelated cost position. A state contains x_i1=20 °C and x_i2=293.15 K. The policy “select Celsius temperature” returns S={i1}. The all-matches policy “all Temperature positions” returns {i1,i2}, with x|_S in the original order and both Scale meanings retained. UINDM converts neither value. This use retains the reference to CS7 and a separate characteristicSpaceDescriptionRef to the episteme defining that basis.

For the first policy, a separately admitted USCM scoring use can read the Celsius value under its declared method g_C(c)=(c-0)/(40-0) on Celsius interval [0,40], returning 0.5 at 20. The interval endpoints are part of this illustrative scoring declaration. Supplying 293.15 as though it were Celsius violates that method’s input basis; a TemperatureRef alone would not reveal the error. All-matches selection requires a receiving method that accepts those two Scale-specific positions, or a narrower selection; it is not a universal default.

Suppose a different space CS8 reverses these two positions. Select CS8 and its description episteme, preserving CS7, its description and the earlier result. Resolving “Celsius temperature” now selects its CS8-local position i2. Reusing CS7’s naked integer 1 would select kelvin and is invalid. If the policy says “Temperature” and supplies no rule for the two matches, return its unresolved/abstain disposition. A missing indicator policy retains the defined abstain result, and evidence gating still applies only when the selected policy requires it.