Skip to content

Measurement Scaling & Decision Boundaries

v9 compatibility note: This page documents the retained v6 portable-state semantics. V9 preserves the contract as a frozen substrate and may wrap/project it through Logical Artifacts without redefining it. See Version-Stack-v6-v7-v8-v9.

Kristal v6 separates the shape of a value from the decision derived from that value.

A binary action does not imply that the underlying world is binary.

Value shapes

Kristal valuations may use several semantic forms:

Shape Example
boolean document received: true / false
categorical native / introduced / cultivated
set applicable dimensions = {energy, information}
ordinal low < medium < high
scalar wear = 0.72
interval pressure = 350–400 kPa
probability P(cause | observations) = 0.72
distribution {cause A: .55, B: .30, C: .15}
vector support + severity + cost + testability
partial order A cheaper, B faster; no universal winner
state machine draft → review → approved
temporal instant / interval / duration / recurrence

Binary is often a projection, not the source data

For example:

risk = 0.37

may be mapped by policy to:

< 0.20        automatic
0.20–0.60     human_review
> 0.60        human_decision

The thresholds belong to policy. The scalar remains the measured representation.

Ordinal grayscale

The historical Kristal certainty scale was an ordinal grayscale rather than a true continuous analog value.

unknown / speculative / low / medium / high / established

The important property was ordering, not equal numerical distance.

In v6 this becomes explicit:

dimension = diagnostic_support
value_semantics = ordinal
value = high

A different family might use:

dimension = constitutive_necessity
value = high

Same ordinal shape, different semantic dimension.

Meta-states are not low values

These must remain outside the measured scale:

  • unknown — dimension applies, value unresolved;
  • not_applicable — dimension does not apply;
  • indeterminate — available information cannot produce a stable value;
  • not_measured — dimension applies but has not been measured.

Never map these blindly to zero.

Multi-dimensional values

Many operational decisions should not be collapsed into one score.

Example diagnostic candidate:

support      = 0.82
severity     = 0.95
testability  = 0.60
cost_of_error= 0.90

The policy may use all four to route work. A single “health score” or “risk score” can hide important tradeoffs.

Partial order and incomparability

Some alternatives are not naturally rankable without organizational preferences.

A = cheaper, slower
B = faster, more expensive

Kristal should preserve that incomparability instead of forcing an arbitrary total score.

Scaling without semantic breakage

A valuation scales safely when:

  1. the dimension remains explicit;
  2. units / semantics remain stable;
  3. unknown states are preserved;
  4. projections are documented;
  5. thresholds live in policy rather than being baked into observations;
  6. richer values can be down-projected for legacy consumers without changing high-stakes meaning.

Actionability is a derived boundary

nuanced state
    ↓
measurements / categories / rules
    ↓
policy
    ↓
actionability
    ↓
automatic | human_review | human_decision | prohibited

This lets Kristal preserve a nuanced world while still making operations simple where they genuinely are simple.