Data dictionary

The definitional layer of GII-US: scorecard dimensions, the Decision Impact Record contract, decision-origin codes, and lifecycle states. Field-level JSON Schema ships with the pilot; the definitions below are normative now.

Scorecard dimensions#

Stable identifier#

GII-US-[YEAR]-[ORIGIN]-[TYPE]-[SERIAL]-[VERSION]
  e.g. GII-US-2027-EXE-AGY-RULE-0042-v1   (illustrative only)

A materially revised analytic record increments its version while retaining root event identity; distinct implementing events receive distinct identifiers linked to the same lineage, so they are different events rather than versions of one. External source identifiers (including PEPI-native IDs) are preserved verbatim in the adapter layer only and are never surfaced in GII records.

The shape is enforced, not merely published. A record whose identifier is malformed — or whose embedded origin or instrument type disagrees with the record’s own — fails the quality controls and is not scored. Note that an origin code itself contains a hyphen, so an identifier cannot be split on hyphens and read positionally: EXE-AGY-RULE read that way gives origin EXE and type AGY, which looks entirely plausible and is wrong.

Decision origins#

Instrument types#

A closed enumeration, and the second column is a rule rather than a description. Watch-stage instruments receive no observed-impact score — they may be stored and tracked, and the headline is barred categorically, whatever else the record satisfies. An instrument kind that is not on this list is treated the same way: a kind nobody has classified is a kind nobody has decided about.

The five clocks#

Government impact does not run on one date. Each clock must fall on or after the one before it, and a record whose outcome is measured before the decision that caused it is refused rather than reported — that ordering describes an impossibility, and an impossibility in a dataset is nearly always a data error. The one permitted reversal is anticipation: affected parties really do act on an announced change before it takes effect, and a record claiming it must name anticipation as a threat in its causal statement.

National series attribution runs on the exposure clock, not the publication date. A decision with a long implementation lag does not begin contributing on the day it issued; it begins when someone felt it.

The required causal statement#

Every observed-impact record states these before data analysis, and an observed headline is withheld without them. They are separate fields rather than one prose block on purpose: nobody can say which of the eight a paragraph omitted, and everybody assumes it covered them.

The last of these — the conditions under which the attribution would be weakened or rejected — is the one most likely to be skipped, because it is the only one that asks the author to describe how they might be wrong. It is required, and its absence withholds the headline.

Decision Impact Record — field groups#

Lifecycle and maturity states#