Methodology and validation status

Validation status: not validated. The Miniato Scale is a methodological proposal, rules pilot-1.0. No inter-rater reproducibility study, public-comprehension experiment or sensitivity analysis has been completed. Nothing on this site is an established standard, a legal determination or an emergency-priority signal.

What the level measures

The level classifies realised consequences up to a declared evidence cut-off, under a declared rules version. It is the maximum substantiated band across six domains (human health, rights, economic assets, operations, biosphere, societal systems), except the terminal levels 9 and 10. Loss of control, autonomy and evidential confidence are recorded and displayed next to the number and never added into it. A hazard, an evaluation-contained incident or a blocked attempt is not a realised harm.

Where evidence runs out, the result is a set of admissible levels (an interval such as 3–4, a lower bound such as ≥ 2, or an explicit set), never a single guessed number and never zero. “No deaths confirmed” is not “zero deaths”.

Source: Miniato Scale preprint v1.0, rules pilot-1.0; Product Charter §7; Constitution, Public Trust Principles 2 and 4.

What Miniato.org adds beyond the paper

The paper’s rules and reference implementation adjudicate nothing: they receive domain levels already established by people. Miniato.org will add product extensions (evidence graph, review states, automated provisional proposals, challenges). These extensions are labelled as such and are not part of rules pilot-1.0. An automated proposal is never presented as human or scientific validation.

Source: Decision Register D-005, D-006; Constitution §9.

Reproducibility

Frozen reference implementation: 61d7007 of santismm/miniato-scale, vendored with SHA-256 provenance.
Deployable engine @miniato/rules 0.1.0, verified value for value against that oracle, including every one of the 531 441 possible domain profiles.
Every assessment records its rules version, engine version, evidence cut-off and provenance; superseded assessments stay retrievable.
Rule changes go through versioned Miniato Improvement Proposals with a corpus impact report, never silent edits.

Source: Rules, Versioning and Reproducibility Policy §§2–4; ADR-0001, ADR-0003, ADR-0005 in the repository.

Known limitations

Thresholds (θ = 10⁷ € per statistical death, w = 0.1 per serious permanent injury) are declared conventions, to be varied in the validation programme, not derived facts.
Coverage is uneven: more reported incidents do not mean a period, company or country is less safe.
Levels are ordinal; a mean of levels is not mean severity. Trends are counted, not averaged.
Causality follows a conservative material-contribution test; enabling or contextual AI use does not receive a canonical level.

Source: Product Charter §§6–8; Decision Register D-010, D-011.

See also the scale and the rules metadata API.