CJI Validation

Testing the axis conventional indices miss.

The Contributive Justice Index is tested as an orthogonal stress measurement: whether a society can look prosperous and formally democratic while losing the contributive, agency, and common-life conditions that make institutions work.

Honest status

Directional validation is live. Measured-panel findings are not claimed yet.

The current validation package proves method logic, internal coherence, orthogonality, cross-validation behavior, and preregistration mechanics. It does not yet license public real-country CJI scores as findings.

Open sealed CJI commitments
44
directional country panel
33%
independent variance after V-Dem plus prosperity
0.994
blind-spot in-sample / leave-one-out correlation
44/44
blind-spot verdicts sign-stable under leave-one-out

Validation Stack

The website should show the method as something under pressure: orthogonality, retrospective probes, blind-spot cross-validation, negative controls, and sealed future commitments.

Core Thesis

Orthogonal Stress Axis

CJI is tested as a distinct stress measurement, not as a renamed GDP or formal-democracy score. The 44-country panel reports that V-Dem and prosperity together leave about one third of CJI variance unexplained.

core_thesis.pyorthogonality_test.pytest_core_thesis.py
Taxonomy

Conversion Class Validation

The product is not only a number. The engine classifies how stress converts: capture, rupture, constraint-stress, recomposition, delayed conversion, dependency, legitimacy-stress, resilience, or monitor.

panel_class_validation.pyclassifier_audit.pyannual_release.py
Blind Spot

Concealed Stress Gap

The blind-spot gap asks how much stress conventional prosperity and democracy profiles would conceal. Leave-one-out cross-validation tests whether that gap survives holding countries out of the baseline.

blindspot.pyblindspot_cv.pyblindspot_robustness.py
Retrospective

Historical Probe Suite

Retrospective probes are used carefully: they find method limits, check ordering, test matched pairs, and expose bugs. They are not treated as the strongest predictive evidence.

retro_test.pycascade_test.pychain_leadlag_test.pymatched_pair_test.py
Controls

Negative And Split Profiles

Denmark tests the down direction. Singapore tests whether a collapsed dimension gets hidden by an average. The method must not become a high-score machine.

negative_control_denmark.pysplit_profile_test.pysensitivity_test.py
Future Credit

Pre-Registration And Retraction

The strongest evidence comes later: sealed CJI class predictions, fixed resolution criteria, Brier scoring, and annual retraction thresholds when enough calls resolve.

prereg.pycanonical_criteria.pyretraction_monitor.py

Orthogonality

This is the core validation claim. CJI must carry information that formal democracy and prosperity do not already contain.

Core result
33%

Independent CJI variance after fitting conventional formal-democracy and prosperity proxies on the directional 44-country panel.

V-Dem alone
Explains about 60% of CJI variance; leaves about 40% missed.
Prosperity alone
Explains about 55%; leaves about 45% missed.
V-Dem + prosperity
Explains about 67%; leaves about 33% independent.
Twin tests
US/Germany and Denmark/France look similar on conventional axes but diverge sharply on CJI stress.

Retrospective Tests

Retrospectives are useful when they are humble. They are used to expose bugs, test the direction of the method, and separate stress from conversion pathway, not to claim hindsight-proof predictive power.

France divergence case
CJI can flag high stress in a formally healthy democracy without mislabeling it as captured.
Denmark negative control
The method reads a genuinely healthy democracy as low stress after the amplification correction.
Poland / Hungary matched pair
Similar stress separates into recoverable versus captured classes through veto-point and resilience differences.
US chain lead-lag
Contributive-justice stress appears before democratic-agency stress in the directional 1999-2020 probe.
Cascade test
Chain lead, flag, and high-stress diagnosis fire in order on the US probe, but this remains secondary internal-coherence evidence.

What These Tests Do Not Prove

The public validation surface is stronger if the limits are visible.

Not measured-panel validation yet

Most tests use directional codings from the public record. Real measured validation needs source-bound country-year indicators and frozen reference panels.

Not proof of causation

Lead-lag and ordering do not prove that moral-infrastructure stress caused the later outcome. Rival drivers remain possible.

Not a public country-score product yet

Any real-country CJI number remains a calibration artifact until connector coverage, expert coding, confidence, panel hash, and the 2027-2030 windows mature.

Prediction is secondary

Pre-registration is the accountability layer. The working product today is orthogonal stress measurement plus conversion classification.

Country-Score Release Guard

No public or API surface should present a real-country CJI number as a finding unless the score carries the full measured-panel proof envelope.

Frozen measured-panel hash

The country-year score must name the exact reference panel used to transform the cohort.

Payload hash

The scored country payload must be reproducible from the exact input envelope.

Connector coverage report

Every emitted source-bound indicator must be present, absent with a stated reason, or expert-coded under the protocol.

Expert-coded cell evidence

The nine judgement-coded cells need citations, coder/review status, and quality.

Confidence band

The score must carry weakest-link confidence, not only a point number.

Validation status and §16.4 caveat

The public surface must state whether the score is validated, pending, directional, or only a calibration artifact.

Path To Stronger Evidence

The public story should show where validation goes next, not pretend the destination has already been reached.

1

Assemble measured panel

Bind every indicator to real source coverage, expert-codebook cells, confidence bands, and a frozen panel hash.

2

Publish validation runs

Expose the annual stress-class release, blind-spot gap, error notes, and retraction monitor without turning calibration artifacts into findings.

3

Resolve preregistered calls

Let sealed commitments mature against canonical events, then report Brier, recall, false positives, and retraction thresholds.