{
  "schemaVersion": "ess-legitimacy-distribution-plan/v1",
  "storyId": "ESS-03",
  "version": "1.1.0",
  "status": "registered",
  "registeredOn": "2026-09-19",
  "parentAnalysisVersion": "1.0.0",
  "parentAggregateSha256": "cb9015dccf7ff3b9f8b9cd2c38bcb157b82dcfbfc64c305ea845ae391b502fde",
  "inferenceStatus": "Secondary exploratory extension proposed after inspection of the original mean-model results. Registration fixes this extension before its estimation; it does not turn the original exploratory question into independent confirmation.",
  "rounds": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11],
  "modelRound": 11,
  "variables": ["economic_satisfaction", "democratic_satisfaction", "age", "gender", "education_years", "financial_strain", "unemployed"],
  "sourcePolicy": "Reuse the exact eleven releases, questionnaire mappings, reviewed full designs and private prepared core. No new source edition, raw-source mutation or local blocked survey engine.",
  "weights": "Original ESS anweight, with the existing historical pspwght times pweight derivation. No equal-country reweighting.",
  "descriptiveDomain": "Both satisfaction answers observed and valid on the integer 0-10 scale. Histograms and joint probabilities share that paired denominator; conditional probabilities use the paired economic-score domain.",
  "descriptiveScopes": {
    "countryRounds": 204,
    "allCountryRounds": 11,
    "fixedCountryRounds": [9, 10, 11],
    "fixedCountries": "Use the existing reviewed 19-country intersection, not a new country registry.",
    "totalScopes": 218
  },
  "descriptiveOutputsPerScope": {
    "marginalProbabilities": 22,
    "jointCells": 121,
    "conditionalCells": 121,
    "democraticCdfCutoffs": 10,
    "conditionalCdfCells": 110,
    "responseCoverage": ["paired-among-eligible-design", "complete-among-paired"]
  },
  "proportionUncertainty": {
    "method": "Full-design stratified-cluster Taylor linearization with the existing adjust/adjust-domain-lonely policy.",
    "interval": "Pointwise 95% logit-transformed t interval for interior probabilities; report the standard error on the probability scale.",
    "domainNMinimum": 50,
    "lowPrecisionDomainN": [50, 99],
    "sparseEventThreshold": 10,
    "sparseEvents": "Flag when either the event or its complement has fewer than ten observed records; this is distinct from the denominator-based precision flag.",
    "boundary": "An observed weighted probability of zero or one is a sample boundary, not population certainty. Show the observed value and support but no manufactured zero-width interval.",
    "empty": "Retain empty and suppressed domains explicitly; never convert unavailable probability to zero.",
    "coherence": "Marginals, joint cells, conditional cells and cumulative probabilities must reconcile before and after serialization. All observed CDFs must be nondecreasing."
  },
  "probabilityModels": {
    "status": "registered",
    "family": "binomial",
    "link": "logit",
    "modelCount": 10,
    "modelIds": ["cdf-0", "cdf-1", "cdf-2", "cdf-3", "cdf-4", "cdf-5", "cdf-6", "cdf-7", "cdf-8", "cdf-9"],
    "domain": "Exactly the original 46091 complete Round 11 records; keep the full eligible design for covariance.",
    "predictors": "The original adjusted-spline design: country fixed intercepts, the three registered economic-satisfaction basis columns, age and age squared, source gender, fractional schooling, financial-strain category effects and main-activity unemployment.",
    "interactions": "No country or financial-strain interactions in the cutoff suite. Those questions have separate Gaussian models.",
    "optimizer": { "tolerance": 1e-12, "maximumIterations": 200 },
    "cutoffs": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],
    "defaultCutoff": 3,
    "economicScores": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
    "interpretation": "Probability of a democratic-performance rating at or below the chosen cutoff. A low rating is not rejection of democracy.",
    "targets": ["Each of the 30 included countries", "ALL: all 46091 complete records", "FIXED: complete records in the existing 19-country recent set"],
    "standardization": "For each target, hold its original-anweight complete-case empirical covariate distribution fixed. Replace economic satisfaction and its three basis columns at each registered score, preserve every reference respondent's other covariates and country, and average inverse-link probabilities. Do not evaluate the inverse link at mean covariates.",
    "standardizationUncertainty": "Average individual response-scale gradients with the same fixed target weights, then use the complete corrected coefficient covariance. These intervals condition on the declared empirical reference distribution; they do not additionally estimate uncertainty in that reference composition.",
    "interval": "Pointwise 95% model-residual-df logit-transformed t intervals using the delta-method variance of the standardized probability.",
    "contrasts": "For each cutoff and target, the two registered economic-score differences and their upper-minus-lower difference. Use differences of standardized gradients and the same model covariance; report probability-point units.",
    "coherence": "Check unrounded predictions at all eleven economic scores and all ten cutoffs for each target. All probabilities must be strictly interior and nondecreasing across cutoffs. If any crossing occurs, the entire ten-cutoff target CDF is unavailable; do not selectively omit cutoffs, clip, sort or project.",
    "coherenceFailure": "Keep rejected raw predictions and covariance diagnostics only in private reference material. Publish the target's unavailability and crossing count, with no fitted target curve or contrasts. Individually validated coefficient matrices remain a model audit, not a coherent-CDF claim.",
    "crossCutoffUncertainty": "Separate fits do not supply cross-cutoff covariance. Do not difference intervals to create category-mass intervals or claim simultaneous coverage.",
    "fitGuards": ["approved-public-svy-binomial-api", "finite-coefficients", "full-rank-fisher-information", "weighted-score-solved", "iterations-strictly-below-maximum", "ordinary-psu-covariance-agrees", "corrected-covariance-positive-semidefinite", "positive-residual-design-df", "independent-R-reference"],
    "fitFailure": "Convergence, separation, rank or covariance failures stop acceptance rather than becoming silently dropped countries or success-shaped fallbacks. Any later alternative method needs an explicitly revised registered plan.",
    "publicApi": "svy0.27.0 sample.glm.fit(y,x,intercept=True,family='binomial',link='logit',where=complete,drop_nulls=False,tol=1e-12,max_iter=200). The pinned stack has no independently supported ordinal family.",
    "legacyCompatibility": "New code must retain byte-identical v1 owners. Reuse their runtime/domain/covariance guards; add canonical-logit influences in a new module.",
    "executionGate": "This final model, target, support and coherence contract is registered before any extension estimates are computed."
  },
  "heterogeneityModels": {
    "countryShape": {
      "domain": "Round 11 paired responses",
      "formula": "Country intercepts and interactions of country with all three registered economic-satisfaction spline columns.",
      "question": "Does the common-shape restriction hold across the included countries?",
      "comparison": "Country-specific [m(8)-m(5)]-[m(5)-m(2)] contrasts with their full-model covariance; the original common-shape estimate is a reference, not an independent comparator.",
      "test": "Joint design-based Wald test of all country-by-spline interaction terms."
    },
    "financialStrainShape": {
      "domain": "The same 46091 complete Round 11 records used by the original adjusted model",
      "formula": "Country intercepts, the original demographic and main-activity covariates, financial-strain category effects, and financial-strain interactions with all three registered spline columns.",
      "question": "Does the conditional association differ by reported household financial difficulty?",
      "comparison": "Four group-specific curvature contrasts evaluated under the same fixed covariate reference, with model n and actual group support distinguished.",
      "test": "Joint design-based Wald test of the financial-strain-by-spline interaction terms."
    },
    "basis": "Retain the original fixed knots 0, 3, 7 and 10; do not select a breakpoint after observing results.",
    "intervals": "Pointwise design-aware 95% t intervals. Do not interpret overlapping or nonoverlapping intervals from different fits as a between-model test.",
    "multiplicity": "Exploratory joint tests and explicitly labeled pointwise intervals; no significant-country ranking or simultaneous coverage claim.",
    "failures": "Rank, convergence, design-degree or support failures are explicit unavailable results only when anticipated by the final plan, never silent dropped-country or iid fallbacks."
  },
  "history": "Descriptive distributions cover the same eleven separate waves. The recent fixed-country view retains the same 19-country membership in Rounds 9-11. This does not hold demographic or population-weight composition constant and is not a respondent panel.",
  "reference": {
    "runtime": "Pinned independent R survey 4.5",
    "absoluteTolerance": 1e-9,
    "relativeTolerance": 1e-7,
    "reproductionRuns": 2,
    "requirements": ["synthetic-singleton-and-domain-fixtures", "all-published-probability-estimates-and-intervals", "model-coefficients-and-full-covariance", "all-predictions-and-contrasts", "sample-and-scope-identities", "cdf-coherence", "aggregate-only-output"]
  },
  "serializationSignificantDigits": 10,
  "publicBoundary": "Only schema-validated aggregates and their receipts may leave the approved research runner. No respondent, household, sampling-unit or credential fields.",
  "publicationAllowed": false
}