No moving goalposts
Primary outcomes, horizons, thresholds, and exclusion rules are frozen before evaluation.
FRAMEWORK 1.0 / Falsifiability
Falsifiability
A framework becomes scientific only when it states what evidence would count against it.
Claim register
| ID | Registered claim | Evidence against the claim | Required response |
|---|---|---|---|
| H1 | The multiplicative C × Lₑ × Eₙ relation predicts bounded stability better than independent additive effects. | Repeated preregistered tests show no incremental out-of-sample value or inferior calibration across the declared domain. | Reject the multiplicative default or narrow it to supported domains. |
| H2 | Very low trust-legitimacy constrains returns from added usable capacity. | Capacity gains produce stable benefits independent of Lₑ across credible identification designs. | Remove or revise the threshold/complementarity claim. |
| H3 | Rising entropy pressure predicts longer recovery after comparable shocks. | Matched-event and survival analyses show null, reversed, or non-replicable association. | Redefine E, revise mechanism, or reject H3. |
| H4 | Unequal distribution lowers durable stability relative to equal aggregate capacity. | Multilevel models show no consistent distributional effect after measurement and selection corrections. | Narrow the claim and publish the failed comparison. |
| H5 | Protected dissent and peaceful correction improve recovery after shocks. | Institutional comparisons and event studies show no benefit or systematic harm under declared conditions. | Revise the correction mechanism and domain. |
| H6 | Greater network modularity reduces cascading failure without eliminating coordination. | Network simulations and infrastructure cases show greater cascade risk or no robustness advantage. | Reject universal directionality; specify topology-dependent conditions. |
| H7 | A defined CCWS ensemble improves calibrated warning over domain baselines. | Prospective blind evaluation fails calibration, lead-time, false-alarm, or utility thresholds. | Do not deploy; revise or discontinue the ensemble. |
| H8 | Trust-building interventions improve cooperation only when perceived procedural fairness also rises. | Credible field or natural experiments show robust cooperation gains without fairness change. | Revise the moderation claim. |
Decision rules
Failure is publishable
Null, negative, reversed, and replication-failure results remain in the public record.
Claim-specific status
Evidence upgrades or downgrades a specific claim and version—not the founder, organization, or entire field.
Ethical failure counts
Systematic rights violations, discriminatory harm, surveillance abuse, or inability to abstain can disqualify practical use even when prediction is accurate.