Mass, Direction and Historical Polity Breakdown

T-002 tested a narrow implication of Probability Fields at the scale of states.

PrF distinguishes Mass — a structure’s effective capacity to make outcomes real — from Direction — where that capacity is being applied.

The test asked whether a preregistered empirical proxy for a state’s Direction adds useful information about its future beyond a proxy for its material Mass alone.

Material capability was represented by a Material Mass Proxy (MMP) based on the Correlates of War Composite Index of National Capability.

MMP is not PrF Mass itself. It is a narrow empirical proxy for one material expression of Mass.

Direction was represented by an Institutional Direction Proxy (IDP) combining four institutional measures: public-goods orientation, consultation, impartial administration and stewardship of public resources.

IDP is not PrF Direction itself. It is a narrow empirical proxy for one possible institutional expression of Direction.

The primary question was:

Among states with substantial material capability, does a higher Institutional Direction Proxy — representing a more Forward-oriented institutional expression of Direction under this operationalisation — predict a lower probability of structural breakdown over the following 50 years?

The test used historical state-level data from COW, V-Dem and Polity5.

T-002 was preregistered publicly on OSF before breakdown outcomes were accessed. The predictors, outcome rules, validation procedures, feasibility thresholds and evidence categories were frozen in advance.

Registered result

Invalid / Uninterpretable

This does not mean that the registered polity-scale prediction derived from PrF was supported or contradicted.

T-002 required a minimum amount of historical information before the confirmatory models were allowed to run:

• 600 eligible state-decade landmarks

• 60 distinct polities

• 80 distinct qualifying breakdown onsets

• breakdown events across at least 30 polities

The final dataset contained:

• 543 eligible landmarks

• 98 polities

• 53 distinct breakdown onsets

• 46 polities containing breakdown events

The polity requirements passed, but the landmark and event requirements did not.

Under the frozen preregistration, the confirmatory analysis therefore had to stop before model fitting.

The registered conclusion is:

T-002 did not contain enough qualifying historical information to determine whether the registered Institutional Direction Proxy improves prediction beyond the Material Mass Proxy.

The thresholds were not lowered after seeing the data, and the test was not redesigned to obtain a more favourable result.

Why feasibility failed

The historical pipeline began with 2,673 candidate state-decade landmarks.

Strict predictor-history, material-capability coverage, state-membership, entity-matching and continuity requirements reduced the available sample substantially.

A further 714 landmarks were excluded because complete 50-year non-event follow-up could not be observed.

The shortage of distinct breakdown events was more fundamental.

Only 53 qualifying breakdown onsets survived the registered construction, compared with the required 80.

Creating more overlapping observation windows would increase the number of rows, but it would not create more independent historical breakdown events.

This became one of the main methodological lessons from T-002: future tests need a stronger feasibility stage before confirmatory sample and event thresholds are frozen.

Exploratory follow-up

After the registered T-002 verdict had been permanently frozen, we ran a separate exploratory analysis using the existing data.

This exploratory analysis does not alter T-002’s registered evidence category.

The purpose was to understand what pattern existed in the data and to improve the design of future tests.

In grouped cross-validation, adding the Institutional Direction Proxy to the Material Mass Proxy:

• improved held-out log loss by 7.26%

• produced a high-MMP IDP probability contrast of −39.35 percentage points

In simple terms, the pooled cross-polity data showed a substantial association in the preregistered predicted direction.

Among states with high Material Mass Proxy values, higher Institutional Direction Proxy values were associated with substantially lower predicted breakdown probability.

However, the predictive improvement itself was highly uncertain.

The 95% bootstrap interval for the log-loss improvement ranged from −52.68% to +23.14%.

That means the exposed sample does not establish that adding the Institutional Direction Proxy reliably improves overall prediction.

The IDP probability contrast was more stable within the grouped cross-polity analysis, but that alone is not enough to establish the registered polity-scale prediction derived from PrF.

Forward historical validation

The most important problem appeared when the models were forced to predict genuinely forward through historical time.

For each historical forecast period, the model was trained only on information that would already have been available at that point.

Under this stricter validation, adding the Institutional Direction Proxy worsened pooled temporal log loss by 209.56%.

The high-MMP IDP contrast also reversed from the predicted negative direction to +4.74 percentage points overall.

This means that the strong pooled cross-polity relationship did not reliably transport through historical time.

The failure was especially concentrated in earlier historical periods, where the available training samples were small and historically narrow.

Later forecast periods performed more favourably, but the unfavourable earlier periods were retained rather than removed.

Variation across time and regions

The exploratory grouped-CV estimates became more favourable in later historical periods.

Before 1900, adding the Institutional Direction Proxy worsened predictive performance.

Between 1900 and 1949, the improvement was approximately flat.

From 1950 onward, the exploratory grouped-CV improvement was substantially more favourable.

Regional performance was also mixed.

In leave-one-region-out validation, predictive performance improved in three regions and worsened in three, even though the IDP probability contrast remained negative across all six.

This heterogeneity means that the observed relationship cannot yet be treated as a stable general law across historical periods and regions.

What we learned

T-002 did not produce confirmatory evidence for or against the registered polity-scale prediction derived from the PrF distinction between Mass and Direction.

But the exploratory follow-up revealed something worth investigating further.

The exposed historical panel contains a substantial MMP/IDP cross-polity association consistent with the predicted PrF Mass/Direction relationship, particularly in later modern states.

At the same time, the predictive gain was uncertain, historical forward transport was poor, and performance varied across periods and regions.

The narrow conclusion is therefore:

The historical data contain a substantial MMP/IDP cross-polity pattern consistent with the predicted PrF Mass/Direction relationship, but T-002 does not establish that the registered Institutional Direction Proxy reliably predicts state breakdown beyond the Material Mass Proxy.

That distinction matters.

The result does not justify claiming that PrF has been demonstrated at civilisation scale.

Nor does it adjudicate the underlying PrF distinction between Mass and Direction; MMP and IDP were only one empirical operationalisation of that distinction.

Instead, T-002 identified where the next test needs to improve: event coverage, historical follow-up, entity continuity, temporal stability, feasibility planning and independent validation.

What happens next

T-002 remains permanently recorded as Invalid / Uninterpretable.

Its exploratory findings are preserved as exploratory evidence for hypothesis development and successor-test design only.

Any future polity test must be treated as a genuinely new test rather than a rescue of T-002.

A successor design should establish feasibility before preregistration, use a defensible sample-size and event-information plan, improve treatment of censoring and historical continuity, and preserve strict forward temporal validation.

The central question remains open:

At the polity scale, can a defensible empirical proxy for Direction reveal prospective structural consequence beyond a proxy for Mass alone?

T-002 did not answer that question conclusively.

It did, however, give us a clearer way to test it next.

Transparency

T-002 was preregistered before outcome access.

OSF DOI: 10.17605/OSF.IO/GAFMY

The original preregistration, public pre-outcome clarification, execution audit, final confirmatory report, exploratory follow-up, analysis code and reproducibility materials are preserved.