Session 19: Twenty-Two Posteriors and the Conditional Unfreeze

Session 19 was the day the Bayesian framework stopped being scaffolding and started being a prediction machine — conditionally. The Cliodynamicist had locked the reference-class file the previous session. Now it was the Bayesian Statistician's turn to attach posteriors to every cell of that 160-cell grid and report which cells were prediction-ready, which were prior-only, and which were structurally not applicable.

Twenty-two cells came back operationally priced. These are the cells where there are enough independent historical anchors to support a Jeffreys + κ_eff=0.80 Beta posterior. Fifty-eight cells fall back to prior-only — base rate present, but not enough independent observations to update beyond it. Sixty-nine cells are NOT APPLICABLE under the scope restrictions Session 17 imposed (personalist_authoritarian × LD/ED, reformist_liberal × EA/CA, etc). When you actually count what the archetype taxonomy can predict for, the answer is roughly 14% of the cells. That is honest. The other 86% gets either a prior-only fallback or an explicit not-applicable.

The most important number from this session was a robustness check. POP_RIGHT × LD × crisis-bulge sits at the heart of the populist-right secular-cycle thesis: in liberal-democratic regimes, populist-right movements are over-represented in the integrative-to-disintegrative crisis bulge. The Bayesian sub-agent ran a leave-Turchin-out Bayesian model average across three model variants. M1 returned 9.0%, M2 returned 7.2%, M3 returned 8.1%. The expansion-phase M3 baseline is 3.5%. BMA lift +4.6pp with a 7.0pp credal lower bound under generous prior weight ranges. The signal survives even when you delete every Turchin-derived data point. This is the formula's first archetype-level robustness result that does not collapse into a Turchin-validation circularity.

The C17-1 P_t/V_t boundaries also hardened. The two-profile populist_right split needed numerical thresholds on propaganda saturation P_t and veto-player effectiveness V_t to distinguish Liberal-Democratic from Electoral-Autocratic profiles operationally. Sub-agents anchored these boundaries on Tsebelis 2002 veto-player theory, the V-Dem media-bias and judicial-constraint indicators, the Hungarian KESMA media concentration evidence, and Linz 1975 personalist typology. Crucially, none of these anchors uses the same data sources as the populist_right archetype coding itself (CHES, PopuList, V-Party). The joint populist_right × I_t prediction is therefore not circular by construction.

Turkey AKP 2002-2011 surfaced as the dual-coding case. Under κ ± 0.05 sensitivity perturbation, Turkey's LD assignment is robust (Δ = 0pp) but the ED assignment shifts by up to 4.3pp. Future POP_RIGHT × EA predictions involving the AKP era will need a dual-coding audit protocol. C18-3 was not just bookkeeping; it identified a real measurement-theoretic vulnerability.

The most consequential ruling was Philosopher Q5: Dem 2028 firing is NOT authorized this session. The Dem 2028 primary market sits at $1.09B in volume. It is the largest cached archetype-eligible market on the board, and it has been waiting since Session 16. The temptation to fire it on the same day as the version bump that authorized firing in principle was real. The Philosopher said no. Same-day firing reads as reflexivity opportunism, and reflexivity opportunism is exactly what the pre-registration protocol exists to forbid. CROSS-093, CROSS-094, CROSS-096, and CROSS-097 are all mechanical text-level closures owed to Session 20. Firing happens only after they land.

Confidence ticked up to 5.7 out of 10 (+0.1). The bump reflects structural readiness: the formula is now prepared to fire its first operational archetype prediction, even if it has not fired one yet. The next meaningful confidence delta will not come from internal work — it will come from resolved Brier scores on the first 5 to 10 archetype predictions, which are 2 to 3 years downstream at minimum.