Validate populace-be against EUROMOD-BE and Federal Planning Bureau scores

Part of #259 (populace-be epic).

What

Belgium has no incumbent dataset inside our stack, so the US/UK self-parity gate is replaced by external oracles. Two tracks:

1. Baseline comparison vs EUROMOD-BE. EUROMOD is open-access; where the SILC licence permits running it on the same input vintage, compare baseline outputs: poverty rates (AROP), income decile means, PIT revenue, social-contribution totals, main benefit aggregates. Where we cannot run it, compare against published EUROMOD-BE country-report statistics. Document every comparison's vintage and concept bridge.
2. Reform validation vs Federal Planning Bureau costings. The JCT/OBBBA analog: a curated list of Belgian reforms with official published scores (BFP analyses; SPF Finances estimates where available), split in-sample (targets the calibration saw) vs out-of-sample (never seen). Wire into the reform_validation.json release contract with the existing out_of_sample_simulated requirement.

Dependencies

• Reform runs need counterfactual materialization through the Axiom adapter — sequenced behind the adapter issue's reform-extension decision.
• Rules-level (single-household) cross-engine validation against EUROMOD/other calculators belongs to the Axiom oracles work (TheAxiomFoundation/axiom-oracles); this issue is the dataset-level analog — population aggregates and distributions, not per-case rule correctness.

Acceptance

• A release ships reform_validation.json against named BFP/EUROMOD-BE comparators for the encoded slice, with in/out-of-sample labeling and source URLs per row.
• Baseline comparison table (populace-be vs EUROMOD-BE vs admin) published in the release diagnostics.

Start by tracing the existing release contract and the reform_validation.json output, then review the Axiom adapter dependency before planning the reform runs. Define the named EUROMOD-BE, BFP, and administrative comparators, their vintage and concept bridges, and the in- and out-of-sample labels; done means the release diagnostics contain the baseline table and a sourced reform_validation.json for the encoded slice.