Add Belgium target and population-input boundary

[MaxGhenis]

Summary

• declare a value-free Belgium target surface: 15 typed count references plus a separate three-row monetary inventory
• add ten nullable, Microcosm-owned person-status inputs and one-to-one typed scheme-population mappings for GRAPA and regional child benefits
• keep all unresolved source, population, input, period, geography, unit, and completeness conditions fail-closed
• keep every regional child-benefit row validation-only; no participation or take-up row enters calibration
• prevent generic behavioral names or formulas such as takes_up_* from entering Axiom or the Microcosm input contract

Chronicle owns publisher facts and provenance. Microcosm constructs/calibrates the population and populates measured or latent nullable flags. PolicyEngine consumes those flags and owns later non-legal take-up and labor-response mechanics. Axiom may receive only concepts specifically grounded in public policy documents.

Unknown is null, never false. Activating a mapped target requires all typed readiness fields plus a bound frame receipt proving n_unknown == 0; changing a readiness string alone is insufficient. No activation path is added here.

Target boundary

The calibration-candidate count surface contains Statbel population, Statbel fiscal-income returns, and ONEM complete-unemployment recipients; each is still blocked on its named bridge or input. The separate monetary profile contains SPF Finances PIT receipts as a blocked calibration candidate, with ONSS worker contributions and NBB household disposable income retained for historical validation only.

GRAPA, pension, and all nine regional child-benefit rows are validation-only. The child rows nevertheless define the typed Microcosm inputs needed for future receipt/participation validation. They do not construct eligibility denominators, take-up rates, or behavior. No wealth target is declared because an official pinned HFCS package and exact interview-period/support bridge are still absent.

Chronicle #212/#213 reconciliation

Chronicle #212 declares the relevant record identities and administrative scopes. The newer source-faithfulness audit in Chronicle #213 establishes that the Opgroeien dashboard capture, Iriscare material, and Ostbelgien manual transcription are not activation-grade publisher artifacts.

Accordingly, five mappings carry publisher_source_readiness="native_publisher_artifact_required_missing": one Opgroeien, two Iriscare, and two Ostbelgien rows. GRAPA and the four Walloon child rows have pinned official source material, but remain independently blocked on Microcosm population/input/period/completeness gates. Publisher household rows remain excluded until their unit is proved equivalent to a Microcosm household.

Source follow-ups are tracked in Chronicle #215 (HFCS), #216 (Iriscare), #217 (Opgroeien), and #218 (Ostbelgien).

Reviewed coordinates

• target: main@d1e3e397bdc4b7e6b9e05dc73cf6345e7111e6db
• head: 654cb6c2cba558ee8aec32144d22f8d04495866c
• Chronicle Cross-check: Populace US 2024 net income & tax ~37–43% below Enhanced CPS (benefits match) #212: 0f75a2bb5fae8a197e4e1f6541a5af34708fc206
• Chronicle Publish a small calibrated national sample (populace_us_20k) for Colab/CI/tutorials #213: fe3fd81670f106c5509847312ba61dd4c4742626
• merged monetary primitives from Microcosm Add generic monetary target primitives #825: 510e3e6c9c2cd7f61854c56f6edd2acb14bc1cd8

Validation

• 730 focused country, gate, Ledger, monetary, profile, population-input, and bundle tests pass after the Cross-check: Populace US 2024 net income & tax ~37–43% below Enhanced CPS (benefits match) #212/Publish a small calibrated national sample (populace_us_20k) for Colab/CI/tutorials #213 reconciliation
• 1,688 runnable shared-spec tests pass, with 43 expected skips and two explicit US-parity deselections
• calibration shard: 212 passed
• repository Ruff passes
• lock check: unchanged, 123 packages
• CI test grouping: 319 tracked test files, verification=ok
• JSON parsing and git diff --check pass

The two deselected US parity-regeneration cases fail identically on main@d1e3e397: the local feed exposes a CBO source_projection while the committed US reference remains observed_only. This branch does not touch that data or assertion path.

Independent exact-head review

Served claude-fable-5 adjudicated the exact d1e3e397…654cb6c2 range and returned VERDICT: APPROVE with no actionable correctness, source-faithfulness, architecture, or coverage finding. Attested Subfleet run: 20260831-090042-microcosm-824-exact-head-fable-v4, session 1b93f442-ea78-4ae2-af0a-c6f76622853e (requested: claude-fable-5, served: claude-fable-5).

Because the Claude read-only wrapper currently rejects --safe-mode, the review ran with workspace permission only in a disposable detached copy. The reviewer added three local-only commits ending at 7bf22f8d; the complete delta from the PR head is exactly two adjudication Markdown files (ADJUDICATION_824_PROGRESS.md and ADJUDICATION_824_REPORT.md) and zero product files. Nothing from that copy was pushed. The adjudication report explicitly binds approval to PR head 654cb6c2, not the local journal commits.

The reviewer independently passed 508 focused tests and found the full no-engine microcosm-build shard clean except four pre-existing policyengine_us import failures in an unchanged US test file. GitHub's engine-equipped US matrix is green at this head.

This PR remains draft and must not be merged here. The publisher artifacts, Microcosm population/input/period/completeness work, dependency review, and merge-time journal historicization remain follow-up gates even after exact-head approval and CI.