[MaxGhenis]

Draft. Tightens five G tests that review 213 v4 (#1183 comment 6615417145) found checking less than their names claim. Tests only: no runtime, golden or resource file changes. It targets main since G (#1478) landed as f8f543a4. After the retarget, GitHub kept the PR's merge ref on G's old head f43e8fbb, so the hub merged main into the branch (121b5c02). That merge is clean over the single base f43e8fbb, and against main it changes only this PR's two files.

What changes

1. test_each_coarse_row_selects_its_native_rows_committed_labels (test_nz_target_references.py, review nit 1). Each finite coarse band's first native label now starts $1 above the band's lower cut, and its last label ends at the band's upper_cent (reference_intervals.json, table ird_taxable_income_standin_ty2024). The three native rows whose label repeats the previous row's upper endpoint (row_60, row_90, row_230: $50,000, $80,000 and $300,000) start at the cut instead, and the test pins exactly those three, in the style of _WAGE_SHARED_ENDPOINT_ROWS. The open tail's one label reads "Over $1,000,000", and the cut below it is $1,000,000. Before, moving row_59 from b01 to b02 in the labels, in metadata.native_rows and in the two bands' expected_member_count passed every feed-free test.
2. test_measurement_prefit_fails_the_toy_only_on_its_unwired_evidence (test_nz_transport_toy_graph.py, nit 2). The toy entry's roster_total_bounds, band_bounds and maximum_discrepant_bands must equal N/gates.json measurement_prefit's, its parameter keys must equal the packaged ones, and its two support floors must equal TOY_PREFIT_SUPPORT. The reviewer's mutant R2 (bounds [0, 100], 99 discrepant bands) passed before.
3. test_property_a_positive_standin_above_the_first_cut_lands_in_one_band (nit 3). An explicit @example puts wages at 0 and on each aged cut (16,500, 44,000 and 66,000), each one either side, with every other component 0. In 5,000 random examples the reviewer found no record on or within one of an edge, and closure mutant R1 passed every toy test.
4. New test_the_toy_standin_measure_keeps_the_packaged_central_rules (nit 4). The toy declaration's entity, add-backs, floor and scale equal N/income_measurements.json measures.central's, it names the same three outputs, and both give every component sign +1. Mutant R3 (no rental add-back) passed every toy test before.
5. test_target_profile_coverage_passes_on_the_toy_with_income_required (nit 5). The gate's matches_by_requirement["income"] must name exactly the six toy stand-in rows, so this test now pins the gate's own match list. The binding hands the gate the calibration problem's row names, which read name@period (gate_bindings.py:803-823, target_kernels.py:552-554), so the test drops the period before it compares. test_each_toy_coarse_row_enters_the_problem_matrix_once_with_support already catches a row that leaves the income family; this test adds the gate's view.

Invariants the tests now hold

• Each coarse income band covers exactly the native rows between its own cuts, read from committed files alone.
• The toy pre-fit gate carries R2C-3's frozen ratio numbers; only its support floors differ, as TOY_PREFIT_SUPPORT declares.
• The toy measure follows the packaged measure's rules: same entity, add-backs, floor, scale and output roles.

Checks so far

• ruff check and ruff format --check pass on both files.
• The hub replicated item 1's assertions on the committed JSON in plain Python: they pass at f43e8fbb, and the row_59 move turns them red (b01's last label ends at $49,000).
• Review 221 (Activate the NZ graph against pinned Chronicle facts (NZ graph G8) #1183 comment 6616218228, at 92ff931b) ran both files and five mutants. test_nz_target_references.py gave 20 passed and 41 skipped (each skip: NZ_CHRONICLE_ARTIFACT unset). Fixes 1–4 each turned red on their mutant: row_59 moved (M1), the R2C-3 bounds widened (M2), right-closed cuts made inclusive (M3, on the new explicit example) and the rental add-back dropped (M4). It owed two fixes. Item 5's assertion failed at the head, since the gate's names carry @2002 and the test's didn't: 8edd52a3 drops the period before comparing. Item 5's text overstated the gap the test closes: this body rewords it. The reviewer ran the same comparison without the outer sorted: it passed at 92ff931b and turned red when one stand-in row left the income family (M5). The outer sorted changes neither result, since the gate already sorts its matches.
• Delta check of 92ff931b..8edd52a3 (an in-session Opus agent; read-only, plus a plain-Python replay of both comparisons on the six names): APPROVE_WITH_NITS. The fix holds and both cites support the comment; the hub applied its body nits here. Brief 222 (Activate the NZ graph against pinned Chronicle facts (NZ graph G8) #1183 comment 6616766311) ran it at 8edd52a3: APPROVE. test_nz_transport_toy_graph.py gave 32 passed, 0 skipped; under M5 (toy_income_standin_amount__b2 moved to family toy), the coverage test passed its gate-status assertion and went red on the new one.
• Not run locally (the hub's machine has no slot for these files tonight). CI run 41745653899 runs both files against main.

🤖 Generated with Claude Code

[MaxGhenis]

Merged by the NZ hub (session local_e407a527) at 3ab4860 with --merge. Gates: gh pr checks exit 0; MERGEABLE; not draft; no CHANGES_REQUESTED review. Reviewed head: 8edd52a; the merged head differs only by a base update, and the PR's own changed lines are identical to the reviewed head's. Review 222 (#1183 comment 6616766311): APPROVE at 8edd52a. CI run 41851099404 tested 3ab4860 merged into main 875506e. Main since then changes 23 files from #1473 and #1506, none of them NZ's or #1501's, plus H2 (#1485, merged 18:21:54Z as 358b1c1). H2 and #1501 share only test_nz_transport_toy_graph.py, where #1501 edits three existing tests and adds one in the middle and H2 appends at the end; the merge is clean, with no duplicate names or imports (the hub's 11/03 check).