farm_income is an unpinned export dimension (109 selected carriers) — parity gate fails Build M attempt 12

[MaxGhenis]

Found by the export-mass parity gate on Build M sparse attempt 12 (release id populace-us-2024-buildm-sparse-rmloss100-6584dfa-20260716T105513Z, the first run under the #437 batched terminal report): farm_income exported $23.45B against the live-default reference's $62.39B (−62.4%, beyond the ±50% band). It was the run's ONLY failing gate group.

The column is an unpinned free dimension, and the wander is the unpinned-solve signature (#432's "amplifies unpredictably"):

artifact	weighted farm_income mass
live-default reference c2065b64 (buildg eCPS)	$62.39B
base-m pool at base weights — 353 nonzero persons of 865k, ONE-signed (no negative leg), byte-identical pre/post-#435	$10.24B
in the frozen rmloss100+keogh-swap selection (109 of 353 carriers, base weights)	$4.10B
Build J sparse export (75d5add)	−25.6% in-band (~$46.4B)
Build M attempt 11 export (27c8e64, pre-#435 base)	−43.2% in-band (~$35.4B)
Build M attempt 12 export (6584dfa)	$23.45B, −62.4%, FAIL

Nothing pins the column anywhere in the stack:

• the v8 Ledger consumer feed contains zero farm rows (grep source_measure_id.*farm over consumer_facts_buildh_v8.jsonl → no matches), and
• the compiled 2024 registry contains zero specs with farm in name/measure,

so the reference's $62.39B is an incidental artifact of the incumbent solve (the estate_income/Build H class), and the export is 109 selected records' solved weights — pure side-effect mass. Attempt 12's solve would have had to stretch the selected carriers 7.6× (from $4.10B) to reach the band floor ($31.19B) toward an unpinned incidental value.

Note farm_income (353 carriers, one-signed, PUF E02100-era concept) is a different column from farm_operations_income (the #298/#435 signed Schedule F leaf with ±legs) — #435 did not change farm_income at all.

Remedy, in order:

1. A reviewed parity exclusion ships with Build M (register entry cites this issue) — the rental_income/estate_income doctrine: a correctly calibrated column cannot be held to an incidental reference band.
2. Durable fix (the base-m collapses partnership_self_employment_net_earnings to signed near-cancellation (+47.7B/−48.1B legs) #432 remedy-3 pattern): identify the farm concept with a real target — SOI Publication 1304 Table 1.4 farm net income (Schedule F, income and loss legs) into the Ledger feed — so the solve stops treating farm as a free dimension. The exclusion lifts with that identification.
3. When the Ledger farm facts land, also consider protecting thin-carrier columns (353 pool / 109 selected) in the sparse selection the way the keogh swap (Keogh carriers lack selection support in the frozen rmloss100 57k: protect-swap for Build M + restoration-time support checks #434) protected keogh_distributions.