[Parity]: [P3] Verify TDT header counts vs the retail table and under-construction credit

Imported from the engineering backlog (_bmad-output/implementation-artifacts/backlog.md, story tdt-header-counts-verify).
Priority P3 (do-first order) | Severity: low | Status: open | Epic: SimTower parity | Type: review-deferral | Recorded: 2026-07-08

Divergence from the 1994 original

Gated: needs the harness (real-game load). From the v1.11.3 header-count fix review (bmad-code-review, 3 layers; no confirmed bugs, Blind Hunter's over-count "High" refuted by the repo-access layer). Two canon questions the diff review couldn't settle without the real game: (1) commercialCount vs the empty §7 retail table, the fix writes a nonzero commercialCount while the exporter still writes the 512-slot retail table entirely empty; if the 1994 reader cross-checks the header count against that table (variants/status/finance lines), the mismatch could mislead it the way a zeroed recyclingCount did. Recycling was verified against the real game; commercial was not. (2) Under-construction crediting, buildTDT counts under-construction facilities toward the header aggregates (consistent with "count what's emitted", they emit with a negative type byte); confirm the real game credits a still-building recycling/security/etc. toward its advisories, or only operational ones. Both LOW; verify once the DOSBox-X save/load harness lands.

Wait for the DOSBox-X save/load harness, then inspect buildTDT and the exporter’s 512-slot retail table. Compare commercialCount with the empty retail table in a real-game load, and verify whether under-construction facilities contribute to advisories. Done means both behaviors are confirmed against the 1994 game.