RSIT residual stages, two homes for the measures, and the source of stretch (locked-core re-expression, 30 October 2026)

Working note, not a journal paper. Locked-core re-expression of the residual door and the interface sequester so that each measure has a concrete home. Does not revise the locks, does not replace prior deposits, and does not claim a detection. One construction, three locked doors, zero extra continuous free parameters. No fourth door unless one of the three fails in public. Residual class A remains the terminal first-coverage mark only (working centre ≈ 0.040 inside [0.03, 0.05]), of an unnamed first whole with no edge, not looked for after the start. Intermediate unfinished selection is distinct and does not re-enter residual A. The geometric pair is identified, not fitted. The interface recovers general relativity and the Standard Model when the residual gate is closed. Two homes keep the accounts separate: Book A retains the cutoff sum with null gravitational effect when the gate is closed; Book B holds the radiatively stable residual that sources the locked geometric path and is measured by the data. Public readings recorded: fixed-point BAO + PantheonPlus Δχ² ≈ −3.3 in favour of the geometric point; joint free-parameter MCMC on Planck + DESI DR2 BAO + PantheonPlus shows mean −log posterior difference ≈ +1.33 (equivalent Δχ² ≈ +2.7), with free w₀ settling ≈ −0.838 next to the lock. Competitive, not decisive. Companion to Zenodo records 23653695 and 23674560. Prior deposits are not replaced.