2026-08-09 (later): Second-round critique from Nicolas's parallel Claude session

Nicolas ran the synthesis past another Claude and brought back a sharper review. Verdict on the pick survives; several corrections integrated into program.md (Method guardrails): - Credit-channel argument STRENGTHENED: TESS's own CTOI evaluation documents that promoted community candidates are exactly our slice (FFI fainter than QLP Tmag 10.5 cutoff + single transits rejected by periodic searches). - Calibration: ~3,400 CTOIs, ~20% to TOI, 168 TOIs via CTOI. Plan 1 TOI per 5 well-vetted submissions. - Gaia dismissal overshot; narrowed to "spectroscopic-confirmation science out of reach". DR4: one timeboxed week in December max. - Three hard amendments: (1) self-detrend from simple-aperture flux (wotan biweight 0.5-1d), injection-recovery through the IDENTICAL detrending path; (2) blended EBs are the dominant contaminant, centroid vetting module mandatory (difference imaging, per-pixel LCs, Gaia DR3 neighbor sweep, archival ZTF/ATLAS/ASAS-SN); (3) hunt duotransits, sweep all sectors for a second event to get a schedulable alias comb of periods. - SOHO odds corrected down to 0.05-0.2/yr; kept for artifact-discrimination training. SOHO verified still operating (extended to >=2029). - Practical: TESSCut/astrocut server-side cutouts over bulk FFI pulls. claude setup-token confirmed available in v2.1.226 for auth.