Proves two fixes shipped 2026-08-23 end-to-end in prod: PR #3920 (the identity-gate
detector now forks on a bare parent/HQ root paste with ≥2 discovered /city-chapters/
siblings — previously this silently absorbed the whole organization) and PR #3923
(the unnamed-arm fallback label reads "The part I run" instead of "This chapter").
Pasted https://www.theoffline-club.com (no /city-chapters/ path,
no first-person operator copy) into the Brain "Research my business" field on a wiped
Peyton WP tenant. The free peek completes and the identity gate opens with the
conservative unnamed-arm ask — this is the exact case that previously never forked at all.
Title: "We found more than one face on this site — which one is yours?" Fallback labels: The part I run / The whole business — not the retired "This chapter". Neither radio is pre-selected. "That's me — run ✦" is disabled (gray).
Radio for "The part I run" selected. Business Name field auto-fills The Offline Club — the real extracted name, not the literal UI fallback string (confirms the #2997 placeholder-persistence guard: the fallback label is never persisted as the confirmed business name). "That's me — run ✦" is now enabled (white). Subline updates to "Everything we map starts from your pick — nothing from the other one sneaks in."
Closed the gate without confirming. This sitting's one paid read was already spent on the Luma URL in a prior session (CT-BALI-04); CT-BALI-06 only needs to prove the fork is choosable and HQ is not auto-selected — which it does.
| Table | Check | Result |
|---|---|---|
brain_profile | business_name | Empty — no row. Confirms nothing was persisted since "That's me" was never tapped. |
brain_read_audit | newest row vs. this walk's timestamp | Newest row predates this walk (from a prior sitting) — no new paid synthesize-profile row fired for theoffline-club.com from this free peek. Matches pass criteria exactly. |
This walk targets CT-BALI-06 only — the one scenario gated on the two fixes above, and the one the queue's own tracking (dated 2026-08-23) marked as "NOT YET DONE." CT-BALI-00..04 already PASSed in an earlier sitting; CT-BALI-05b remains BLOCKED on missing fixture screenshots; CT-BALI-07..15 and Sitting B remain outstanding — those are unrelated to this pair of fixes and are left to whichever session resumes the full dress rehearsal. The tenant was found already empty (assistants: 0, brain_profile: 0) when this walk started, despite an earlier same-day queue note describing a completed CT-BALI-04 paid read — a likely sign of concurrent sessions sharing this one QA tenant. Flagging for whoever picks this back up, not diagnosing further here.