Tours need a phone

Getting the notes from the Zoom call itself: Virtual tour notes, from the Zoom call to the guest card (Oct 30, proposed).

A Western Slope Zillow lead, Jaleigh, asked for a virtual tour. She was told "virtual, here's the Zoom link." Everyone on our side was told in person. Not a new break: the booking has worked this way since July.

Why: the booking only reads "virtual" from the email that sets the time. The earlier ask is ignored.

Gaps we still have

1. No phone before booking. Zillow sends none and nothing asks yet. The ask-then-book chain is half in: first piece live, second armed, more open.
2. Relay email is not on her PropFlow contact. On purpose, since an August fix for split duplicates. The AppFolio card has it.
3. Reminders are text only. A lead with no phone gets none.
4. The virtual ask never reaches the AppFolio card as a note.
5. Wrong label. Her team email was tagged "resident inquiry" (a lookup fixed since July). Fixed and merged: prospects now tag Leasing.
6. Zoom waiting room unknown. The link opens with no sign-in and no registration; we can't see whether a waiting room is on.
7. Calendar body has no video link. Kat's calendar event body has no video link (only the title and location say virtual); the staff booking text has the link but not the word "virtual".

What is being done

Unrelated check: Kat's note on Elizabeth, a different prospect, did land on her AppFolio card.

Clara: "Friday, October 23 at 3:30 PM works. What's the best number to reach you? I'll book it once I have it."as built, passes the booking-promise guard · name and phone both missing: "Can I get your full name and the best number to reach you?" · Spanish: "El viernes, 23 de octubre a las 3:30 PM funciona. ¿A qué número le puedo contactar? Lo reservo en cuanto lo tenga."

If the slot is gone when the number arrives, Clara offers the next times. No number: no tour exists, nothing to expire, nothing blocked. A phone already on file: booked as today, no ask. Staff bookings: no ask.

Decisions (Fede, Oct 19, his words)

Extended the existing lead-outcome replay family (no new harness), synthetic records through the real reply and booking path. Red on main (12 of 19 assertions fail), green on the merged branches (19 of 19), and each of 8 planted defects turns the file red. Registered as mutation controls; atlas row added.

Defect the harness found, now fixed: a number typed in an email reply and saved through one prospect tool got no reminder-text consent (only the booking call's phone field granted it), so such a prospect would have been booked without the 24h and 1h texts. Fixed on the branch; that case is red before the fix and green after, and a number pulled from a parsed lead email still gets no consent. 19 of 19 assertions green on the merged branch. Known gap kept: a second, different number is not saved (D5 follow-up). Case 12 proves the calendar is re-read, not the race for the slot itself.

Replay result (local build, Oct 18 evening). A script rebuilt each person's contact records as they stood when the tour was created and ran the real gate decision from the local branch over all 33 tours.

What Clara would have said, from the real leasing prompt and the real booking tools fed the real prospect message (the replay harness cannot set its clock back to a past slot, so this is the reply path, not the full loop). All three asked for a phone; the control run fed the line production actually sent said "You're all set, tour confirmed" all three times.

Two gaps the replay builder found by reading the gate, both sent back to the gate builder to fix on the branch: (1) a held tour could still be confirmed with no phone if the prospect's next message asks to move the time, because the reschedule path never checks contact (main has the same hole for the name rule); (2) a prospect id passed by the model counts as a staff booking and skips the rule. Neither fired in the 33 real tours.

Local build, Oct 18 evening (branch fede/guest-card-contact-intent, 8 small commits). A new "update contact" write kind and its queueing function (one row per card per field; the row carries the field name, never the number). A set-contact method on the shared guest-card writer, its AppFolio version, and a robot client call to the fill-empty route. The drain reads the person's current phone or email at send time and ends in one of four states: filled, no change, already set, no value. The trigger fires only when a field was newly filled and the number came from the prospect or a staff edit, never from the AppFolio sync (that is the loop guard, with a planted-defect control that turns red when the sync's provenance is added to the push list). It rides on the existing per-building "guest card after a call" setting, no new switch.

Ownership: the write intent and the outbox are in the PMS layer this session owns (lane 8); the after-reply adapter and side-effects belong to lane 10 and are not touched. No AppFolio name outside the adapter files. The robot checkout on this machine is behind; the route exists on its main.

Both branches are rebased onto current main. Unit tests are listed as unit-covered, never as "tested"; the only validation that counts is the end-to-end run in production after merge, at Willows first.

Known follow-ups: a volunteered second number (D5) is not built, reminders go to one number per person; a time the team must confirm (no calendar connected) is still recorded as a request with no phone; the harness's fake robot does not know the new fill route.

Read-only inspection, Oct 18, of every note Clara queued for a guest card since the outbox began on Sep 18 (596 note-bearing writes across every customer; counts are rule-based over all of them, not a sample). Western Slope is the only live customer with notes: Situs and Camellia have none, and Camellia's 83 cards in the Oct 9 export carry no notes at all.

Gap count, the 20 Western Slope prospects who wrote back since Oct 1. "Happened" is read from the real threads; "got a note" is any note on the card for that milestone.

The deferrals that got no note: Angele T. ("I will reach out"), Ethan S. (busy "for a couple weeks"), Natalee M. ("let me get back to you on timing"), and Erika L. (a home "out of my first range"). One prospect, Tessa L., wrote 20 times and called once, and her card has nothing from Clara at all. Every email logged on a card is a first email; Clara's later emails and texts are never logged.

The best Clara notes on any card today are Kat's answers to the after-tour check-in, relayed by Clara ("Post-tour check-in from Kat: good fit, looking for end of December move-in"). That is the shape the design below copies. Full report, per-note sheet and 15 scrubbed samples are kept with the session; the Fairhaven test property shows 127 failed notes (no time zone on the property), a test-data defect, not a customer one.

• Kat Barker's email, Oct 16 5:06 PM and Fede's reply (Slack #western-slope, ts 1792188405 / 1792189615). Jay Taylor's Oct 9 email (ts 1791594212). Oct 14 weekly with Western Slope, raw transcript (#transcripts thread 1792015510).
• Leasing hub: leasing-engine-designs, sections 6 (prospect notes field audit), 10 (what we write to AppFolio), 22 (guest-card updates write-intent audit), 25 (guest-card notes, team style). Name rule: PR #19305. Robot fill-empty contact route: appfolio-browser-agent #490.
• Testing: testing playbook; harness atlas. Branches (local, no PR): fede/tour-phone-gate, fede/guest-card-contact-intent.
• Numbers: read-only DynamoDB queries over propflow-prod on Oct 18. Tours: by property partition since Sep 17; person claims by type and created time; mirror guest cards since Sep 17 by lead source; reminder marker rows. Notes: write-intent outbox by status (776 sent, 724 failed, 1 pending, 0 writing); Western Slope conversations since Oct 1 (264: 178 voice, 61 email, 15 video, 10 text) with the 86 non-voice threads read in full; staff notes from the Oct 9 AppFolio history pull (3,274 Western Slope cards) and the 2022 to 2025 robot card dump (1,707 cards). Per-tour sheet, per-note sheet and scripts kept with the session.