Monday Jul 27 — Parallel Execution Guide

1 Status: agent lane is DONE — everything left is yours

What the agents finished this afternoon (all ledgered, all reversible):

• Reporting rewired to the password-proof system-user token — the 06:10 sync now covers BOTH accounts (it found 3 CL campaigns the old sync never saw), backfilled Jul-20→27 matching Graph to the cent (B2B $114.11, CL $190.70). Dashboard live at stpierre-ads-live-dashboard.pages.dev with an honest "everything is paused, $0/day" banner + a CL section ($12.49/convo). Ledgers 91707/91709/91717.
• Privacy cleanup: Sarah's personal iMessages/photos purged from the lead DB and taken off the public media pages (verified serving nothing). 4 contaminated rows quarantined reversibly; a 3-layer ingest guard is live in the bridge and replay-proven. Clean restated CL numbers: 10 qualified, $18.45/QL. Ledgers 91710/91712/91713.
• Transfer rail hardened: DB-level dedupe (one sent per lead), 24h same-phone guard, terminal errors park immediately, retry cap 12, six runaway rows reset, Mike Yeo unstuck. Replay of Jul-22 = exactly one Cindy delivery. Also caught + turned OFF an auto-send an earlier session had left on. Ledgers 91714–16.
• Bridge enrichment verified live end-to-end + AI fallback added for messy text (proven on Cindie: "seventy six two six two" → 76262). Her card is now complete; media mirrored. Ledger 91711.
• Copy gate CLEAR: all 420 live ads across R2/R4/CTRL scanned — zero retired "inspections in 7 days" lines, zero banned handles, all Learn More CTAs. Local copy pool cleaned (434 replacements, 21 files, git rollback).
• Nurture verified NOT sending (published flag is real, but zero outbound emails, all enrollment tables empty, auto_send FALSE everywhere).
• Scripts v2 staged on a preview URL; production untouched. 125 uncaptured sessions archived to the brain.

2 What changed since The Report (verified live)

1. The machine token already exists. The system-user credential works on BOTH ad accounts — verified, and now wired into the sync + dashboard. Decision 8 is closed: no Meta re-authorisation needed, ever again for this failure class.

2. Dan's winning campaign is off too. Zero ACTIVE campaigns on B2B (55 checked) and zero on CL (12 checked) — including "CL | Dallas | $50pd" (~$18/qualified vs the $150 ceiling). Who paused CL is still unanswered. The relight is a two-account decision.

3. The Jul-22 relight spent ~$114 total before it was pulled (R2 $58.93 + R4 $47.43 + CTRL $7.75). The learning phase never got a read.

4. The contamination root cause is inside GHL: an email-reply automation adds the qualified tag to internal contacts, which fires the whole lead pipeline. The n8n guard now blocks the damage, but the trigger itself is builder-only — it's click #4 on your list below.

3 Your lane — everything that's left (about 2.5 focused hours)

• 1Get the humans unstuck (10 min). Forward Scott + Cindie to Dan (Scott's callback is overdue; Cindie is buying in early August). Reply to Nathan Green in the Facebook inbox. Forward the two recovered packet drafts too: _scratch/deliverables/2026-07-27/packets/twyler-webster.txt and mike-yeo.txt — both leads are Messenger-only, so Dan reaches them there.
• 2Start A2P registration (20 min). Multi-day carrier queue — every hour not started is added to the end. Keep the backup path ON until a real text lands on your handset. Save +1 626-325-0175 in contacts; check Unknown Senders.
• 3Text Sarah (5 min). Confirm she's done with the compromised-account cleanup, and ask whether she also paused CL Dallas. Her reply is the last relight input.
• 4Settle the SMS conflict BEFORE you relight (5 min — new, found Jul-27 PM). Production has the scripted-SMS and auto-follow-up rails armed (SP_SCRIPTED_SMS_DISABLED=false, verified live on the sp-dollar-start worker). They're silent today only because there's no traffic — the first applicant after relight gets an automated text. Two of your own rules disagree: your Jul-22 architecture directive says every automation ends as a draft in the inbox and you click send; AGENTS.md line 88 (Jun-30, binding) says scripted funnel SMS is automated speed-to-lead and forbids adding approval gates to it. Jul-22 is newer and more specific, so it should win — but the doc still says otherwise and only you can edit governance files. Pick one, then I deploy the matching state and amend the doc that lost. My read: keep speed-to-lead automated for the *first* scripted reply (a $3–15K ticket punishes a slow response), keep everything off-script as drafts. Note the registry-level disarm does NOT cover this path — the code reads only the env var.
• 5Relight (5 min). Copy gate is cleared, reporting is live and waiting. Flip R2 + R4 + CTRL on B2B, and CL Dallas unless Sarah says it was paused for cause. Open Ads Manager and see it with your own eyes. The dashboard and 06:10 sync will show spend the same day.
• 6Four GHL builder clicks (25 min). Agents are locked out of the builder: fix the crossed field-save · update the extractor prompts · add the text-on-review-pile step · kill/scope the email-reply → qualified trigger on internal contacts (the contamination root cause).
• 7Two one-tap approvals (5 min). (a) Scripts v2 preview — approve at ten-scripts-v2-preview.stpierre-brand-cold-start-docs.pages.dev/docs/ten-scripts and it gets promoted to the page you shoot from. (b) Packet template wording — approve the v3 format (Name/Phone/City/ZIP/Service/Media/Note, missing = "unknown") and auto-send gets flipped on.
• 8One voice note — remaining decisions (5 min). Silence accepts the stars: relight per #4 (7★) · review pile = sweep + text step (10★) · book Dan's Zapier call this week (11★) · Collins HOLD (12★) · Meta optimises on qualified-only, set before relight (13★). Plus three new calls: hard-delete Sarah's R2 originals (kept for restore — your call)? · approve one SQL pass to clean 326 retired-line rows in the angle bank (blocks future image-card waves, not the relight)? · expansion fork — (a) sell a contractor first or (b) build the St. Pierre lead-gen brand?
• 9Shoot V01 + V07 (60–90 min) once the preview is approved and promoted. Read the vein-cut aloud first; gut-check V06's "three weeks of money left" line on a beach backdrop.
• 10Close the day (15 min). Daily review: re-score the four stale week-28 commitments. Run shutdown.

4 Standing holds — unchanged

• Collins: pack queued, nothing sends (scope + pay unsettled).
• EyeFly: change nothing, ever. No expansion ad sets until the WP-6 fork is called.
• No Meta writes without preflight; no activation without your explicit go.
• Wide-matrix reads wait 72h / $150–200 after relight — no early kills.
• Drift backlog (406) and task-board cleanup (1,251) stay cut this week.
• Known debts, parked deliberately: talking-head direct-offer videos have the retired line baked into the renders (re-render needed) · 12-email nurture is published against the signoff rule (not sending; unpublish when convenient) · watchdog still shows 5 stale rails outside this scope (morning-read, finance chain, scorecard huddle).

5 Done-when — tonight's scoreboard

• DONE — reporting live on both accounts; sync heartbeat green; dashboard honest.
• DONE — contaminated rows quarantined, Sarah's content purged + off the public pages, guard replay-proven.
• DONE — transfer rail: no attempt count above 12, dedupe enforced, auto-send gated.
• DONE — copy gate clear on all 420 live ads; local pool cleaned.
• DONE — 125 sessions archived; scripts v2 on preview.
• YOURS — Dan has Scott, Cindie, Twyler, Mike; Nathan has a reply.
• YOURS — A2P submitted with a reference number; backup path still on.
• YOURS — both accounts relit (or an explicit decision to stay dark); spend recording today.
• YOURS — four GHL clicks done; a messy test lead reaches your handset untouched.
• YOURS — V01 + V07 on disk from the v2 scripts; shutdown logged.