Lead Tracking Playbook

1Mission + where we are

The 30-day focus: Dan (Level King, DFW concrete leveling) makes an extra $100K ≈ 60 jobs ≈ 25 qualified leads transferred per week. Everything else is parked.

• WORKING Delivery: rebuilt + proven end-to-end Jul 28 (instant-send, ~5 s).
• WORKING Conversion signal to Meta for Dan's campaign: 27 live "Lead" events since Jul 21.
• WORKING Bookkeeping (since Jul 28 evening): every real conversation logs to the brain with its ad fingerprint, every hand-off is QC-gated and logged; one unified transfer log ties convo → QC → transfer → delivery (see the tracking SOW).
• HALF-CLOSED The last hop to Dan is still manual, but outcomes no longer live only in the thread — a read-only morning ingest reconciles the Dan thread into the brain daily (D4, live).

2Dan lead scoreboard — reconciled from the real thread

Read directly from the Akash↔Dan iMessage thread (148 messages, Jul 20–28). This corrected several stale "next actions" that the morning list still carried.

3The tracking machine — atlas vs live

Full audit lives in the Facebook Lead Tracking SOW. The one-screen version:

4Decisions made (co-CEO, Jul 28) — do not re-litigate

• D1 — Tracking rows live in the brain database with dashboards on top. No Google Sheets mirror unless Akash later says "sheets too".
• D2 — Layer 2 is rebuilt inside the proven n8n transfer rail (extract → QC gate → needs-review vs transferred → log + notify). The dead "*3 Auto Transfer Tracking" GHL workflow gets deleted once the rail carries its semantics — it has never fired; there is no history to lose.
• D3 — One log, two doors: the manual transfer form feeds the same rail and the same log as auto transfers.
• D4 — Daily Dan-thread ingest: YES. A read-only morning step reads the Akash↔Dan iMessage thread, updates lead outcomes in the brain, and flags anything owed. Never sends anything.
• D5 — Pipeboard retired from the SP relay: DONE Jul 28 (vault system-user token, proven with a test fire, change-ledgered, reversible).
• D6 — Every new market page gets a messaging dataset as part of the page-factory checklist — added as an Akash click-step (UI-only in Events Manager), then routed through the universal sender.
• D7 — Photo required at intake: the bot asks for a photo before transfer (Dan's hard requirement). Ships as a config change through the normal approval gate since it touches live bot behavior.

5Execution checklist — agent work, in order Claude

• 1. Verify the migrated SP CAPI relay's first scheduled run succeeded (n8n executions for rHCGWJnWv3oSGv3w); mark the ledger row verified.
  DONE First post-migration hourly run succeeded 18:00 UTC (execution 19553, 2s, clean) after erroring hourly on the dead token; ledger row 91914 marked verified.
• 2. Layer 1 readback: pull "*1. Qualified Convo Tracking" live config, diff field-by-field against the atlas attribution list, add missing fields via API, re-point the log row from the EyeFly sheet to the brain (D1). Prove with one synthetic first message.
  DONE All 9 atlas attribution fields already exist as live custom fields (nothing to add). New brain table sp_qualified_convo_tracking + n8n receiver live; synthetic first message proved the row lands with full fingerprint. GHL-side swap of the Sheets action = 2-min Akash click-path (section 6).
• 3. Build Layer 2 in the rail (D2): complete extraction to the atlas field list, add the "client-facing?" QC gate + placeholder/AI-leak checks, needs-review branch, transfer log row. Prove with one dirty + one clean synthetic lead.
  DONE QC gate ("client-facing?" + placeholder/AI-leak checks) now runs on every lead in the bridge; QC fail → needs-review + transfer refused; drain only sends QC-checked qualified leads (10-min backstop keeps speed-to-lead). Proven: dirty synthetic blocked, clean synthetic SMS-delivered 17:45 UTC. One log two doors: v_sp_transfer_tracking.
• 4. Build the D4 daily Dan-thread ingest (read-only, morning chain), backfilled from today's reconciliation.
  DONE Live: reads the Dan thread read-only each morning (06:45 chain step 2.95), classifies outcomes, updates the brain, flags owed items. Backfilled all 10 leads; first run already caught Sam Etoum → ENGAGED + Dan’s CRM-entry suggestion.
• 5. Ad-URL fingerprint sweep (read-only) across live ads; deliver the fix list as an approval pack.
  DONE Swept both ad accounts: zero live ads anywhere; all 11 staged relight ads are Messenger-type (ad ID passes natively, no URL to tag). Verdict: no fixes needed; LP-bound ads get the URL-tag rule at relight. Pack in _scratch/deliverables/2026-07-28.
• 6. Media mirror for Joseph, Scott, Spencer (durable lead-media pages); nudge list for Joseph's stale outcome into the next Dan packet.
  DONE All three pages live (Joseph’s newly mirrored, 1 photo): sp-lead-media.stpierre.workers.dev/{joseph-anwah, scott-edwards, spencer-johnson}. Joseph nudge saved as a P1 task due Jul 29 + flagged in the outcomes table.
• 7. Draft the two FB let-down replies (Cindie, Scott) for Akash to send — drafts only, no sends.
  DONE Both drafts ready in _scratch/deliverables/2026-07-28/fb-letdown-drafts.md — nothing sent.
• 8. Prepare the photo-ask intake change (D7) as an approval pack.
  DONE Approval pack staged (soft-ask default ★, hard-gate alternative): _scratch/deliverables/2026-07-28/photo-ask-intake-approval-pack.html. Deploy is one reversible API call after your go.
• 9. Fix the 5 open audit items from the morning close-out: Nathan G. duplicate, re-create Cindy S. + Richard W. deleted contacts (+ root cause), scrub debug text from packet notes, fix dead follow-up nudges.
  DONE Nathan G. duplicate stub quarantined · Cindy S. + Richard W. GHL contacts recreated (deletions were unledgered — now tagged do-not-delete) · Spencer + Sam junk ZIP/notes scrubbed · Anas’s stale contact link fixed. Follow-up nudges: root-caused — no follow-up workflow exists at all; the 6-step cadence build is on your GHL click list. BONUS: found + fixed the deleted delivery contact that had silently broken all lead SMS since 15:20.
• 10. End-to-end proof + screenshot pack (SOW Phase 5), then close the SOW.
  DONE One synthetic lead ran the whole chain: conversation counted with ad fingerprint → qualified → extracted + QC pass → transfer logged → SMS delivered 17:57 UTC, all visible in one v_sp_transfer_tracking row. SOW closed.

6Akash-only list ~15 min total

• Send the two FB let-downs (Cindie Stewart, Scott Edwards) — drafts will be ready per item 7 above. You told Dan you'd handle both.
• Phase 0 clicks: open "Qualified Tag → Lead Submitted Event" → click the Meta Conversion API step → disable if on. Confirm Jordan is the selected agent in the sleep-bot steps of *3 and *5.
• Mint messaging datasets for TX + ATX (then each new page): Events Manager → Messaging → select page → create. ~1 min each, UI-only.
• Text Dan (one message): the DripJobs/Zapier key re-ask is already in the thread from this morning — bundle the booking-link ask if he hasn't replied.
• Nudge Joseph Anwah's status next time you talk to Dan (only lead with no outcome). His job photo: durable media page.
• NEW — finish the Layer-1 re-point (~2 min): open "*1. Qualified Convo Tracking" → on the qualified branch click "#1 Update Qualified Convo_Raw Tracking" (the Google Sheets step) → add a Custom Webhook action right after it: method POST, URL https://eyefly.app.n8n.cloud/webhook/sp-layer1-convo-b38cb8fcdbb342c5, custom data: first_message={{message.body}} · utm_source={{contact.attributionSource.utmSource}} · utm_campaign={{contact.attributionSource.utmCampaign}} · utm_medium={{contact.attributionSource.utmMedium}} · utm_content={{contact.attributionSource.utmContent}} · fbclid={{contact.attributionSource.fbclid}} · ad_url={{contact.attributionSource.url}} · ad_id={{contact.lastAttributionSource.adId}} → Save + Publish. Keep the Sheets step until the first real row lands in the brain, then delete it.
• NEW — delete "*3 Auto Transfer Tracking" (the rail now carries its semantics; it has never fired, nothing to lose). Automation list → *3 → delete. API can't delete workflows.
• NEW — approve the photo-ask (D7): read the approval pack (item 8) and reply "approve photo ask" — deploy is one reversible call.

7Learnings + gotchas (today's, worth keeping)

• The transfer database only knows "queued → sent to Akash." Real outcomes (declined, proposal sent, meeting booked) live in the iMessage thread — which is why the morning action list carried stale items. D4 closes this permanently.
• The atlas itself shipped the double-sender. The qualified workflow fired both the n8n relay and GHL's built-in CAPI action. The built-in action is the one that silently broke in April (blank Messenger ID → cost per qualified conversation ~$58 → ~$428). One sender only, forever.
• Messaging datasets ≠ web pixels. Messenger leads never hit a web page; only page-linked messaging datasets see them, and they only accept LeadSubmitted/Purchase. Dataset creation is UI-only (API permission-blocked).
• The EyeFly fleet relay has been silent since Jun 22 (4,025 lifetime events). If any EyeFly client campaign still spends, Meta is optimizing blind for it — worth one check against active spend, not yet asserted.
• Wrangler deploys need --branch=main or the custom/prod URL never updates (hit again today).
• Atlas ≠ gospel: the recording captured a machine with its own bugs (Layer 2's missing trigger, EyeFly-branded sheets). Rebuild the semantics, not the artifacts.

8Links + IDs