True North

Builder script exists (launchpad_build_template.py, 732 lines). Turf messaging pack complete (turf.json — 5 workflows, custom fields, 6 pipeline stages). But NO template has been built yet. All snapshot_id = null in registry. Custom fields bootstrap DONE (Apr 24). Steps: (1) run --build --industry turf, (2) configure 5 workflows in AGM UI using copy from Golden Template Google Doc, (3) review pipeline + custom fields, (4) take snapshot in AGM UI, (5) run --register --industry turf --snapshot-id <ID>.

CORRECTED Apr 24, then WIRED Apr 24. Backend /score endpoint now returns contact_id to frontend. Quiz results page has 'Get Started with AGM Pro Tools' CTA that links to Launchpad intake with pre-filled name/email/phone/company/agm_contact_id. Full report (page 6) also has 'Get Started Now' CTA with same pre-fill. Remaining: deploy modal_automation_score.py to Modal (contact_id in response), push agm-automation-score to GitHub Pages (frontend CTAs).

Full Stripe test mode walkthrough: fill intake form, pay with test card, verify webhook fires, pipeline runs, close-the-loop fires, AGM sub-account created from golden template (not interim), OAuth URL appears, dashboard deployed, lifecycle email fires Day 1. Never tested. Target: under 10 min.

4-rep Costco training failed 2026-05-13 due to SQLite-on-Modal-Volume concurrent-write loss. Day-1 shipped: max_containers=1 throttle on signing_ui (live at signature.heavenlygreens.com); Neon project agm-contract-engine provisioned (PG17, Launch tier, us-west-2); canonical Postgres DDL in signing_pg.sql; one-shot migrate_sqlite_to_pg.py with --dry-run / --verify; 1,905 rows migrated with perfect parity (245 sessions + 1,483 events + 30 audit + 1 counter + 146 jobs); psycopg2-binary in image; Modal secret contract-engine-pg created. Day-2 (next session): shadow-write code in signing.py + bind contract-engine-pg to every init_db call site + parity-monitor cron. Day-3: flip reads to Postgres. Day-4: decommission SQLite + lift throttle + 15-rep canary. Full plan in docs/NEXT_SESSION.md top entry. Blocks the 6-10 rep ramp until Day-4 completes.

Lifecycle clock stamps timestamp on each pending state, emits state_stale event on stale threshold cross. Customer cadences (signing 3-day Costco / 7-day HG Direct, deposit 14-day HG Direct only). Per-rep 8 AM email with stalled-deals queue across all stages. Custom field updates: agmpro_contract_status_label, agmpro_contract_age_days. Phase 4A rep cadence already live — this is the customer side and the digest.

Delta on top of the just-shipped Site Videos tile grid (b17e90d). Add ~15 lines to format_normalize.normalize_video to extract a 1.0s frame thumbnail (ffmpeg -ss 1 -frames:v 1 -vf scale=480:-2) and upload to S3 as <basename>_thumb.jpg. Add ~25 lines to _layout_videos in installer_packet.py to fetch the thumb via boto3, write to tmpfile, page.insert_image as tile background under existing PLAY badge. Fall back silently to the current green tile if thumb missing. Backfill script for the 5 existing HGC-TEST-001 videos.

Build google_business_signals.py: given an operator domain, capture total review count + last-30d delta from Google Business Profile. Populate signals.review_velocity_30d + review_velocity_gap_pct (15pt of the FR-pest rubric — currently unsourced, capping cohort at 67/100). Wire into enrichment/__init__.py after geography. Flip HOT_SCORE_THRESHOLD back from 60→80 in pipeline.py. Re-run backfill — expect 4-8 of existing 8-warm cohort to move into ≥80 with review-velocity on record. Open arch question: Places API (~$0.70 one-off for 41 ops) vs. public scrape — lean Places API for reliability.

Recon shipped 2026-05-06 found the FieldRoutes (formerly PestRoutes, same software, ServiceTitan-owned) marketplace is empty real estate — 17 partners total across 9 categories, only 4 in Marketing Solutions, only Scorpion as a real platform-shaped competitor (and Scorpion sits 30-100x above AGM in price as a $10-25K/mo agency play). The $300-$3,000/mo price band is completely empty. Combined with the Valley Wide owner's no-flinch at $297 (would have closed at $597), the pricing thesis: AGM is structurally underpriced. Recommended new ladder: $97 Diagnostic / $397 P2 Headless (Troy's call) / $597-697 P3 Pro / $997 P3 Plus / $2,497 Enterprise (NEW tier). Existing subs grandfathered. Five-attribute defensible moat (FSM-agnostic, diagnostic-first, headless, self-serve, vertical-aware) — no competitor combines all five. Week-1 actions: (1) email FR partner team re: listing requirements + Scorpion positioning question, (2) draft marketplace listing copy using Jobber template Liam shared, (3) update Andrew's pitch deck with new ladder, (4) run 5-prospect elasticity test before locking. June-July: build quantified Sertbo/Valley Wide case study. Oct 7-9: IGNITE Orlando — pursue peer-led speaker slot (no public CFP, curated/invited), booth as fallback. Full doc + email shipped + project memory saved.

Top-to-bottom plan (drafted 2026-05-11) for productizing the FieldRoutes opportunity. Five phases: intake (self-serve form), headless provisioning (clone pest snapshot → new AGM Pro Tools sub-account → generate config + Modal apps), initial backfill, steady state (15-min Modal cron already running for Valleywide), and activation monitoring (per-client AGM Pro Tools dashboard + AGM-internal multi-client matrix with single activation score). BLOCKER #1: pest snapshot in AGM Pro Tools has never been built — launchpad_template_registry.json shows it null, falls back to turf template, which means headless onboarding for any new FR client can't run today. Build order: (1) pest snapshot — Andrew, (2) provision_fr_client.py generator, (3) FieldRoutes branch of Launchpad intake, (4) per-client dashboard as reusable template, (5) multi-client activation matrix, (6) activation score formula + Slack alerts on red transitions, (7) wire review_request_tag on completed-ticket detection, (8) marketplace listing. Includes embedded 2026-05-06 market research deep dive (Part 10 of the doc) covering: FieldRoutes ≡ PestRoutes lineage, 17-partner marketplace footprint, five-attribute defensible moat, IGNITE 2026 strategy, pricing thesis with outcome math, recommended ladder ($97/$397/$597-697/$997/$2,497). Seven decisions needed from Troy (pest snapshot owner, activation score weights, multi-office handling, per-frequency tags, backfill SLA, marketplace timing, pricing).

Strategic frame ratified 2026-05-05 for everything happening on the contract engine. Two distinct products fall out of one engine: (1) Costco Contract Engine as a Service — sold to Costco-approved vendors, 'already approved by Costco' as moat, small TAM / high margin / white-glove sales, gated by Costco approval relationship; (2) HG Deposit-to-Delivery Engine — sold to any FSM with deposit-then-execute workflow (turf, pool, fence, painters, hardscape, solar), broad TAM / mid margin / self-service signup. Same primitives, different products, different sales motions. Engine is already multi-tenant-aware in its bones (client_id first-class, track + ready_for_install_trigger anticipates deposit handling). Per-Modal-app split deferred until paid tenant #2. Andrew gets scoped Modal deploy so Troy is not on the critical path for client incidents (two-layer enforcement: pre_deploy_check.py Check 7 + scoped Modal token). Open non-engineering items: Costco approval conversation, pricing model, compliance posture for HG-direct deposits (Stripe Connect with tenant merchant accounts), tenant onboarding UI, per-tenant SLA + observability, Costco change-management process. Order of execution: finish Costco on current architecture → audit four large modules (delivery, pipeline, assembler, installer_packet) for tenant-cleanliness → ship HG-direct as a track → start Costco approval conversation in parallel → layer Andrew scoped deploy → first paying non-HG tenant triggers per-Modal-app split + tenant onboarding UI sprint.

PAUSED 2026-05-06 morning for Jobber developer onboarding call. STARTER_TRACK_STEPS scaffold already in execution/launchpad_pipeline.py (lines 158-177); 9-module shared library ready in execution/lib/. Resume after Jobber call: add $97 Stripe Price in parent agm-stripe acct_1TKjcNRN9q5afIIK alongside existing $297, build execution/stripe_price_registry.json, migrate modal_webhook.py:2772 from inline price_data to Price ID lookup with tier param, implement 5 STARTER_TRACK_STEPS bodies (client_provisioned Starter variant, dashboard_deployed multi-tenant tenant, welcome_sent Starter copy, starter_registered), wire CAS quiz CTA at automation_score/index.html:1921 to pass tier=starter, end-to-end smoke test.

Two conversion-frame pieces deferred from 5/3 audio-first bundle, bundled with Stream B (shared file: modal_webhook.py:2772). Pre-Stripe: 'What happens next (60 sec)' page between report-CTA-click and Stripe redirect. Post-OAuth: first 60 sec after Jobber connect must show real leads + dollar values immediately ('You have 17 leads >4hrs old, est loss $2,180').

Two distinct gaps surfaced 2026-04-30 when Marcia Serpa-Garcia 'Positive Review Survey' alert routed to Anthony Hall instead of auto-converting to a public Google review. (1) HG RETAIL: review workflow stack EXISTS (Ask For Review > Send Survey, Review Rating Results, Job Reviewed — all published) but is rep-dependent: satisfaction survey → manual rep follow-up. Needs rewiring to auto-fire Google review CTA; rep alert kept only for negative responses. (2) COSTCO: NO post-install review workflow exists at all — none of 24 review/survey/install workflows reference Costco. Greenfield build needed: trigger (SF install_complete or AGM fully_executed + N days), message (SMS+email with Google review URL), cadence (+3, +7, +14 then stop), branching (positive → public CTA, negative → ops escalation). Full research at docs/research/2026-04-30_review_automation_audit.md — includes workflow IDs, open questions, and Troy's confirmed framing that SF owns install triggers and Costco review side is dark.

Automated weekly build cycle report: git commits + chapter notes → contractor-friendly newsletter + companion YouTube video. 7 historical issues (Mar–Apr) drafted for May rollout. Issue # + publish date override ready. Campaign calendar at docs/BUILD_CYCLE_CAMPAIGN.md.

Managed monthly newsletter addon for AGM subscribers. $197/mo Pro Tools addon ($497 weekly). SPEC DELIVERED 2026-04-23 — Draft v0.1 Google Doc live: https://docs.google.com/document/d/1Q-4YZScmet1499k5tb1w0bzAHlOBU-LlS4qk8ve6P_k/edit (13 sections: scope, 10-beat architecture, intake/assembly/send pipeline detail, template system, packaging, pilot roadmap, risks, success criteria). ORIGINAL SPRINT (Apr 15-17) MISSED — HG Costco/change-order build owned the window. REPLANNED ship date in spec: Fri 2026-05-01 for AGM dogfood issue #1. SIX OPEN DECISIONS in §9 awaiting Troy: (1) ship date confirmation, (2) SMS number strategy (shared vs per-customer), (3) Quill voice calibration — add Andrew for pilot QA?, (4) pricing sanity check $197 vs $297, (5) Apple Shortcut v1 vs v1.1, (6) unsubscribe plumbing SES-managed vs custom AGM-writeback. GREEN-LIGHT PATH: Troy reviews → Harlan starts MJML template + SES identity for agmpro.com subdomain → build through pilot roadmap Phase 1. RED-FLAG PATH: inline edits in doc → Harlan revises v0.2. Andrew's role remains DNS-only for Chad rollout, NOT editorial. Chad's intake trigger is AGM workflow engine, NOT Jobber. Jobber stacks on later for HG/ATE.

Bring more reps online to run Costco in-home leads representing the Shaw product line. These are seasoned sales vets — they need training on the Costco system (proposals, contracts, workflow), not on selling. Training deck is the key deliverable that enables this.

Finalize the Costco system training deck. Covers how to use the system, write proposals, create and process contracts through the workflow. These are experienced sellers — the deck is about the system, not the sale. Backbone of multi-rep capability.

Logistics plan for Costco lead coverage: store locations, dates, times, personnel assignments, and materials required. Need a clear schedule so every store is covered with a trained rep and the right materials — no gaps, no scrambling.

Generate per-rep QR codes for HG sales reps working Costco in-store events. Each rep gets a unique QR that routes leads into the HG pipeline with rep attribution baked in — scan at the booth, lead enters the system pre-tagged to the rep who sourced them. Ties directly into multi-rep rollout and staffing plan. Build: QR generation script, rep-to-URL mapping, printable card/badge template.

Backend deployed, audio-first pivot shipped 5/3 (URL-first delivery, click-to-play green icon, Shock→Personalize→Quantify→Direct email frame, visibility-only positioning). Cross-browser verified Chrome/Safari/Firefox desktop + Mobile iOS. Stream B Stripe wiring paused for Jobber call. REMAINING: LinkedIn GTM post — drafted + approved, send when Stream B is live so the quiz CTA routes correctly. Email-client matrix (Gmail/Outlook/iOS Mail) verification still on the punch list before broad GTM push.

Built end-to-end Apr 6. V2 is AGM-centric per Troy's feedback ('pretty dang good'). Open question: Automation Score URL referenced but funnel not live — decide whether to cut paragraph, hold email, or swap content. Once approved, send to Texas Turf/Andrew first.

Format drift between record_chapter.py and create_chapter_book_doc.py caused dead links in Google Doc. Fixed for Chapter 30+. Chapters 1-29 need: sweep script to verify S3 URLs, backfill missing objects, update chapter entries, re-run doc generator.

Add J. Chris to the update loop and get the live gift card delivery workflow operational.

AGM pipeline opportunities show $0 because Zapier never synced monetary values. Backfill from SF contract amounts for all synced opportunities.

Code-side wiring complete. Service Type custom field + 5 appointment fields created in AGM Pro Tools sub-account, sync writes them all. 6 workflow shells published. Premium API key unlocked 2026-04-09 (quota blocker cleared, domain migrated pestroutes.com → fieldroutes.com everywhere, 15-min cadence locked). Tenant 1 of the AGM Pro Tools for FieldRoutes product — see full plan for productization context. Still pending go-live: SERTBO nurture workflow copy review + client cadence confirmation + flip active: true.

Audit every AGM workflow, calendar automation, and trigger that sends email or SMS for Heavenly Greens. Map: what sends, who it comes from, who receives it. Flag anything using {{user.name}} or rep identity when it should be call center. Partial progress: HG Lead Alert split sender from costco@ to HGLeads@ + Andrew cut over to agmprotools.com (commit 4d60510, May 5). Contract emails now route FROM rep with callcenter fallback (commit f5529e2). Three earlier-found instances fixed (appointment confirmation email/SMS, welcome email/SMS). Systematic sweep across all remaining workflows still pending.

TODAY. Developer onboarding call with Jobber team (Liam) at 9:00 AM PT. Engineering ALL GREEN (8 bugs fixed, live drill passed Apr 7). Partnership channel active. Demo prep status to verify before call: support@agmprotools.com inbox, 5 screenshots, Loom demo video, privacy policy URL, ToS URL, app icon 512x512.

TOP PRIORITY. AGM-driven workflow: on install stage, fire webhook to Modal worker, write BarCloud Stock Issue with turf type + dye lot, push draw record back to AGM + SF. Exec memo sent to Tim + Troy Apr 9 — significant scope. First step: BOM schema + dye lot data model in AGM. No canonical bill of materials exists at HG today — this build forces a canonical turf-type → materials map to exist for the first time. BarCloud API has no webhooks (pull-only). Eliminates manual BarCloud entry by office staff.

Curated dossier of 30+ speaking opportunities for Troy to position AGM Pro Tools. Covers industry conferences (Lawn & Landscape Tech Conference, Service World Expo, Equip Exposition, PestWorld), SaaS events (SaaStr, SaaS North), 20+ podcasts (Tommy Mello, Jobber, Owned and Operated, Landscape Disruptors), and local events (Phoenix/Austin expos). Three pitch angles by audience type. Several have OPEN speaker applications now — Lawn & Landscape Tech Conference (Jul, Scottsdale) is the slam dunk.

Automated delivery DARK since Mar 27 (18 days). Content pipeline Echo stage (Modal cron content_pipeline_scheduled, Tue 5 AM PT) fires but skips with "no trends file" — Monday Scout not persisting to /tokens/flywheel/monthly_trends.md. Last publish_request: 2026-03-27_214245. Morning triage: Modal logs + volume inspection. See docs/NEXT_SESSION.md top priority.