Overview
3 Weeks to Live
Week 1: Signal Engine validated and live — real distress data flowing, companies scored, pipeline operational. Weeks 2–3: Engagement Engine built and validated — from DPS trigger to outreach to mandate pipeline running. Both engines working together by end of Week 3.
2
Weeks — Engagement Engine
Week 1
Signal Engine
Data flowing · Companies scored · Pipeline live · Validated
Week 2
Engagement Engine — Core
Warm path · Intel package · Outreach sequencing · Drafting
Week 3
Engagement Engine — Full
Meeting prep · Proposals · CRM · End-to-end validated
Week 4+
First Live Outreach
Real companies · Real bankers · Real mandates
What's Already Built
| Component | Status | Notes |
|---|
| PostgreSQL + TimescaleDB schema | Done | 7 tables, live on localhost:3400 |
| Python DPS scoring engine | Done | Temporal decay + cross-signal multipliers |
| Utah SOS ingest (UCC filings) | Done | Live, running |
| openFDA ingest (recalls) | Done | Live, running |
| CourtListener ingest (federal court) | Done | Live, running |
| WARN Act ingest (layoffs) | Done | Live, running |
| News RSS ingest | Done | Live, running |
| 28 CPG companies seeded | Done | SLC metro, $50M–$250M debt range |
| REST API + dashboard | Done | localhost:3400, all endpoints live |
| Engagement Engine | To Build | Weeks 2–3 |
Build Model
All code is written by Claude Code (Max plan — unlimited, no API cost). Phineas specs, coordinates, and validates. Darren reviews and approves at each milestone gate. No exceptions.
Week 1
Signal Engine — Build & Validation
Goal: Signal Engine fully validated. Real distress data flowing from all sources. DPS scores accurate. Pipeline ranked and actionable. Gaps identified and remediated.
Day 1–2 · Data Ingestion Audit
1.1
Run full ingest across all 5 sources
Execute all ingest modules against the 28 seeded CPG companies. Review logs: records fetched, signals created, errors. Identify any sources returning zero results or errors.
Claude Code
Day 1
1.2
Utah SOS re-export — expand NAICS coverage
Re-run Utah SOS ingest with expanded NAICS codes: 311, 312, 3256, 32541, 4242, 4461. Add any new companies discovered to the database. Validate no duplicate imports.
Claude Code
Day 1–2
1.3
Fix any broken ingest modules
Remediate any sources returning errors — rate limit issues, API changes, selector changes for scrapers. All 5 sources must produce signals before proceeding.
Claude Code
Day 1–2
Day 2–3 · Scoring Validation
1.4
Score all companies — review DPS distribution
Run scoring engine across all companies. Review distribution: what percentage fall in each band? Are scores plausible given the signals? Flag any companies with unexpectedly high or low scores for manual review.
Phineas
Day 2
1.5
Calibrate scoring weights
Review 5 companies manually — check their signals against their DPS score. If scores are too compressed (everyone 20–40) or too extreme, adjust base point values or multiplier thresholds. Document any changes.
Both
Day 2–3
1.6
Temporal decay validation
Verify decay is applying correctly — signals older than 365 days should contribute 0 points. Signals from the last 30 days should be at full weight. Test with a manually constructed edge case.
Claude Code
Day 3
Day 3–4 · Dashboard & API
1.7
Dashboard review — fix any display issues
Full walkthrough of dashboard at localhost:3400. Check company rankings, signal timelines, score history charts, company detail pages. Fix any broken views, missing data, or UI issues.
Claude Code
Day 3
1.8
Add DPS threshold alert system
When any company crosses a score band threshold (e.g. moves from 40 → 41, entering Elevated), log the event to a new threshold_alerts table and send a notification to Phineas via webhook. This is the trigger for Engagement Engine in Week 2+.
Claude Code
Day 3–4
1.9
Scheduled ingest cron
Set up daily cron job to run all ingest sources and re-score all companies automatically. Log results. Signal Engine should be self-updating without manual intervention.
Claude Code
Day 4
Day 5 · Validation Gate
1.10
Pipeline review with Darren — Week 1 sign-off
Walk through the live ranked pipeline. Review top 10 companies by DPS. Are the signals plausible? Are the scores directionally correct? Are any companies missing that should be there? Sign-off required to proceed to Week 2.
Both
Day 5
1.11
Contact database — seed executive data
For the top 10–15 companies by DPS, populate the contacts table with CEO/CFO names, LinkedIn profiles, and email addresses (manual research or LinkedIn). This feeds the Engagement Engine's warm path scan.
Phineas
Day 4–5
Weeks 2–3
Engagement Engine — Build & Validation
Goal: Full engagement lifecycle operational — from DPS alert through warm path analysis, outreach drafting, meeting prep, and proposal generation. End-to-end validated against a real company from the pipeline.
Day 6–7 · Trigger & Intelligence
2.1
DPS alert → Engagement Engine trigger
When Signal Engine fires a threshold alert (DPS 41+), Engagement Engine automatically creates an engagement record for the company: status = APPROACH_PLANNING, timestamp, current DPS, triggering signals logged.
Claude Code
Day 6
2.2
Company intelligence package generator
Agent assembles a structured intelligence brief for each triggered company: signal timeline (from DB), capital structure estimate (from UCC data), known contacts, news summary, preliminary path options. Output as formatted document. Banker-facing only — never client-facing.
Phineas
Day 6–7
2.3
Warm path network scan
Cross-reference target company contacts (executives, board, investors) against banker's known network (LinkedIn connections, deal history, professional associations). Output: ranked list of warm paths with recommended connector and suggested framing. Flag if no warm path exists.
Phineas
Day 7
Day 7–9 · Outreach Sequencing
2.4
Outreach sequence engine
Build the cadence manager: 5-touch, 14-day sequence (Email 1 → LinkedIn → Email 2 → Phone prep → Email 3). Each touch is queued with a send date. Sequence pauses automatically when prospect responds. Status tracked per company.
Claude Code
Day 7–8
2.5
Outreach message drafter
Agent drafts all 5 outreach messages for each company. Messages are sector-specific, reference industry trends only (never company-specific signals), and are calibrated to the warm vs. cold path. Banker reviews and personalises before sending. Never auto-sends.
Phineas
Day 8–9
2.6
Outreach review dashboard
Simple UI showing: all active outreach sequences, current touch status per company, next scheduled touch, drafted messages awaiting banker approval, and response tracking. Banker can approve/edit/send from dashboard.
Claude Code
Day 9–10
Day 10 · Week 2 Gate
2.7
End-to-end test — 1 real company
Pick the top company from the DPS pipeline. Run full Week 2 flow: trigger fires → intel package generated → warm path scanned → 5-message sequence drafted → reviewed in dashboard. Darren reviews all outputs. Sign-off to proceed to Week 3.
Both
Day 10
Day 11–12 · Meeting Preparation
3.1
Meeting briefing package generator
When a prospect agrees to meet, agent generates: updated intel package, one-page meeting brief (key facts, emotional state assessment, recommended opening, topics to avoid), sector benchmarking data, conversation map (3–4 themes), and a branded leave-behind sector piece.
Phineas
Day 11
3.2
Phone call briefing sheet
For the Day 7 phone touch, agent generates a one-page call brief: 30-second opening script, 3 key talking points, likely objections and responses, target outcome (book a meeting). Banker-facing only.
Phineas
Day 11
Day 12–13 · Proposals & Engagement Letter
3.3
Preliminary assessment generator
Agent drafts a complimentary preliminary assessment (5–7 pages): financial overview from available data, capital structure map (UCC-sourced), regulatory risk summary, peer benchmarking (anonymous), preliminary path scenarios. Banker reviews and adjusts before presenting.
Phineas
Day 12
3.4
Engagement letter template library
Build template library for engagement letters: retainer-based, success-fee, hybrid. Agent populates deal-specific fields (company name, fee terms, scope, timeline, banker names). Banker adjusts fee terms and signs. Legal review before any live use.
Both
Day 12–13
Day 13–14 · CRM & Pipeline Management
3.5
Engagement CRM
Build lightweight CRM on top of the existing outreach_log table: company status (MONITORING → APPROACH → OUTREACH → MEETING → PROPOSAL → ENGAGED → CLOSED), last touch, next action, banker owner, notes. Integrated into dashboard.
Claude Code
Day 13
3.6
Pipeline recycling — non-converts
Companies that don't respond re-enter Signal Engine monitoring with enhanced tracking. Companies that responded but didn't engage get a quarterly check-in queue. Companies at DPS 81+ that haven't responded get escalated to banker for direct action.
Claude Code
Day 13
Day 15 · Full Validation Gate
3.7
End-to-end validation — full lifecycle
Run the complete lifecycle for 3 companies from the pipeline: DPS trigger → intel package → warm path → outreach sequence drafted → meeting brief → preliminary assessment → engagement letter. All outputs reviewed by Darren. Sign-off = ready for live outreach.
Both
Day 15
3.8
First live outreach — Week 4 kickoff
Select 3–5 companies from the top of the DPS pipeline. Banker reviews and approves the warm path analysis and first outreach message. Messages sent (by banker, not by agent). Engagement Engine monitors responses and queues next touches.
Both
Week 4, Day 1
Validation & Next Steps
Validation Criteria & What Comes Next
Signal Engine — Validation Checklist
| # | Validation Test | Pass Criteria |
|---|
| SE-1 | All 5 ingest sources run without errors | Zero errors in ingest logs. Each source produces ≥1 signal across the company set. |
| SE-2 | DPS scores are distributed across bands | At least some companies in Elevated (41+) band. No artificial clustering at 0 or 100. |
| SE-3 | Temporal decay applied correctly | Signals >365 days old contribute 0 points. Signals <30 days old at 100% weight. |
| SE-4 | Cross-signal multipliers firing | A company with Tier 1 + Tier 2 signals scores higher than one with either alone. |
| SE-5 | DPS threshold alert fires on crossing | When a company is manually moved from DPS 40 → 41, a threshold alert is logged and delivered. |
| SE-6 | Daily cron runs unattended | Cron runs at scheduled time. Results logged. No manual intervention required. |
| SE-7 | Darren pipeline review sign-off | Darren reviews top 10 ranked companies and confirms DPS scores are directionally correct. |
Engagement Engine — Validation Checklist
| # | Validation Test | Pass Criteria |
|---|
| EE-1 | DPS trigger → engagement record created | Within 1 minute of threshold alert, engagement record exists with correct status. |
| EE-2 | Intelligence package quality | Darren reads a generated intel package and confirms it is accurate, useful, and banker-ready. |
| EE-3 | Warm path scan — covers all sources | Scan checks LinkedIn, deal history, associations. Output clearly identifies warm/cold path. |
| EE-4 | Outreach messages — no signal references | Darren reviews 5 drafted messages. Zero references to company-specific signals, liens, or financial data. |
| EE-5 | Cadence management — auto-pause on response | When response is simulated, sequence pauses and banker is notified. No further automated messages sent. |
| EE-6 | Meeting brief quality | Banker reads generated meeting brief and confirms it provides useful preparation without exposing surveillance. |
| EE-7 | Preliminary assessment quality | Assessment is coherent, sector-appropriate, and sufficiently detailed to demonstrate Maverix capability. |
| EE-8 | Full lifecycle — 3 companies validated | Darren signs off on complete lifecycle output for 3 companies. Ready for live outreach. |
What Comes Next
- Week 4: First live outreach — 3–5 companies. Real bankers. Real messages. Engagement Engine manages cadence, bankers own all conversations.
- Multi-sector expansion — Healthcare (SNF/ALF) is the next vertical. Signal types already built for CPG; sector-specific signals for SNF/ALF to be added.
- Commercial data integrations — D&B PAYDEX, Experian Business, PitchBook for stronger Tier 2 signals.
- Engagement launch module — When a mandate is signed, agent immediately creates workspace, data room, data request list, and starts the 13-week cash flow model skeleton.
- Referral flywheel tracking — Alumni network, post-engagement relationship management, reverse referral cultivation.
Built by
Maverix Studio
darren@maverixstudio.com · maverixstudio.com
Read alongside: Maverix Restructuring — Project Brief v1.0