13 Agents · 1 Masterfile · 4:30 AM → 6:30 AM Daily
The Signal Newsletter Intelligence System
A complete AI-powered newsletter engine that converts free subscribers into paying customers. 13 specialist agents. Daily email + 9-email onboarding cadence. A hook engine that gets smarter every week.
13AI Agents
1Masterfile
5QA Layers
5Segments
9Onboarding Emails
The Mission
What We're Building
The Signal is not a newsletter. It's a revenue engine dressed as a newsletter. Every financial publisher that crossed $100M — Stansberry ($409M), Motley Fool ($400M+), Morning Brew — ran the same playbook: free signal builds trust, trust converts to subscribers, subscribers convert to paying customers through a deliberate cadence.
$0
The Signal (Free)
$149
Premium / Year
$999
Pro / Year
$5K
Platform / Year
The 3 Jobs of Every Email
1. Create Trust
Lead with a real, data-backed signal. TradeAlgo composite score. Options sweep data. Insider buy confirmation. The reader must feel that opening this email gave them an edge they couldn't have gotten anywhere else.
2. Show the Value Gap
Free subscribers see the signal. Pro members get the full playbook — entry, exit, stop, position sizing. Every email makes this gap visible. Not aggressively. Just clearly. The gap is the sale.
3. Move the Reader
One CTA per email. Matched to the subscriber's position in the 4-week cadence. Week 1 earns trust. Week 4 asks for the upgrade. The Optimizer tracks what moves people and adjusts the system weekly.
The Core Principle: The Masterfile is the institution. Agents are replaceable. Models are interchangeable. The voice, signal standards, cadence strategy, and product positioning persist regardless of what AI powers the system underneath.
System Architecture
One Open. A Cadence Begins.
The moment a subscriber opens their first email, the upsell engine activates. Every interaction — an open, a click, a link visit — advances them along a behavioral cadence designed to convert free subscribers into paying customers at exactly the right moment.
First OpenCadence triggered
→
Signal 2Deepen value · Day 3
→
Show the GapWhat Premium got · Day 7
→
Prove ItTrack record · Day 10
→
Soft AskDay 14 · One CTA
↓
No Engagement30-day win-back fires
→
Click DetectedAccelerate cadence
→
Upgrade Page VisitUrgency email same day
→
ConvertedPro welcome sequence
Data Sources
TradeAlgo Scoring Engine
500+ stocks scored 0–100 daily. Composite of momentum, flow, fundamentals, and sentiment. The backbone of every featured pick. Min score 72 to include, 88 to feature.
Off-exchange block trades above $500K. Printed near key levels (52W highs, round numbers, prior close). Institutional accumulation signal when paired with options flow.
SEC EDGAR Form 4
C-suite open-market purchases above $250K. Discretionary buys only — excludes grants, option exercises, and 10b5-1 auto-plans. The cleanest insider signal available.
Multi-Agent Pipeline
13 Agents. 4 Phases. 1 Newsletter Director.
A full editorial and intelligence team operating at machine speed. Each agent is a specialist governed by the Masterfile. The Newsletter Director orchestrates every phase and makes the final send decision. Three new agents — Narrative Director, Hook Performance, and Conversion Intelligence — close the loop between every send and the next improvement.
Command Layer
DIR Newsletter Director (Claude Opus)
Reads the Masterfile. Initiates the pipeline at 4:30 AM. Delegates to all 10 agents in sequence. Manages revision cycles when quality gates fail. Reviews all three reviewer scorecards. Makes the final send/hold decision. The only agent with Beehiiv send authority. If any reviewer scores below 8/10, the Director sends back to the relevant writing agent with specific revision instructions. Maximum 3 revision cycles before escalating to human review.
Controls: Full Pipeline · All 10 Agents · Final Send Decision
↓
Phase 1 — Intelligence · 4:30 AM
01 Market Intelligence Agent
Scans all five data sources simultaneously: TradeAlgo scoring engine, Polygon options flow, Polygon dark pool, SEC EDGAR Form 4, and market regime API. Cross-references all signals to assign convergence tiers (TRIPLE / DOUBLE / SINGLE). Ranks stocks by composite score within each tier. Applies Masterfile filters: min score 72, exclude sub-$5 stocks, max 2 picks per sector. Builds the full MarketSnapshot that every writing agent depends on.
Pulls Beehiiv subscriber analytics for each segment: open rate, click rate, unsubscribe rate, top-clicked sections, A/B winner history. Reads the cadence tracker to determine which week each segment is on (1–4) and assigns the correct CTA. Flags demographic patterns from Beehiiv data. Identifies at-risk subscribers. Outputs an AudienceBrief that tells the writing agents who they're writing to and what will move them today.
Output: AudienceBrief per segment — cadence week, CTA assignment, tone notes, engagement health
03 Competitive Intelligence Agent
Scans the last 3 sends from major competing newsletters (Morning Brew, Pomp Letter, Unusual Whales digest, The Daily Upside). Identifies what angle they led with today. Finds the gap — what did they miss or underweight? Reports to the Signal Writer so The Signal can take a differentiated angle rather than repeating what the reader may have already seen. Also surfaces any competitor subject line patterns worth A/B testing against.
Receives the MarketSnapshot, AudienceBrief, and CompIntel brief. Writes everything from the hook through the risk watch section. The hook is the most important output — it must reference a specific number, feel written at this exact moment, and make the reader feel the signal. The featured pick includes score, convergence tier, entry zone, data-backed rationale, and a specific risk note. Supporting picks are compact — one paragraph each. Every claim traces to the MarketSnapshot. No editorializing.
Takes the Signal Writer's draft and adds the narrative layer. Writes "What This Means" for long-term and beginner segments. For Week 2 cadence, writes "The Gap" — showing what Pro members received that free subscribers didn't. For Week 3, writes the outcome follow-up on last week's picks with actual price data. Places exactly one CTA at the end, matched to the cadence week from the AudienceBrief. The CTA must feel earned — the value must be demonstrated before the ask is made. Total word count must stay within Masterfile limits.
Output: Complete email body — Signal Writer draft + story layer + single CTA
06 Subject Line Agent
Reads the full email and writes 3 subject line variants — one data-driven (specific number), one curiosity-gap (creates tension to resolve), one actionable (direct value proposition). Checks A/B winner history from AudienceBrief and sets the strongest historical angle as the primary. Validates all three against Masterfile spam word list. Writes matching preview text for each. Outputs all three formatted for Beehiiv A/B configuration with 34/33/33 split.
Phase 3 — Triple Review · 5:30 AM (All Must Score 8+/10)
07 Fact Checker
Verifies every factual claim against the original MarketSnapshot. Checks: all tickers real and in snapshot, prices within 1.5% of snapshot, scores match exactly, convergence tiers correct, options/dark pool/insider data not embellished, SPY and VIX levels accurate, earnings dates correct. Any price off by more than 1.5% = FAIL. Wrong executive title = FAIL. Phantom ticker = immediate FAIL. Returns scorecard with specific line references for every flag.
Output: Fact Check Report — PASS/FAIL per claim, specific fixes required
08 Compliance & Deliverability Agent
Two-track review. Compliance: disclaimer present in footer, unsubscribe link present, no guaranteed return language, no personal investment advice ("you should buy"), no implied non-public information. Deliverability: subject line spam word scan, body spam trigger check, text-to-link ratio, HTML template variable resolution, preview text length, exclamation mark count. Returns combined scorecard. Hard fail on any compliance issue. Deliverability warnings allowed if count is low.
Output: Compliance + Deliverability Report — dual PASS/FAIL with specific fix instructions
09 Brand & Tone Reviewer
Scores the email on 5 dimensions — all must hit 8/10: (1) Voice Authenticity: does this sound like TradeAlgo's sharpest analyst? (2) Signal Strength: is the featured pick specific and data-backed or generic? (3) Segment Fit: is the tone and content right for this audience? (4) CTA Effectiveness: is there exactly one CTA, is it earned, is it placed at the end? (5) Readability: can it be read in under 90 seconds, are paragraphs max 3 sentences? Forbidden phrase scan: "it's worth noting," "in today's volatile market," "as we navigate," "going forward," "at the end of the day." Also reviews all 3 subject line variants.
Output: Brand & Tone Scorecard — 5 dimensions × 10 + subject line review + PASS/FAIL
All 3 reviewers must independently pass. If ANY fails → returns to writing agents with specific fixes. Max 3 revision cycles. On 3rd failure → escalates to human review, send is held.
↓
Phase 4 — Send & Learn · 6:00 AM + 9:00 AM
10 Send & Distribution Agent
Receives the reviewed email + all 3 subject variants. Renders HTML via the html-renderer. POSTs to Beehiiv API — sets publication ID, audience segment, scheduled send time (6:30 AM), and configures the A/B subject line test (34/33/33 split, open-rate winner, 4-hour decision window). Archives the full run to runs.log with all metadata. Confirms Beehiiv post ID. Reports send confirmation to Newsletter Director.
Output: Beehiiv post ID, send confirmation, run log entry
11 Narrative Director
Runs a story test on every draft before it reaches the reviewers. Asks three questions: (1) Does this email tell a story or recap news? (2) Is ATLAS AI the hero, not just a feature? (3) Does the subscriber feel the problem before the solution is offered? If any answer is no, rewrites the opening and the footnote, returns to Value Writer for integration. Governs onboarding sequence tone — ensures every cadence email advances the arc, never repeats, and maintains Atlas's voice across all 9 emails.
Output: Story test scorecard · Rewritten opening if failed · Approved narrative arc
12 Hook Performance Agent
Owns the hook library. Selects the daily hook from a pool of 10 active variants, each tagged by angle: fear, curiosity, social proof, data, identity, time-pressure, outcome, contrarian, aspiration, specificity. Logs every send with the hook used and resulting open rate. Every Monday at 8 AM: retires hooks below 30% open rate, promotes hooks above 45%, generates 10 new variants based on winning angle patterns. By week 10 the hook library is a proprietary competitive asset — no other newsletter has learned what opens your specific audience.
Tracks the Free → Premium → Pro funnel. Runs every Wednesday at 10 AM. Maps which emails, hooks, and lock-block reveals drove upgrade page visits in the prior 7 days. Identifies where subscribers stalled in the onboarding cadence. Flags churn risk (30+ days no open). Feeds conversion patterns back to the Value Writer and Hook Performance Agent — so every send gets better at the one job that matters: turning a free subscriber into a paying customer. Proposes onboarding cadence adjustments (email order, timing, CTA placement) backed by cohort data.
Newsletter Director confirms send at 6:30 AM. Signal Reviewer scores performance at 9 AM. Hook Performance Agent logs the A/B winner at 10:30 AM. Conversion Intelligence Agent reports the funnel every Wednesday. Optimizer reads 7 days of scores every Monday and proposes Masterfile improvements. Each week the system gets incrementally smarter — higher open rates, better conversion, tighter onboarding — because the loop never stops closing.
Every new subscriber enters a 9-email arc the moment they sign up. It's not a drip campaign. It's a progressive story with a villain (time), a hero (ATLAS AI), and a single ask at the end. Each email follows Carlos's exact structure — opening, three things, market context, featured content, the number, Atlas footnote.
The Villain
Time. Life. The market moves while investors are in meetings, with family, living their life. You can't monitor 15,412 stocks. The market doesn't care.
The Hero
ATLAS AI. Runs at 3:47 AM. Processes 2.3M data points. Finds the signal before the crowd. By 6:30 AM the analysis is done — you just decide what to do with it.
The Fear
Being late. Everyone else knew. You found out on CNBC after the move. Three signals. Three times close but not in. The question isn't whether the edge is real — you've watched it work.
Email
Subject
Core Job
CTA
Day 0
"You just got early."
Set tone. Deliver first value. Build anticipation.
None — earn trust first
Day 1
"The market moved while you were in a meeting."
Introduce the villain. Three real examples from this week. The math of 1,638 hours.
None
Day 3
"3:47 AM. You were asleep. ATLAS AI wasn't."
AI as hero. 2.3M data points. 4:12 AM rankings. Proof of capability.
None
Day 5
"Premium subscribers knew 4 hours before this hit the news."
Exact side-by-side: what free got vs what Premium got. First soft CTA.
Unlock Premium →
Day 8
"The S&P returned 24% last year."
Performance proof. S&P comparison. The math: one signal pays for 364 days.
Track record →
Day 10
"Let me guess what you're thinking."
Objection busting. "Is this hype? What if it's wrong? Why $149?" Direct honest answers.
Track record →
Day 14
"You can't monitor 2.3M data points. It can."
200 stocks (human ceiling) vs 15,412 (ATLAS AI). Identity: which investor do you want to be?
See what Premium gets →
Day 21
"Atlas called it. The outcome landed."
Three missed signals listed. Compounding regret. "Third time I've watched this happen to you."
Stop being late →
Day 28
"Stop watching. Start knowing."
Two options. Identity language in both. Clean direct ask. No pressure — free stays free forever.
Get Premium $149/yr →
Behavioral Accelerators: Any link click advances the cadence 2 days. An upgrade page visit triggers a same-day follow-up (2hr delay). Three consecutive opens skips to Day 14. No open in 14 days triggers the win-back sequence. The cadence adapts to behavior — not a calendar.
10 Hooks a Week. The System Learns What Opens Your Audience.
Most newsletters write a subject line and hope. The Signal tests 10 hook variants every week, identifies what's working, retires what isn't, and generates 10 new hooks based on the winning patterns. By week 10 the hook library is a proprietary asset no competitor can replicate.
10 Active HooksIn the pool
→
3 Tested Per SendBeehiiv A/B split
→
Winner Logged10:30 AM daily
→
Monday ReviewRetire losers
→
10 New HooksBased on winners
The 10 Hook Angles
😰 Fear
"$6.9M moved before the open. Most people missed it."
🤔 Curiosity
"Breadth narrowed again. One stock didn't care."
📊 Data
"NVDA scored 93/100. Here's the convergence."
🪞 Identity
"The investor who was in it before the crowd knew."
⏱ Time
"3:47 AM. You were asleep. ATLAS AI wasn't."
📈 Outcome
"Atlas called it. The outcome landed. +18.4%."
🔄 Contrarian
"Everyone's watching NVDA. ATLAS AI is watching something else."
🛡 Protection
"VIX at 20 and breadth narrowing. Atlas sees a pattern."
👥 Social Proof
"Premium subscribers got this at 6:30 AM. The news broke at 2:17 PM."
🎯 Specificity
"One stock. Score 94/100. Here's what ATLAS AI found."
Winner Threshold: 45%+ open rate = winning hook, stays in pool. Below 30% = retired, replaced Monday. The winning angle pattern (not just the specific hook) informs all 10 new variants. After 10 weeks, open rate patterns are unique to TradeAlgo's audience — no competitor can buy or copy this data.
Subscriber Segments
5 Audiences. 5 Voices. 5 Paths to Conversion.
The Masterfile defines exactly how each segment is written to, what they receive, and what product they're being moved toward. Every agent reads the segment definition before writing a single word. Fear/loss prevention governs all CTA framing — fear-driven subscribers retain at 70–85% vs 40–60% for greed-driven.
Ambitious Accumulator
Age 30–50, $150K–$400K portfolio, growth-focused
Tone: Direct, data-first, no fluff
Fear driver: Missing the move others caught
CTA target: Premium ($149/yr) → Pro ($499/yr)
Hook: Lead with score + convergence tier
Anxious Retiree
Age 55–75, $500K–$5M assets, capital preservation
Tone: Measured, protective, never alarmist
Fear driver: Outliving savings, sequence-of-returns risk
Always the featured pick. Highest conviction. Full analysis.
⚡ DOUBLE
Any two of three sources
75+
Supporting pick. Compact paragraph. Entry zone included.
🏛 SINGLE
One source only
88+
Include only if score is 88+. Brief mention.
Quality Standards
7-Dimension Rubric. All Must Score 8/10. No Exceptions.
Three independent reviewers score every email before it goes out. Each scores 7 dimensions. All must hit 8/10 or the email goes back. This is the standard that separates a newsletter people trust from one they unsubscribe from.
Dimension
What Gets Evaluated
Min
Data Accuracy
Every price, score, convergence tier, and dollar amount verified against the MarketSnapshot. No rounding that changes meaning. Correct executive titles.
8/10
Signal Strength
Featured pick is specific and data-backed — not generic market commentary. Entry zone present. Convergence tier called out. Risk note specific and honest.
8/10
Voice & Tone
Passes the TradeAlgo voice test. No forbidden phrases. No em-dash abuse. No AI tells ("certainly," "it's worth noting"). Sounds like a sharp analyst, not a chatbot.
8/10
Segment Fit
Content and tone match the segment. No options jargon in beginner email. Active traders get entry zones and speed. Long-term investors get context and conviction.
8/10
CTA Effectiveness
Exactly one CTA. Placed at the end. Matches the cadence week. Text matches the destination page. Value demonstrated before the ask. No buried or doubled CTAs.
8/10
Readability
Readable in under 90 seconds for active traders. Paragraphs max 3 sentences. No sentences over 30 words. Tickers and numbers scannable. Clear section headers.
8/10
Compliance
Disclaimer in footer. Unsubscribe link present. No guaranteed return language. No personal advice framing ("you should buy"). No spam triggers in subject or body.
8/10
The Revision Protocol: Any dimension below 8/10 returns the email to the responsible writing agent with specific line references and required fixes. Maximum 3 revision cycles. Third failure triggers human review and the send is held. Zero compromises on quality — one bad email erases weeks of trust.
Upsell Engine
The 4-Week Cadence That Builds Brand-Loyal Customers
Every subscriber enters the cadence when they sign up. The Audience Intelligence Agent tracks which week each segment is on. The Value Writer matches the narrative to the cadence. The Optimizer can extend or compress weeks based on engagement data.
W1
Establish Credibility — Pure Signal
No hard sell. 3–5 picks with full data backing. The goal is to make the reader feel The Signal is the smartest thing in their inbox. CTA links to the methodology page — zero friction, zero ask.
Pure signal qualityTrack record buildingCTA: See our methodology →
W2
Show the Gap — What Pro Members Got
"Free subscribers saw the NVDA signal. Pro members received the full playbook at 6:04 AM — entry trigger, target, stop, position sizing." Show the gap factually, not aggressively. CTA: 7-day free Pro trial.
Social proofValue gap visibleCTA: Start Pro free →
W3
Prove the Outcome — Follow Up With Results
Go back to Week 1's picks. Show what happened. "NVDA: +10.5% in 6 days. JPM: +2.1%. KKR: flat." Real outcomes, real numbers. No cherry-picking — report everything. This is the most powerful trust signal in financial publishing.
Outcome trackingFull transparencyCTA: See full track record →
W4
Make the Ask — Direct Close
Full pitch. "If you've been reading The Signal this month, you've seen us call NVDA, JPM, and META before they moved. Pro members got the full playbook in real time. $999/year. Less than one bad trade." Direct, honest, no pressure.
Direct closePrice anchoringCTA: Upgrade to Pro →
Cadence Override Rule: If open rate drops below 35% for any segment during the cadence, the Audience Intelligence Agent automatically resets that segment to Week 1 (credibility rebuild). The trust must be re-established before the ask can be made.
Revenue Architecture
The Signal Drives Every Product in the Stack
The email list is the distribution. The Signal is the trust-builder. The products are the revenue. Every send is an opportunity to move a subscriber up the ladder. The Optimizer tracks which emails drive the most product clicks and continuously refines the approach.
✉
The Signal
Free · Daily · 6:30 AM
→
28d
Onboarding Arc
9-email story cadence
→
$
Premium
$149 / year
→
$$
Pro
$499 / year
→
$$$
Elite
$2,500 / year
→
AUM
Managed
$50K+ AUM
Performance Targets Per Send
Metric
Target
Industry Avg
Why It Matters
Open Rate
42%+
21%
Primary signal of subject line quality and sender reputation
Click Rate
8%+
2.5%
Every click is a product touchpoint — this is what the Optimizer chases
Unsubscribe Rate
<0.3%
0.5%
Above this triggers immediate pipeline review — quality gate tightens
Pro Conversion
2–5% of clickers
<1%
The revenue metric — Signal Reviewer tracks attribution per CTA
Execution
90 Days to a Self-Improving Revenue Machine
The pipeline is operational. The first email produced a 93/100 quality score. Two API keys away from live sends. The 90-day plan takes The Signal from first send to fully optimized, self-improving cadence.
1–2
Foundation — Wire Up & First Sends
Beehiiv API key + TradeAlgo API URLs plugged into .env. First real send to test segment. Install cron (4:30 AM pipeline). First Signal Reviewer run at 9 AM. Calibrate quality gates. Target: 40%+ open rate on first send.
First live sendCron installedQuality calibrated
3–4
Cadence — All 3 Segments Running
All three segments live. Cadence tracker initialized — each segment on Week 1. First Optimizer run at end of Week 2. A/B subject line data accumulating. First outcome follow-ups from Week 1 picks posted.
3 segments liveFirst optimizer runA/B data building
5–8
Optimization — Loop Closing Weekly
Optimizer running every Monday. Masterfile getting updated weekly from real engagement data. Churn prediction identifying at-risk subscribers. Week 4 cadence sends for first cohort — first direct Pro upgrade ask. Track conversion rate.
Weekly optimizationFirst Pro askConversion tracked
9–12
Scale — Self-Improving Machine
System has 6+ weeks of engagement data. Masterfile has been updated 6 times by Optimizer. Subject line win rates stabilized. Conversion rates tracked by cadence week. TradeAlgo can now tie newsletter sends to product revenue directly.
42%+ open rate sustained8%+ click rateMeasurable Pro conversions
Every agent is defined in a standalone .md file formatted for training in Claude Code. The Masterfile governs all of them. The pipeline scripts execute the full workflow automatically.
the-signal/
├── agents/
│ ├── 00-masterfile.md# The institution — voice, tone, thresholds, cadence, product stack
│ ├── 01-market-intelligence-agent.md# Scans all data, builds MarketSnapshot, convergence scoring
│ ├── 02-audience-intelligence-agent.md# Subscriber analytics, cadence assignment, AudienceBrief
│ ├── 03-competitive-intelligence-agent.md# Monitors competing newsletters, finds differentiation angle
│ ├── 04-signal-writer.md# Hook, market pulse, featured pick, supporting picks, risk watch
│ ├── 05-value-writer.md# Story layer, gap narrative, outcome follow-up, single CTA
│ ├── 06-subject-line-agent.md# 3 A/B variants (data/curiosity/actionable) + preview text
│ ├── 07-fact-checker.md# Every claim verified against MarketSnapshot — truth gate
│ ├── 08-compliance-deliverability-agent.md# Legal language + spam score + inbox placement
│ ├── 09-brand-tone-reviewer.md# 5-dimension score, all must hit 8/10 — TradeAlgo voice gate
│ └── 10-optimizer.md# Weekly analysis → max 3 Masterfile proposals → --apply
├── atlas/
│ └── masterfile.yaml# Executable config — segments, thresholds, A/B, benchmarks
├── content/generation/
│ ├── pipeline.py# Orchestrator — 4:30 AM, runs all agents in sequence
│ ├── data-ingestion.py# Signal Writer data layer — async fetch all sources
│ ├── prompt-builder.py# Assembles Claude prompt from snapshot + masterfile
│ ├── quality-check.py# 5-layer validation gate — all must pass before send
│ ├── subject-line-generator.py# 3 variants, spam validation, Beehiiv A/B config
│ ├── html-renderer.py# Markdown to Beehiiv-ready HTML with full CSS
│ └── mock-data.py# Full test snapshot — runs pipeline without live APIs
├── analytics/optimization/
│ ├── content-scoring.py# Signal Reviewer — composite score per send, JSONL storage
│ ├── masterfile-optimizer.py# Optimizer — weekly Claude analysis → proposals → --apply
│ └── churn-prediction.py# At-risk subscriber detection, re-engagement triggers
├── automation/
│ └── daily-pipeline.yaml# Cron: 4:30 AM pipeline · 9 AM scorer · Monday optimizer
├── .env# Anthropic + Polygon live. Beehiiv + TradeAlgo pending.
├── .env.example
├── requirements.txt
└── README.md
Component
Status
Notes
Masterfile (00-masterfile.md + masterfile.yaml)
Built
3 segments, signal thresholds, 4-week cadence, optimizer guards, benchmarks
Agent 01 — Market Intelligence
Built
Full .md definition + data-ingestion.py running with mock data
Agent 02 — Audience Intelligence
Built
Full .md definition — awaiting Beehiiv API for live analytics