Most investors have the same information. The difference isn't insight—it's infrastructure. HARVEST PROTOCOL is a complete investment operating system with quantitative scoring, deployment gates, and behavioral circuit breakers — shipped as a Claude Code plugin with a license-gated server-side engine. Bring your own API keys; we never see your portfolio.
Behavioral finance research documents an average 1.5–2% annual return loss from preventable decision errors — selling at panic lows, deploying at FOMO peaks, holding positions whose thesis broke months ago.
Morningstar Mind the Gap 2024; DALBAR QAIB 2024 — investor returns lagged fund returns by 1.68% annualized over 10 years.
On a $250,000 portfolio, that is $4,375 per year. Over 10 years compounded: over $60,000 in avoidable wealth destruction. Not from bad picks. From bad process.
"The most valuable thing the system does is tell me to hold back. Not because the company is bad—but because the math doesn't justify the risk at this price, in this environment."
— Early Access User · $480K Portfolio · Software Engineering Director
Representative examples. Individual outcomes vary. Historical screens for illustration only. Verify independently. Not financial advice.
THE AI IS THE REASONING ENGINE. THE PROTOCOL IS THE PRODUCT.
HARVEST PROTOCOL gives Claude a 500-page protocol—with a Deployment Gate, 10-factor scoring, Expected Value Framework, and 9 exit triggers—that converts analysis into a sized, routed, behaviorally-enforced decision. The AI is the reasoning engine. The protocol is the product.
| CAPABILITY | AI ALONE | WITH PROTOCOL |
|---|---|---|
| Analysis quality | High—but unfocused | 40+ quantitative checks, structured |
| Position sizing | ✗ None | ✓ EV-weighted, risk-adjusted |
| Market timing gate | ✗ None | ✓ VIX/CAPE/sentiment deployment gate |
| Exit criteria | ✗ None | ✓ 9 pre-committed triggers |
| Behavioral circuit breakers | ✗ None | ✓ VP disconfirmation, EV decay |
| FEATURE | HARVEST | Robo-Advisors | Stock Pick Services | Asking Claude/GPT Directly |
|---|---|---|---|---|
| Deployment Gate (VIX/CAPE) | ✓ | ✗ | ✗ | ✗ |
| 9 pre-committed exit triggers | ✓ | Auto rebal only | ✗ | ✗ |
| EV-weighted position sizing | ✓ | Fixed % | ✗ | ✗ |
| 40+ quantitative checks | ✓ | ✗ | Varies | Unstructured |
| Behavioral circuit breakers | ✓ | ✗ | ✗ | ✗ |
| FIRE / tax routing integration | ✓ | ✗ | ✗ | ✗ |
| Local — your data stays yours | ✓ | Full account access | Cloud | Cloud |
| Cost | $297 once | 0.25% AUM/yr | $99–299/yr | $0–20/mo |
| Decisions or just analysis? | Sized + routed | Decisions | Ideas only | Essays only |
| You stay the operator | ✓ | ✗ | Partial | ✓ |
HARVEST PROTOCOL isn't a stock picker or a newsletter. It's a complete investment operating system with four integrated components that work together to convert research into repeatable, behaviorally-enforced decisions.
30 chapters. Knowledge layer spanning Graham to Dalio. Compiled mental models, frameworks, and decision heuristics from 8 decades of investment literature.
10-sweep BEST protocol with 40+ quantitative checks. Composite scoring, expectation gap analysis, Beneish M-Score, Piotroski F-Score, and VP modeling.
FIRE planning integration, buy cycle management, tax-advantaged routing, and market-condition deployment gates that prevent bad timing.
License-gated Cloudflare Worker hosts the proprietary scoring math. Plugin invokes harvest_score / harvest_ev_band / harvest_kelly / harvest_screen / harvest_regime tools. Deterministic numeric output; no client-side install.
mcp.getharvestprotocol.com. Deterministic numeric output, no client install.claude plugin install harvest-protocol. Slash commands (/best-cycle, /quick-screen) work from any Claude Code session. No file drops, no Project setup.
Not a stock tip. Not a newsletter pick. A complete mechanical analysis — every number sourced, every score traceable, every decision pre-committed. Here's a real analyze MSFT output.
invest $X) is the signature monthly workflow — 7 cold sweeps across the market, multi-candidate composite scoring, sized portfolio of picks + index sleeve. Answers "what should I buy this month?"analyze MSFT output.
Framework output shown for educational illustration. Scores reflect methodology application, not investment recommendations. Not financial advice.
See the scoring engine in action. 3 free analyses per day. No account required.
Placeholder scenarios — illustrate realistic system outcomes. Replace with 3+ real beta testimonials (name, role, portfolio size, concrete outcome) before opening cart. Individual results vary. Not financial advice.
In March I was 20 minutes from buying NVDA at $950 on momentum. The Expectation Gap came back unfavorable. I passed. One pass paid for the system 50x over.
Software Engineer, 34, $340K portfolio
Composite 81—but EV came back at 7%. Passed. It dropped 28% two quarters later.
Finance Professional, 41, $680K portfolio
They all give you ideas. None of them give you a process. This is the first thing that tells me not just what to buy but when NOT to.
Physician, 38, $220K portfolio
Representative feedback. Names anonymized. Results vary. Not financial advice.
Behavioral finance research documents an average 1.5–2% annual return loss from preventable decision errors. On a $250K portfolio, eliminating that gap is worth $3,750–$5,000 per year. The system costs $297 once.
One avoided panic-exit on a 15% decline in a $10,000 position pays for 4 years of the system.
Behavior gap estimate from Morningstar Mind the Gap 2024 and DALBAR QAIB 2024. Individual results vary.
Buy-side research desk: $2,400+/mo · Seeking Alpha Premium: $299/yr · Robo-advisor on $250K: $625/yr · HARVEST PROTOCOL: $297 once
No Claude subscription required to start. Begin with a free Claude account. Upgrade to Pro ($20/month) when your workflow demands it.
30-day money-back guarantee on all tiers · Personal use license · Not financial advice
claude plugin install harvest-protocol), paste your license key from your purchase email into a single env var, and register your own Alpha Vantage / FMP / EDGAR API keys with one CLI command. Then run any of the 14 ready-to-paste workflow prompts — Morning Watchdog, Quick Screen, Monthly BEST Cycle, Quarterly Review — or invoke the slash commands (invest $5000 for the full BEST cycle, /quick-screen MSCI for a targeted screen) directly. The proprietary scoring math runs server-side; nothing to install or configure beyond the plugin. If you can install a CLI tool, you can run the system.
QC_PROTOCOL.md file forces three behaviors on every workflow: (1) Claude must show a calculation_scratchpad for any numerical claim — named operands, source-tagged values, visible arithmetic; (2) when a number can't be sourced, Claude must mark it NULL and apply the DATA_DEFICIENT protocol rather than estimate; (3) every figure must cite its source (web search URL, broker statement, EDGAR filing). The plugin's scoring tools (harvest_score, harvest_ev_band, harvest_kelly, etc.) return deterministic numeric output from the server-side engine — no hallucination surface for the math itself. Your BYO API keys (Alpha Vantage / FMP / SEC EDGAR) feed verified data into both Claude's reasoning and the engine. You are still the final auditor against primary sources (10-K, 10-Q, earnings transcripts).
/best-cycle, /quick-screen, etc. Claude Code is included with Claude Pro ($20/mo) and Claude Max plans. The bundled MODEL_ROUTING.md tells you which model fits which task: most daily workflows run on Sonnet; the adversarial workflows (Bear Case Stress Test, VP Self-Audit, full Monthly BEST Cycle) benefit from Opus's deeper judgment and Extended Thinking. The proprietary scoring math runs server-side regardless of your Claude tier — your model choice affects reasoning depth, not engine accuracy.
QC_PROTOCOL.md requires the model to show its work — every F-Score, Z-Score, M-Score, EV calculation, and Composite Score appears inside a calculation_scratchpad block with named operands, source-tagged values, and visible arithmetic. If Claude skips the scratchpad, the protocol says you reject the output. Each workflow prompt includes verification cues — specific things to spot-check (e.g., "F-Score total should equal sum of 9 sub-checks", "EV must equal weighted sum of bull/base/bear returns"). When something looks off, the protocol provides recovery prompts: "Re-derive that score from these named operands and cite the source for each input." You're never trusting an opaque output — the engine math is reproducible, and Claude's interpretation around it is auditable.
claude plugin install harvest-protocol and paste your license key from the purchase email; (2) register your own Alpha Vantage / FMP / EDGAR API keys via the BYO setup command; (3) run /quick-screen MSCI (or any ticker) for your first verified output. The bundled userguide.html tab 1 is the 30-minute Quick Start with screenshots and verification cues. By the end you'll have run a real workflow against a real stock. The week-1 calendar lays out your daily and weekly routines (Morning Watchdog, Sunday Evening Brief, mid-week scan). Complete-tier and above buyers also get a "Duplicate as Template" link for the Notion implementation workspace — set it up after your first cycle.
mcp.getharvestprotocol.com and is licensed independently of any AI vendor — your Worker access doesn't expire when Anthropic ships a new model. Buyers receive free updates within v6.x; major versions (v7+) may require a paid upgrade with a guaranteed 18-month backward-compatibility window. Founders Lifetime tier locks in v6.x AND v7.x updates for life.
When markets fall 25%, most investors panic. The ones with a Deployment Gate, pre-committed sell criteria, and a VP disconfirmation trigger don't panic—they execute.
Run one full monthly BEST cycle. If it doesn't change how you make your next investment decision — full refund, no questions, you keep the Playbook.
That's the deal. We don't promise returns. We promise the protocol will change how you decide — or we don't keep your money.
30-day money-back · Personal use license · All future updates included · Not financial advice. All investments involve risk.