Skip to content

P2B-C4 · Real Walkthrough (NVDA)

Key Insight

Running a real case for the first time — applying the 4D framework to one real ticker will reveal where you only thought you understood.

Layer 2 · Analyzing One Ticker — After learning the industry, now apply it to a single stock

L2-C4 (5 chapters total). After this chapter, you'll be able to complete a full thesis cycle with NVDA — from transcript to 4D yaml to KPI predictions to post-earnings update.

1. The Problem: The Framework Works in Your Head, But Freezes on a Real Ticker

In C2 you wrote a yaml template, in C3 you learned the terminology. But when you get the NVDA Q4 transcript, you still don't know:

  • The transcript is 100 pages — which section should you jump to?
  • Management says "We had a record quarter" — is this information or noise?
  • An analyst asks "Vera Rubin timing?", CEO replies "production samples H1, full ramp H2" — is this bull or bear?
  • How do you extract 3-5 KPI predictions from all this text?

→ Running the framework in a vacuum doesn't count. Go through the 6 steps with one real ticker, and you'll see where you only thought you understood C2/C3.

2. The Solution: Run the 6-Step Cycle with NVDA

NVDA is the best ticker to start with because:

  • 8 quarters of transcripts are publicly available (free on Motley Fool)
  • Public 13F filings show institutional holdings changes + earnings timeline for 8 quarters
  • The next catalyst (Q1 FY27 earnings) is right around the corner — you can verify in real time

6-Step Cycle (each step clearly maps to which dimension of the 4D framework):

Step What to Do Primary 4D Dimension
Step 1 Assess the current state (what the company is) Background (C2 yaml header)
Step 2 Read the transcript WHAT + WHY (find segments / guidance)
Step 3 Fill in the 4D yaml All 4 dimensions
Step 4 Break down KPI predictions SO WHAT (catalyst look_for)
Step 5 Multi-perspective sanity check RISKS (anti-thesis)
Step 6 Post-earnings update All 4 dimensions (re-fill)

3. How It Works: 6-Step Detailed Walkthrough

Step 1 — Assess the Current State (5 minutes, fill in C2 yaml header)

Field NVDA Value
ticker NVDA (NVIDIA Corp), NASDAQ
Market Cap ~$3.3T (formerly world #1)
Business AI GPU + Data Center Networking + Autonomous Driving
Latest thesis version v91 v8 (includes CRWV circular financing red_flag)
Latest financials Q4 FY26 (reported 2026-02), revenue $39.3B, gross margin 75%, EPS $0.89

Next catalyst: Q1 FY27 earnings on 2026-05-20 — this is the best time for a walkthrough.

Step 2 — Read the Q4 FY26 Transcript (1 hour, fill in WHAT + WHY)

Start with the publicly available Motley Fool NVDA Q4 2026 transcript (Motley Fool publishes transcripts 1-2 days after earnings).

Beginner's Reading Method:

  1. Skip the opening — The CEO always says "we had a record quarter," no information value. Go straight to the finance section.
  2. Look at segments + YoY (this is WHAT):
    • Data Center: $35.6B (+93% YoY) ← Main driver
    • Gaming: $2.5B (+13%)
    • Auto: $0.6B (+103%)
  3. Look at guidance (this is the SO WHAT anchor):
    • Q1 FY27 revenue: ~$43B (consensus $44B)
    • Gross margin: 73.5-74% (down from 75%, Blackwell ramp costs)
  4. Look at Q&A — Highest density of information:
    • Analysts' real questions reveal what they care about
    • Management's unscripted answers reveal the real situation (vs prepared remarks marketing talk)

Key Q&A Questions Example (NVDA Q4 FY26 Real)

Q (Morgan Stanley): "Vera Rubin shipping timing?" A (Jensen): "Production samples shipping H1 calendar 2026, full ramp H2" → Signal: Rubin is 1-2 quarters ahead of expectations, fill into WHAT supports (bull for FY27).

Q (Bernstein): "How are you managing Samsung HBM3e qualification?" A: "We work with all three [Hynix/Micron/Samsung], qualifying Samsung is ongoing process" → Signal: Samsung not qualified is bear, fill into RISKS red_flag (but NVDA has backup, not fatal).

Step 3 — Fill in the 4D yaml (15 minutes)

Directly plug the signals extracted from Step 2 into the C2 template:

ticker: NVDA
view: bull
confidence: medium
core_thesis: |
  NVDA benefits from AI infrastructure capex, Q1 FY27 earnings will validate demand strength,
  watch customer quality risk (CRWV).

# WHAT dimension
supports:
  - "2026 hyperscaler capex $725B+, NVDA primary beneficiary"
  - "CRWV $99.4B backlog (NVDA $36.6B holdings)"
  - "327 institutions in 13F still hold NVDA as AI core position"
  - "Vera Rubin samples H1 2026 (Q&A Morgan Stanley)"

# RISKS dimension
red_flags:
  - text: "Samsung strike → HBM supply disruption"
    trigger: "Strike starts + lasts >7 days"
  - text: "CRWV-NVDA circular financing concerns"
    trigger: "CRWV customer concentration worsens OR NVDA exposure to CRWV expands"

# SO WHAT dimension
catalysts_90d:
  - date: 2026-05-20
    event: "Q1 FY27 earnings"
    look_for: "Data Center >$30B, FY27 guide >$200B, GM >74%"

price_outlook:
  base_90d: "$215-235"
  bull_90d: "$245-260"
  bear_90d: "$190-205"

Step 4 — Break Down KPI Predictions (15 minutes, fill in SO WHAT look_for)

Good analysis workflows (the hedge fund / Buffett / Anti-thesis series taught in Part 3) all require ≥ 3 verifiable KPI predictions per thesis. NVDA Q1 FY27 KPI example:

predictions:
  - kpi_name: revenue_quarterly
    expected_value: 43
    expected_range: [42.5, 44]
    unit: USD billion
    reporting_quarter: "Q1 FY27"
    verify_at: 2026-05-20
  - kpi_name: data_center_segment_revenue
    expected_value: 36
    expected_range: [34, 38]
    unit: USD billion
    reporting_quarter: "Q1 FY27"
    verify_at: 2026-05-20
  - kpi_name: gross_margin_non_gaap
    expected_value: 74.0
    expected_range: [73.5, 75]
    unit: percent
    reporting_quarter: "Q1 FY27"
    verify_at: 2026-05-20

After earnings (5/20), the KPI verifier automatically runs via cron, compares prediction vs actual, and tags each prediction as confirmed / partial / wrong.

This step is SO WHAT in action: No KPI predictions = no yardstick on catalyst day.

Step 5 — Multi-Perspective Sanity Check (30 minutes, strengthen RISKS)

Don't look at NVDA fundamentals in isolation — look at it simultaneously:

Multi-PM Perspectives (detailed in Layer 3)

  • Value PM (Buffett-style): "P/E 30x isn't cheap, but ROE 100%+ is unprecedented, hold for now"
  • Growth PM (Druckenmiller-style): "FY27 guide $200B+, growth still 50%+, add position"
  • Macro PM (Soros-style): "AI capex is the macro theme, but Fed hawkish + 10Y at 4.5% compresses valuations, hold without adding"

Anti-Thesis

If you were an NVDA bear, how would you argue?

  • "Hyperscaler capex is over-investment, $725B is the top of a bubble, ROI won't materialize" — SemiAnalysis (Dylan Patel) has written this angle
  • Your thesis must have an explicit reply: "I don't think it's a bubble because X / Y / Z" (see P3-C5 Anti-thesis)

→ A thesis without an anti-thesis is an emotional thesis.

Macro Overlay

Current regime = mixed, near-term = neutral — not risk_off, not risk_on. For NVDA: rising rates (compresses valuations) + improving liquidity (support) → net neutral, keep bull thesis unchanged.

Technical Overlay

NVDA trend = uptrend, momentum = neutral, ADX 30.9 (strong trend). - entry: $210-218 (20d SMA 211.3 + Bollinger middle) - exit/stop: $236-240 resistance / stop loss below $200 - → "Fundamental bull + Technical bull but momentum neutral" = don't chase, wait for a pullback to 20d MA to add

Step 6 — Post-Earnings Update on 5/20 (1 hr that night + 1 hr one week later)

Earnings Night (After-Hours)

  1. Read management commentary + Q&A (Motley Fool transcript out in 1-2 days)
  2. Compare against KPI predictions:
    • Actual revenue vs $43B median prediction
    • Actual Data Center vs $36B
    • Actual gross margin vs 74%
  3. Look at next quarter's guidance:
    • Q2 FY27 guide >$50B? = bull confirmation
    • <$45B? = warning

7-14 Days After Earnings

  1. 13F filings released (Q1 holdings around 5/15) — verify against 13F support
  2. Sell-side analyst upgrades/downgrades — look for differentiation
  3. Thesis decision: maintain v8 / upgrade to v9

4. vs C3 — What You Already Know

C3 taught you how to read individual terms. But C4 teaches you how to connect multiple terms into one complete cycle:

Dimension C3 You Can Do C4 You Can Do More
Understand single terms ✓ ✓
Extract signals from transcript ✗ ✓
Plug signals into 4D yaml ✗ ✓
Break down KPI predictions ✗ ✓
Verify after earnings ✗ ✓

C4 is the first time you run a closed loop — input → output → verify → re-input. Run it once, and all the gaps in your C2/C3 understanding will be exposed.

5. Try It: Complete One Cycle with NVDA, Then Pick Your Most Familiar Ticker for a Second Cycle

Task (~3 hours spread over 2 weeks):

  1. **Follow Along with NVDA** (1 hour today):

    • Open the Motley Fool NVDA Q4 FY26 transcript
    • Follow Step 2 to extract segments / guidance / Q&A
    • Follow Step 3 to fill in the complete yaml

  2. Verify with NVDA Q1 FY27 Earnings (30 minutes on 5/20 night):

    • Compare against Step 4's KPI predictions
    • See which are confirmed / partial / wrong
    • Which of your thesis supports are strengthened, which are invalidated

  3. Pick 1 Ticker You Know Best (TSM / AMD / GOOGL recommended because transcripts are easy to read), repeat Steps 1-5 (complete within 2 weeks)

Self-Check (C4 is only done after completing the second cycle):

  • You can explain to a friend "why I extracted the Vera Rubin section into supports, not catalysts"
  • Your KPI predictions have at least 3, each with an expected_range not just expected_value
  • You can articulate your own ticker's anti-thesis with strong arguments + invalidation triggers (not copied from the internet)

All 3 yes → C4 complete. 1 no → don't move to C5, go back and review C3.

6. What's Next

NVDA + 1 of your own tickers = you've run 2 cycles. But in practice, you'll be maintaining 5-10 active theses simultaneously.

You'll need:

  • Your own thesis storage format (markdown / Obsidian / git)
  • A weekly / monthly / quarterly review cadence (otherwise your thesis will fall out of sync)
  • 5 self-checks (to ensure each thesis meets C2-C4 standards)

→ L2-C5 · Write Your Own Thesis — Template + cadence + what your library should look like after 6 months.

7. Deep Dive (Optional): How the KPI Verifier Runs via Cron / How Multi-PM Automation Works

Click to expand prediction verification engineering

KPI Verifier Workflow:

  1. After each thesis is written, predictions are stored in a predictions table (kpi_name + expected_range + verify_at)
  2. On the verify_at night, cron runs kpi_verifier:
    • Queries a financial provider API (Yahoo / Polygon) for actuals
    • Compares against expected_range:
      • actual in [low, high] → confirmed
      • actual deviation < 15% → partial
      • actual deviation > 15% → wrong
    • Writes back to a prediction_results table
  3. Weekly / monthly aggregation: Which KPI types do you consistently get wrong (revenue? gross margin? customer growth?) → Which support sources are unreliable for you

Multi-PM Perspective Automation:

Each thesis runs through 3 LLM perspectives (value / growth / macro), each filling in view + confidence + 1 key reason. Then check if the 3 perspectives show consensus or divergence:

  • All 3 bull → high signal
  • 2 bull 1 neutral → medium
  • 1 bull 2 bear → your thesis has a risk you haven't seen, must rewrite

→ This mechanism transforms "I think I'm bull" into "3 independent perspectives all confirm I'm bull" — a qualitative difference in signal strength.