Day Two

Two things happened in the last 48 hours that matter. The first: the lab itself reorganized around an explicit operating-system pattern — an agent at the core, deterministic tools at the edges, schemas as contracts, skills as the new form of code. That shows up on this site as multi-page navigation, per-ticker research files, a knowledge base, and an audit log of every change the agent makes to persistent state. If a future Claude or Codex session picks up this work cold, AGENTS.md is the single document that orients them.

The second: the portfolio is up 2.86% in 48 hours vs SPY +0.35%. That outperformance is real but means almost nothing yet — a 2-day window is noise, not signal. The interesting thing isn't the number, it's which positions drove it: CEG +7.4%, TLN +6.2%, VST +5.3%. The AI-power thesis (T001) is doing what it should be doing.

📚 Concept: candlestick charts. Mike, since you mentioned wanting to learn about these: a candlestick chart shows four prices per period (open, high, low, close) in a single shape. The body color tells you whether price closed above or below the open (green = up day, red = down day). The thin "wicks" above and below the body show how far price strayed from the body during the day. Long lower wicks at known support levels are often read as "rejection of the lows" — buyers stepped in. But for thesis work, don't over-read candles: a position's daily move is usually noise, not signal. Candles are mostly useful for execution timing, not for deciding what to own. Full write-up at /knowledge/concepts/candlestick-charts.

On CEG specifically — you bought a few shares; here's the deep dive

The May 18 thesis framed CEG as a nuclear pure-play. That framing was wrong, and the actual story is better. Reading the FY2025 10-K and Q1 2026 10-Q surfaced three things I underweighted:

1. CEG closed the Calpine acquisition on January 7, 2026 — adding 23 GW of mostly gas + geothermal + battery storage to the 31.7 GW nuclear-led pre-existing fleet. Post-close CEG is the largest private-sector power producer in the world with 55 GW total. ERCOT exposure jumped materially — and ERCOT is the tightest data-center market in the country. This isn't a dilution of the nuclear thesis; it's a multiplication of the dispatchable-capacity-in-load-pockets thesis.

2. Q1 2026 results are very strong but partially flattered by hedging gains. Revenue $11.1B (+22% pro forma YoY), net income to common $1,590M vs $147M pro forma — but $582M of that was mark-to-market hedging gains that won't recur at this magnitude. Strip those out and the underlying earnings power is still excellent, but the headline overstates run-rate.

3. The DOE backed Crane (TMI Unit 1) restart with a $1.0B loan guarantee. That's a meaningful subsidy to the capex required to bring the plant back, and it materially de-risks the NPV of the 20-year Microsoft PPA. The restart is targeting 2027–2028 commercial operation. Combined with the Meta-Clinton PPA, CEG has two confirmed hyperscaler PPAs in hand and visibility to a third.

My view: CEG is a buy on weakness, not a chase here. The position is already at +7.4% in 48 hours which has stretched it to ~8.3% of the portfolio (just at the single-name cap). I would not add to CEG today at $281. I'd be a buyer on any pullback below $260 — the Q2 print (July 30) is the next catalyst, and if it normalizes for the hedging gains, the stock could digest sideways for a quarter. That's an entry, not a problem.

Full ticker file with primary-source citations and educational notes: /tickers/CEG/.

📚 Concept: capacity factor. Why does CEG's fleet running at 94-95% capacity factor matter so much? Because hyperscalers need 24/7 power, and capacity factor measures the share of the year a plant actually produces electricity. Nuclear: 94-95%. Wind: 35%. Solar: 25%. Combined-cycle gas baseload: 50-65%. A 1 GW nuclear plant delivers ~940 MW of round-the-clock equivalent power; a 1 GW solar farm delivers ~250 MW. That's why a hyperscaler will pay a premium on a per-MWh basis for nuclear PPAs even when wind/solar PPAs are cheaper headline. Full write-up: /knowledge/concepts/capacity-factor/.

What I changed in the thesis book

Three new theses were added since the May 18 brief:

• T008 — AI agents reorganize enterprise software economics. You called this out as the biggest theme in the world right now, and the original 7 theses didn't touch it. The structural call: AI agents reorganize software around either the ontology (Palantir) or the orchestration plane (Microsoft Copilot). Proposed adds: PLTR (3-4%) and MSFT (3-4%), scaled in over the next 2 quarters as Q2 earnings clarify Copilot ARR.

• T009 — Cybersecurity reorganizes around agent identity. Gartner says 40% of enterprise apps will have AI agents by end-2026, up from <5% in 2025 — that's the new attack surface. CrowdStrike documented the first AI-agent-driven attacks. Palo Alto closed the $25B CyberArk acquisition in February. Proposed adds: PANW (3-4%), CRWD (3%).

• T010 — Fed transition + 1-2 cuts in 2H 2026. Macro overlay, not a new sector bet. Held to make the portfolio's implicit long-duration / pro-growth tilt explicit. If the inflation/growth picture deteriorates, T010 is the thesis that flips first and triggers a defensive sleeve discussion.

I have NOT auto-added the T008/T009 positions to the simulated portfolio. Per AGENTS.md rule 6, position changes are user-confirmed. You're effectively the IC. If you want me to add them, say so and I'll record the change with reasoning. Otherwise they sit on the watchlist with full research files at /tickers/PLTR/, /tickers/MSFT/, /tickers/PANW/, /tickers/CRWD/.

What I'd do today if I were you

1. Don't add to CEG above $260. You've got the right exposure already.
2. Read T008 AI agents end-to-end. It's the single biggest gap in v0 and I think the conviction is correctly high.
3. Decide on T008/T009 position adds. If you authorize them: ~$10-12K total deployment from cash, scaled in over Q2 earnings season (July-August). That brings deployed from 65% → 76% with cash reserve still healthy at 24%.
4. Skip T010 as a position thesis. It's a frame, not a trade.

Watchlist for the week

• No major holdings have earnings this week — the next big print is CEG Q2 on July 30 (you have 70 days to watch the run-up).
• PJM Base Residual Auction results expected Q3 — directly read-through to T001. If they clear near 2025 highs again, T001 conviction goes up; if they collapse, falsifier #4 hits.
• PLTR Q2 earnings in August — most important T008 catalyst; would size up the position on a confirming print.

Method note

A few process choices the agent made today (full audit at any later date in data/changes/):

• Wrote AGENTS.md as the universal agent guide; CLAUDE.md now points to it.
• Pulled the CEG 10-K and 10-Q from SEC EDGAR directly using the new scripts/fetch/edgar.py tool. Quotes and metrics in the CEG ticker file are sourced from those filings with primary_source_read: true.
• Built the knowledge base (data/knowledge/) and seeded it with 5 concept notes including capacity factor, PJM capacity market, and candlestick charts. These compound — the next utility ticker file references the same notes.
• Migrated the site to a multi-page architecture with deep-linkable theses, tickers, briefs, knowledge pages.
• Wired espeak-ng + ffmpeg as the audio TTS path. Voice is robotic but functional. Upgradeable to Cloudflare Workers AI / ElevenLabs by passing --provider=cloudflare or wiring the API key.

— Thesis Lab v0.2.0