Case study: a narrow tape, a Green print, and why the book stayed ORANGE

Week of May 11–15, 2026 — archived model-portfolio commentary

This is an archived entry from the Clarion model portfolio — a real-money book run in public as an educational record. It documents how the system processed one specific week: what the regime framework said, how each thesis absorbed the week's news, and where the process held or showed gaps. It is published for what it teaches about the method, not as guidance on what to buy or sell. Prices and facts are as of May 15, 2026 and have not been updated since.

The week is a useful case study because it contained a genuine tension: the daily classifier printed a risk-on signal while the framework's operating posture stayed cautious. How a system resolves that kind of disagreement — and whether it resolves it by rule rather than by mood — is most of what separates a process from a feed of opinions.

The regime tension, and how the framework resolved it

The cross-asset signal printed Green on May 14, but the May 16 living letter still framed the operating environment as ORANGE and applied a 9.69% equity hurdle. The book followed the letter.

The reason is breadth. Through the week, SPY rose +2.27% while the equal-weight RSP rose +0.06%. NVDA alone was +11.46%, pulling the cap-weighted index higher while the average stock barely moved. The Green-day classifier responds to price; the breadth divergence the letter flagged did not go away.

The lesson: a one-day signal and an operating regime are different instruments with different horizons. When they disagree, the framework defers to the slower, breadth-aware read rather than re-rating the whole book on a single print. The tape improved, but broad participation did not — so ideas that depended on cyclicals, small caps, or value re-rates got no help from market internals, and the hurdle rate stayed high.

Bucket map

Bucket weights are policy targets per the Allocation Framework. Position-level weights are not disclosed. The book was heavily cash- and T-bill-weighted by design during this period; equity deployment was a small share of total assets.

How each thesis absorbed the week

IREN — Systematic — thesis validated, discipline tested

On May 7, IREN announced a $3.4B AI cloud services contract with NVIDIA, paired with a $2.1B equity stake. NVIDIA received five-year warrants for up to 30M shares at a $70 strike. The partnership committed up to 5 GW of NVIDIA DSX-aligned infrastructure across IREN's pipeline, including the 2 GW Sweetwater campus. The stock spiked +20% post-market and ran +40% over five days before settling.

The same Q3 release showed revenue and EBITDA missing estimates by ~33% (rev $144.8M vs $216.6M est, adj EBITDA $59.5M vs $125M est), reflecting the deliberate wind-down of Bitcoin mining as AI cloud took over. The market priced past the miss entirely.

What this episode teaches. Validation is the moment a process is most tempted to abandon itself. The thesis catalyst ("wait for the Sweetwater ramp") arrived early and bigger than written — a customer concentrating its own capex through the company's pipeline. The disciplined response was to update the thesis milestone (from "Sweetwater ramp" to "first $1B of contracted GPU revenue shipping"), not to add into a +40% spike. Chasing a validated thesis at post-catalyst prices converts an edge in analysis into a penalty in entry.

TTD — Value — the insider-buy value-trap question

Q1 reported May 7 and was the exact catalyst the thesis named — pointing the wrong way. Revenue $689M (+12% YoY) beat consensus, but EPS came in at $0.28 vs $0.32 expected, and the Q2 guide ($750M vs $772M consensus, ~8% YoY) marked the slowest growth since the 2020 ad downturn. Adjusted EBITDA margins contracted from 34% to 30%. The stock lost 15% in a session.

Management framed the miss as Amazon DSP competitive pressure plus vertical softness, and leaned on the AI-agent story (Kokai, agentic media buying). CEO Jeff Green's $148M insider buy at $24.71 — the anchor of the original bull case — sat ~17% above the May 15 close.

What this episode teaches. This is the canonical stress test of an insider-buy thesis: the highest-conviction signal in the value playbook can still be early if a growth reset is structural rather than cyclical. The framework's job is to separate cycle events (hold through) from thesis events (rewrite or exit), and to name in advance what evidence would distinguish them — here: agentic-AI traction in the next two prints, the Publicis dispute, and Amazon DSP share trajectory. A falling price alone is neither.

BBY — Value — when there is nothing to do

The 2026 annual report flagged six net new US stores in FY27 — the first domestic store-count expansion in more than a decade — and guided Marketplace and Best Buy Ads toward profitability. These are management-confidence signals, not earnings signals. The Goldman Sachs double-downgrade that created the entry opportunity had not been reversed, and the Q1 FY27 print sat ~3 weeks out.

What this episode teaches. Most weeks in a position's life contain no decision. Logging that explicitly — "the print is the test, and it isn't here yet" — is part of the process, because it prevents low-information weeks from generating high-conviction actions.

DECK — Value — sentiment pressure vs. thesis evidence

The most pressured name of the week. Wells Fargo downgraded to Underweight, PT $90 from $115; UBS cut its target to $145 but maintained Buy; Bernstein issued a partially offsetting upgrade citing HOKA + UGG resilience. Q4/FY26 earnings were scheduled for May 22, with the thesis reads sitting on that print: HOKA growth rate, Vietnam/Cambodia tariff impact on the FY27 guide, UGG seasonality.

What this episode teaches. Analyst actions move price but are not primary evidence — three firms looked at the same company that week and produced three different answers. The framework treats the upcoming filing and guide as the evidence, and pre-commits to the specific line items that will be read. Deciding before the print what matters in the print is the cheapest protection against narrating whatever number arrives.

IBM — Value — divergence between product news and price

IBM's Think 2026 conference produced the most thesis-relevant product slate of the year: Watsonx Orchestrate as an agentic control plane (private preview), IBM Sovereign Core generally available, and Red Hat AI Inference announced May 12 with GA set for May 22 — IBM's first managed AI inference offering. An IBM-published survey reported 76% of organizations have appointed a Chief AI Officer, up from 26% in 2025: directly addressable consulting demand.

The stock was down ~12% since the April 22 Q1 print despite that print beating on both lines ($1.91 EPS vs $1.81 est; $15.92B revenue, +9.5% YoY).

What this episode teaches. Price and thesis evidence can diverge for weeks at a time. The named monitor in the thesis — Watsonx ARR growth — got a tangible catalyst (Red Hat AI Inference GA); the price went the other way on sentiment. A written thesis with named monitors is what lets a process call this "constructive week, weaker price" instead of letting the drawdown rewrite the story.

HOOD — Educated YOLO — reinforcement without re-rating

Three developments aligned with the platform-broadening narrative: the Senate Banking Committee advanced the CLARITY Act (crypto-revenue durability moving from "policy risk" toward "policy clarity in motion"); President Trump disclosed buying HOOD in an ethics filing (visibility, not thesis); and HOOD filed confidentially for a second venture fund (consistent with "compounding platform, not just a brokerage").

What this episode teaches. The bucket label is doing real work here. "Educated YOLO" is the framework's honest admission that this position carries more narrative risk than a Value thesis — so a week of favorable headlines reinforces direction but does not promote the position into a higher-conviction bucket or justify new capital. Buckets cap enthusiasm structurally, so the operator doesn't have to do it emotionally in real time.

SOXX — Discretionary Short — a hedge with a named tension

The May 14 letter update opened a short on SOXX — a concentrated, high-beta semiconductor ETF where NVDA, AVGO, AMD, and QCOM combined made up roughly 35% of the fund. The stated rationale was concentration risk: an AI-semis basket priced for perfection in a tape where a handful of names were doing all the work.

The position carried a structural tension worth naming in public: it was sized under the letter's ORANGE framing, and the daily engine printed Green the same day it opened.

What this episode teaches. A basket short used as a hedge is a different instrument from a bearish single-name thesis, and labeling it honestly matters — this was an expression of the same narrow-leadership risk flagged in the regime section, not a claim that every semiconductor company was overvalued. Documenting the regime tension at entry, rather than after the fact, is what keeps a public track record honest.

DRAM — Discretionary Short — a position ahead of its paperwork

The May 16 letter added DRAM as a second discretionary short — a memory-cycle name facing structural headwinds — while stating plainly that thesis scaffolding was still in progress.

What this episode teaches. This entry is included precisely because it shows the process incomplete. A disciplined framework requires a written thesis with a target, kill condition, and maximum-pain threshold; this position existed before its paperwork did. Publishing that gap, and naming the required next step, is the difference between a system that audits itself and one that only narrates its wins.

The watchlist as a discipline, not a teaser

The letter's watchlist — META, MSFT, GOOGL, FFIV, NVDA, KO, PG — is tracked weekly against pre-computed fair values. That week, none converted: MSFT and GOOGL finished flat to modestly higher with no thesis-altering news; META sat above fair value; FFIV ran further away (+5.7%); NVDA added ~11% and widened its gap to any defensible entry; KO and PG remained above fair value.

The lesson: the brief Green print did not, by itself, deliver a single cheap name. A watchlist with written fair values converts "the market is up, should I buy something?" into a checklist question with a usually-boring answer — which is the point.

What the framework was set to read next

The forward calendar, as written that week — preserved here because pre-committing to what matters is part of the method:

• DECK Q4 FY26 earnings (May 22): FY27 tariff guide as the swing variable.
• Red Hat AI Inference GA (May 22): first productized test of the IBM thesis monitor.
• BBY Q1 FY27 (~2 weeks out): same-store-sales commentary as the AI-PC-cycle test.
• NVDA earnings: the regime-level driver — a miss unwinds the breadth concentration; a beat extends it. The same print frames the SOXX hedge.
• SOXX and DRAM thesis pages: both shorts open, both still owing formal kill conditions — the system's own named debt.

Methodology and disclosures

This commentary was generated from the living investor letter as source of truth — every position, watchlist name, and bucket weight was read from the most recent letter update at publication time — layered with May 15 closing prices, the week's news, primary-source SEC filings via the system's PageIndex RAG indexer, and the daily SPY/TLT regime classifier. Each read is tagged for what the news means relative to the thesis as written, not as a price prediction.

The Clarion model portfolio is a real-money book disclosed for educational purposes: the point is to show the framework operating on live conditions, including its mistakes and open gaps. Bucket weights are policy targets; position-level weights and execution detail (entries, stops, limits) are not disclosed. Nothing here is investment advice, and this archived entry is not updated as conditions change.

Sources

• Living investor letter — Clarion Intelligence System 2026 Investor Letter
• IREN / NVIDIA partnership — CNBC; Sherwood News
• TTD Q1 print and guide — Yahoo Finance; MediaPost; Adweek
• BBY store expansion — Retail TouchPoints
• DECK analyst actions — TipRanks; Deckers IR
• IBM Think 2026 and Red Hat AI Inference — Yahoo Finance; Stocktitan; CNBC
• HOOD CLARITY Act, Trump trade, venture fund — Forbes; Quiver Quant; Axios Pro