Back to Blog
MethodologyPlay Portfolio

Inside the Play Portfolio: Committee Discourse, Quarter-Kelly Sizing, and Systematic Discipline

February 24, 2026 · 10 min read

We publish multi-lens analyses, prediction ensembles, and thesis assessments for every equity we cover. But analysis without action is incomplete. The play portfolio exists to answer a natural follow-up question: if a systematic, transparent process took these assessments seriously, what would it actually do?

The goal is not to demonstrate returns. It is to demonstrate that a disciplined framework can bridge the gap from "here is our assessment" to "here is what the framework would do with it" — with every decision fully auditable, every sizing formula visible, and every rejection explained. No real capital is at risk. This is educational infrastructure, not a track record.

Explore the live portfolio

How Positions Enter

The portfolio is long-only. No shorting, no derivatives, no leverage. Only tickers that carry a thesis classification of price-below-value at MEDIUM or HIGH confidence are eligible for evaluation. Everything else — price-at-value, price-above-value, or low confidence — is automatically excluded.

This is not a screening filter that can be overridden. The committee cannot approve a position that the thesis assessment does not support. The analytical pipeline produces the classification; the portfolio framework acts on it. See how we assess price vs. value for how those classifications are computed.

The 4-Step Committee Discourse

Every eligible ticker goes through a structured discourse between four AI-driven committee members. Each step is recorded as a separate JSON file — proposal, risk assessment, challenge, and verdict — and is readable on every position page.

1

Portfolio Analyst — Trade Proposal

Reviews the thesis assessment, prediction markets, and cross-lens signals. Proposes the trade with a specific rationale: which markets have the strongest model agreement, which signals support the classification, and what the implied edge looks like. The proposal also identifies the key risks the committee should weigh.

2

Risk Manager — Kelly Sizing and Constraints

Computes the position size using the quarter-Kelly formula (described below). Checks all portfolio constraints: 10% max single position, 30% max sector weight, 5% cash floor, 20 max positions, and 2% minimum position size. If the computed weight falls below 2%, the position is mechanically rejected — insufficient edge.

3

Devil's Advocate — Thesis Challenge

Stress-tests the proposal. Raises the strongest counterarguments: thesis fragility, catalysts that could invert the classification, tail risks the risk manager may have underweighted. Assesses whether the thesis is robust, fragile, or mixed — and explicitly states what would make the position a mistake.

4

Committee Chair — Final Assessment

Weighs all three inputs: the analyst's case, the risk manager's sizing, and the devil's advocate's challenges. Renders a final decision — approve, reject, or defer — with explicit reasoning for how each challenge was addressed or accepted. Includes monitoring triggers that would warrant re-evaluation.

Full Transparency
Every trade decision is fully auditable. You can read the exact reasoning from each committee member — the proposal, the sizing math, the challenges raised, and the final assessment — on any position page.

Quarter-Kelly Sizing

The Kelly criterion computes the theoretically optimal bet size given an estimated edge and odds. In practice, full Kelly is far too aggressive — it assumes perfect edge estimation, known distributions, and no model uncertainty. We use quarter-Kelly (0.25x) to account for all three.

EdgeclassificationEdge × confidenceMultiplier × dataQualityMultiplier

Classification edge comes from the thesis (0.20 for price-below-value). Confidence scales by thesis confidence level. Data quality reflects model agreement and information gain across active prediction markets.

OddsmagnitudeOdds + tailRiskDiscount × directionAdjustment

Magnitude odds come from the implied price dislocation. Tail risk discounts penalize known asymmetric risks (e.g., customer concentration, regulatory exposure). Direction adjusts for mixed-signal scenarios.

Raw Kellyedge / odds

The theoretical optimal fraction of capital.

Position SizerawKelly × 0.25

Quarter-Kelly. The conservative multiplier that makes this viable in the real world.

Worked Example: DDOG at 3.5%

Datadog entered the portfolio with a thesis classification of price-below-value at HIGH confidence, backed by 8 active prediction markets with 0.88–0.99 model agreement. Here is how the formula computed the 3.5% position:

Classification Edge0.20
Confidence Multiplier (HIGH)1.0
Data Quality (avg 0.96 agreement, 0.80 info gain)0.896
Raw Edge0.1792
Magnitude Odds1.50
Tail Risk Discount-0.20
Direction Adjustment1.0
Adjusted Odds1.30
Raw Kelly (0.1792 / 1.30)13.78%
Quarter-Kelly (× 0.25)3.45%
Final Position Size (rounded)3.5%

The raw Kelly says 13.78% — aggressive enough that a single adverse quarter could cause meaningful drawdown. Quarter-Kelly brings that to 3.45%, well within the 10% single-position cap and above the 2% minimum. The tail risk discounts penalize OpenAI customer concentration and demanding valuation at ~53x non-GAAP P/E.

When the Formula Says No

The 2% minimum position size acts as an edge filter. If quarter-Kelly computes a weight below 2% of capital, the position is rejected — the formula has determined that the edge is not large enough to justify even a minimal allocation.

NFLX — Rejected at 1.95%

Netflix carried a price-below-value classification at MEDIUM confidence. The thesis centered on a deal-failure-is-bullish case — an interpretive inversion where a government override of management strategy would unlock value. The Kelly formula computed 1.95%, just below the 2.0% threshold. The devil's advocate identified three structural concerns: the thesis depends on analytical inversion, the payoff distribution is bimodal (violating Kelly assumptions), and there is a 30% probability of fundamental thesis inversion if the deal proceeds. The formula correctly identified insufficient edge. Not overridden — mechanical discipline matters.

Netflix is not alone. Several tickers — DOCU, OKTA, XYZ — passed the thesis filter but computed Kelly weights between 1.8% and 1.96%, just below the floor. They appear on the screened tickers page as "deferred" — waiting for catalysts that could increase confidence, improve data quality, or shift the odds enough to clear the threshold.

When the Thesis Changes

The portfolio framework does not just manage entries — it monitors and acts on thesis reclassifications. The Moderna position illustrates this.

MRNA — Same-Day Open and Close

Moderna was opened at $51.09 based on a thesis written when the stock traded at $37.74 — a price-below-value classification at MEDIUM confidence grounded in the proximity of the market cap to MRNA's $8.1B cash position. The thesis assessed the pipeline as genuine optionality on top of a near-cash floor.

After the position was opened, a thesis update triggered by new earnings data reclassified MRNA from price-below-value to price-at-value. At $51.09, the market cap implied ~$19.7B — no longer near the cash floor. The market had already moved to price in meaningful pipeline value.

Under price-at-value, the classification edge is zero. Zero edge means zero Kelly weight. The committee approved immediate closure — the same day the position was opened, at the same price, with $0 P&L.

Process Learning
The MRNA case exposed a sequencing gap: thesis prices should be refreshed before entry, not after. The prediction probabilities were unchanged — the same analytical signals that justified opening the position remained valid. Only the price had moved. The committee acknowledged this as a genuine process improvement and recommended checking for thesis staleness before trade execution in future iterations. The correct action given the current state was still to close — holding a zero-edge position to avoid the optics of a same-day round-trip would introduce anchoring bias.

What This Is Not

Not financial advice. No real capital is at risk. No one should replicate these positions. This is educational content demonstrating a systematic process.

Not a track record. Performance is displayed for transparency, not for evaluation. The portfolio has been live for days, not years. Drawing conclusions from the P&L would be statistically meaningless.

Not optimized for returns. Quarter-Kelly is deliberately conservative. The 2% minimum rejects marginal opportunities. The long-only constraint ignores half the thesis assessments. The portfolio is optimized for process integrity, not returns.

What it is: a demonstration that systematic, transparent process can convert thesis assessments into disciplined positions — and that the same process knows when to say no, when to exit, and when to acknowledge its own mistakes.

This report was generated by the Runchey Research AI Ensemble using primary SEC data and reviewed by Matthew Runchey for accuracy.

This analysis is for educational purposes only and does not constitute investment advice. See our Editorial Integrity & Disclosure Policy and Terms of Service.