It Pays to be a Winner: Our ML model and social research layer just went 14-for-16 on Nike’s Q3 call. (Kalshi, Mentions)

Post trade results, and breakdown of our new model's performance.

First Strike Research is not registered as a Commodity Trading Advisor (CTA) with the NFA or CFTC. Past performance figures cited in this article, including win rates, directional accuracy percentages, and P&L results, are historical and are not necessarily indicative of future results. Hypothetical or simulated performance results have inherent limitations and do not represent actual trading by any specific subscriber. Individual results will vary. Subscriber testimonials reflect individual experiences and are not a guarantee of future performance. Prediction market trading involves substantial risk of loss, including the total loss of invested capital.

Our newest model just went 8-for-10 on traded positions across Nike’s Q3 FY2026 earnings call Kalshi keyword contracts, that’s an 80% win rate on traded positions, and 88% directional accuracy across all 16 contracts analyzed

And the part we’re most proud of?

Our two highest-conviction trades both hit. If you would’ve traded this alone you would’ve netted a 200% ROI.

These were positions where our ML probability model and our social research overlay converged and were in full agreement with the highest possible conviction score.

But don’t just take my word for it, our paid subscribers took down some handsome profit—

---

The Three-Layer System Is Getting Sharper Every Cycle

We’ve now run our new architecture through two complete earnings cycles — CrowdStrike and Nike — and the trajectory tells a clear story.

CrowdStrike: 11/13 directional (85%)

Nike: 14/16 directional (88%), +208¢ net P&L

That’s not a fluke. That’s a system learning and improving in real time.

Layer 1 — The ML Model generates raw probability estimates based on historical transcript data and linguistic patterns. It sets the baseline. On Nike, the raw model alone tracked around 10 out of 16 — decent, but not where we want to be. Its strength is consistency: it processes every keyword through the same statistical lens and catches patterns that gut instinct misses.

Layer 2 — The Context Classifier overlays narrative lifecycle analysis. Is a term rising in management’s vocabulary or fading out? This layer caught that Consumer Demand and Amazon were losing narrative momentum even though the model’s base rates still liked them. It correctly flagged Express Lane and Nike Factory as terms drifting out of active rotation. Layer 2 brought us from roughly 10 to 12 out of 16.

Layer 3 — The Social Research Overlay is where the edge lives. Deep research into transcript history, brand strategy signals, analyst commentary, media coverage, and social sentiment catches the event-driven shifts that data alone can’t anticipate. One of our biggest social research win was with Starbuck’s Condiment Bar.

Three critical interventions on Nike show why this layer changes everything:

The Tariff flip saved 97¢. Our model had this as a BUY NO at 63¢. The market had it at 97¢. Our research identified that in the current macro environment, Nike management was almost certainly going to address tariffs head-on. We flipped the position. That single research-driven call was worth nearly a dollar per contract.

The Retailer flip earned 67¢. Research showed wholesale distribution dynamics had become a hot topic in the analyst community — something the model’s historical data underweighted. Our social and media overlay caught the shift.

The Dividend risk flag limited a loss. We identified the April 1 pay date and Nike’s stock sitting near a five-year low as catalysts that could force the word into Q&A. We sized down accordingly. The loss came in at -43¢ instead of what could have been significantly worse.

---

A Tale of Two Misses

We’re celebrating the wins, but it’s important to remember that no model, formula, or system will always be 100% accurate.

Holiday (-81¢) was our biggest miss - Nike’s Q2 call was the holiday quarter — management had already spent extensive airtime on seasonal performance. By Q3, they’d moved on. The model liked Holiday based on historical Q3 base rates, but the word was temporally anchored to the prior quarter. In the future, we’ll more likely than not advise NO ACTION or ‘NO’ contracts.

Dividend (-43¢) hurts less because our research actually identified the risk. We sized down instead of flipping to a full PASS or even YES. Going with the gut here, YES contracts would’ve been more appropriate given NKE 0.00%↑ existential share price retreat.

---

The Scoreboard Speaks for Itself

Two earnings cycles with the new model are now in the books, and here’s what the data says about this system:

Improving accuracy — 85% to 88%. Positive and growing P&L. Highest-conviction picks landing at a perfect clip. And every miss traceable to a fixable process gap rather than a fundamental flaw in the methodology.

The ML model provides the statistical backbone. The context classifier reads the narrative landscape. The social research overlay captures the real-world signals that move markets. When all three layers agree, we haven’t been wrong yet.