Intelligence Synthesis: April 13, 2026

Generated: 2026-04-13 | Baseline: 2026-03-25 | Next Checkpoint: 2026-04-30 CNN Fear & Greed: 39.9 (FEAR) | VIX: 19.38 | 10Y: 4.30-4.39% | Brent: ~$96 | WTI: ~$95.50 ML Model: 245 predictions tracked, 94 resolved | 1d Directional Accuracy: 80.1% | Calibration ECE: 0.309 Regime: VIX normal | Correlation EXTREME HERDING (2.6σ) | Composite: MIXED Gold: ~$4,760/oz | Copper: $5.81/lb | DXY: 98.74 | BTC: ~$71,000

BREAKING: Iran-US peace talks collapsed April 12. Trump announced US blockade of Iranian ports. This supersedes the April 8 ceasefire. See Section 0.

0. GLOBAL NARRATIVE

The two-week ceasefire that markets celebrated on April 8 lasted four days. On April 12, after 21 hours of marathon negotiations in Islamabad, VP Vance declared "they have chosen not to accept our terms" — Iran refused to halt uranium enrichment, dismantle facilities, or fully open Hormuz without tolls. Hours later, Trump announced the United States would blockade Iranian ports. This is not a return to the pre-ceasefire status quo; it is an escalation. The conflict has entered Day 44 with the world's largest navy now actively enforcing a commercial blockade on a nation that still controls physical access to 20% of the global oil supply. Friday's risk-on rally — S&P +3.9% for the week, VIX collapsing to 19.38, core short targets surging 6.6% in a single session — was predicated on a ceasefire that no longer exists. Monday's open will reprice this.

The supply chain consequences are now compounding in ways that the market has not fully absorbed. The Strait of Hormuz has been effectively closed since February 28, with fewer than 10 vessels transiting daily versus 100+ pre-conflict. Four hundred loaded tankers sit queued at the Gulf entrance; 100 empty tankers wait to enter. Qatar's Ras Laffan — 30% of global semiconductor-grade helium and 20% of global LNG — remains offline. The helium constraint is no longer theoretical: Asian fab buffers expired in early April, and South Korea imported 64.7% of its helium from Qatar. European gas prices (TTF) surged to EUR 47/MWh, up 35% YoY, with Goldman projecting EUR 55 and warning of EUR 155 if the disruption extends three months. Meanwhile, copper hit $5.81/lb on a 62σ volume spike — the anomaly detector flagged copper, silver (262σ), palladium (254σ), aluminum (61σ), and gold (6.6σ) as simultaneous volume anomalies. Physical commodity markets are signaling something the equity market's ceasefire-relief rally has not yet processed.

The AI disruption thesis continues accelerating through the geopolitical chaos, creating a bizarre temporal dislocation: the fastest-growing sector of the global economy (AI) is being physically constrained by the oldest bottleneck in geopolitics (a strait that Persia has controlled for millennia). Anthropic crossed $30B ARR (surpassing OpenAI's $25B), with 1,000+ enterprise customers at $1M+/year. Cursor doubled to $2B ARR in three months, now in talks for a $60B valuation. SAP's CEO explicitly declared per-seat pricing dead — "it would be foolish to still charge subscription base." Workday's CTO defected to Anthropic for a staff role, then Anthropic pushed into HR applications. Oracle cut 30,000 workers. The 91,739 tech layoffs in 2026 (47.5% AI-attributed) are cancelling the per-seat licenses those workers occupied. Yet on Friday, WDAY rallied 6.5%, PATH +7.2%, ASAN +5.0% — a mechanical short squeeze driven by regime herding (correlation regime at 2.6σ extreme), not fundamental improvement. The ML model confirms: UP predictions hit at only 16%, while DOWN predictions hit at 57.1%. The rally is noise; the thesis is signal.

What changed since the April 11 report: (1) Iran peace talks collapsed after 21 hours (April 12), replacing diplomatic hope with a US naval blockade; (2) The April 11 trading session was a broad risk-on squeeze — 61 anomalies detected, 49 extreme moves to the upside, with the "extreme herding" correlation regime ensuring everything moved together; (3) Metals markets flashed massive accumulation signals — gold volume surged 6.6σ while price was flat (classic accumulation/distribution pattern), silver and palladium volume hit 260σ; (4) The ML anomaly detector flagged ORCL +11.78% as the day's most significant single-stock move, likely driven by tariff exemption news for tech hardware.

1. THESIS SCORECARD

Thesis tier refinement (updated): - Tier 1 (high conviction): NVDA, MU, TSM, VRT, CEG, VST — physical infrastructure with structural demand. Add RTX, LMT (defense spending escalation). - Tier 2 (moderate): CRWD, PANW — cybersecurity as AI beneficiary, not victim. DDOG at half-size. - Tier 3 (exit/avoid): SNOW, PLTR, NET — thesis broken or severely impaired per deep dive analysis.

2. PORTFOLIO POSITIONS: BASELINE vs CURRENT

Short Targets — Application SaaS

Long Positions — Infrastructure Software (UNDER STRESS)

Long Positions — Hardware & Power

Long Positions — Cybersecurity

Macro Indicators

3. KEY DEVELOPMENTS (past 48 hours)

4. THESIS DEEP DIVES

4a. AI Disruption of SaaS (Short Thesis) — 87% Confidence, UP from 85%

Confirmation evidence: The cumulative weight of evidence has crossed a threshold that moves this from "thesis" to "consensus forming in real time." SAP CEO Christian Klein declared per-seat pricing dead: "it would be foolish to still charge subscription base, because AI is so powerful that it will automate a lot of tasks." This is the largest ERP vendor on earth explicitly validating the short thesis. Workday's CTO Peter Bailis defected to Anthropic after less than a year, taking a "member of technical staff" role — not a C-suite position — to work on reinforcement learning. Anthropic is simultaneously pushing into HR applications, directly competing with Workday's core. This is a talent-signal vote with feet. The 91,739 tech layoffs in 2026 (47.5% explicitly AI-attributed, 35% from Cloud/SaaS companies specifically) continue cancelling per-seat licenses. Oracle's 30,000-person cut ($2.1B restructuring charge) frees $8-10B for AI data center buildout — the capital is literally being redirected from software seats to hardware. Pure per-seat pricing adoption dropped from 21% to 15% in the past 12 months. Usage-based pricing adoption rose to 38% (from 27% in 2023). Gartner expects 40% of enterprise SaaS to shift to outcome-based components by year-end. Anthropic crossed $30B ARR (80% enterprise, 1,000+ customers at $1M+/year). Cursor hit $2B ARR growing at ~2x per quarter.

Challenge evidence: Friday's 6.6% rally across core short targets demonstrates the mechanical risk in this trade. The correlation regime is at "extreme herding" (2.6σ), meaning everything moves together regardless of fundamentals. Short squeezes in oversold names are inevitable and violent. WDAY still has 97% gross revenue retention and $25.37B subscription backlog. The 88% AI agent production failure rate remains real. The Klarna reversal remains the canonical cautionary tale. At current valuations (WDAY ~3.4x revenue, PATH ~2.5x revenue), PE buyout floors provide downside protection. The ceasefire collapse redirects capital from growth plays to energy/defense, which paradoxically helps short targets by reducing selling pressure (no one is paying attention to software when oil is the story).

Adversarial probability: 16% chance the short thesis is fundamentally wrong (down from 18%). The CTO defection and SAP CEO statement reduce the "timeline uncertainty" component. Breakdown: 3% AI timeline is 5+ years (contradicted by WDAY's own CTO leaving), 8% M&A takeout removes targets (tariff/macro suppresses deal activity but WDAY at 3.4x approaches PE floor), 5% technical rally extends 30%+ from oversold levels.

What would change confidence: A major enterprise publicly reporting AI-driven seat INCREASE. Anthropic Managed Agents at-scale production failure. M&A bid for WDAY above $130. PATH successfully establishing "agentic orchestration" as a sticky product category.

4b. Hardware & Power Infrastructure (Long Thesis) — 80% Confidence, DOWN from 83%

Confirmation evidence: TSMC Q1 revenue $35.7B (+35% YoY) at the top of guidance, with gross margin guided 63-65% (highest ever). Full earnings call April 16 — consensus expects EPS ~$3.29/ADR (+50% YoY). CoWoS scaling from 75K to 130K wafers/month, fully booked. NVIDIA Blackwell backlog at 3.6M units; B300 DGX now shipping at $300-350K per system. Hyperscaler capex hits $630B in 2026 (+62% from 2025): Amazon $200B, Google $175-185B, Meta $115-135B, Microsoft $110-120B. Micron announced HBM4 mass shipping (2.3x bandwidth over HBM3E, 20% power efficiency gain) with 2026 fully sold out under binding contracts. VRT backlog at $15B with 2.9x book-to-bill ratio — nearly three years of locked revenue. SOX index hit all-time high 8,928 on April 10.

Challenge evidence — and the reason confidence dropped 3pp: The helium constraint has shifted from "emerging risk" to "binding constraint." Qatar's Ras Laffan has been offline since March 2; Asian fab buffers expired in early April; South Korea imported 64.7% of its helium from Qatar. Spot helium prices surged 40-100%. If the US blockade extends the Hormuz closure, fab production slowdowns will manifest in Q2 earnings across the entire semiconductor stack. This is the single biggest new risk to the hardware long. Additionally: CEG's TMI restart faces transmission delays (FERC waiver filed but uncertain), CEG's 2026 EPS guidance of $11-12 missed consensus $12.11, and the NVIDIA H200 China export impasse has $54B frozen at customs with the AI OVERWATCH Act threatening a two-year Blackwell export ban. Amazon is projected to turn FCF negative in 2026 on $200B capex — unsustainable if AI ROI disappoints.

Adversarial probability: 25% (up from 22%). Helium supply disruption impact on fabs (8%), hyperscaler capex cut if ROI disappoints (7%), semiconductor cycle peak at extreme valuations — SOX P/B 8.2x vs 8.8x dot-com (5%), China restrictions permanently impairing NVDA revenue (3%), recession forcing indiscriminate selling (2%).

What would change confidence: Any hyperscaler announcing capex reduction. TSMC Q2 guidance below consensus. Helium emergency airlift or alternative supply route activated. Hormuz full reopening within 2 weeks.

4c. Cybersecurity (Long Thesis) — 68% Confidence, UP from 65%

Confirmation evidence: Project Glasswing and Claude Mythos represent a permanent escalation of the threat landscape. Mythos found thousands of zero-day vulnerabilities across every major OS and browser, including CVE-2026-4747 (17-year FreeBSD RCE, hidden 17 years), a 27-year-old OpenBSD crash bug, and a 16-year-old FFmpeg flaw. It autonomously chained four vulnerabilities into a working browser exploit. If AI can find bugs humans missed for decades, every enterprise must assume their attack surface has expanded dramatically. CrowdStrike launched Charlotte AI AgentWorks with Anthropic, OpenAI, NVIDIA as partners. PANW completed the $25B CyberArk acquisition — the largest security deal ever — adding identity security to its platform. Check Point launched AI Defense Plane. Cybersecurity spending forecast at $240-244B (+12.5%) with cloud security at 28.8% growth. The EU AI Act August 2 enforcement cliff forces $8-15M initial compliance spend per large enterprise. AI-driven attacks surged 72% in 2026; 87% of organizations targeted; 1,968 attacks per week average.

Challenge evidence — and why confidence isn't higher: Microsoft's Security Copilot bundling represents the most significant competitive threat since the Defender launch. Six autonomous security agents shipped into production at RSA 2026, bundled with E5/E7 at no extra cost. The 6.5x alert detection improvement is a compelling data point for enterprises already on Microsoft licensing. CRWD at 81x forward P/E is priced for perfection in a macro environment that punishes expensive stocks. ZS has declined 45% in 90 days despite raising guidance — the market does not currently reward cybersecurity growth. HACK ETF is -6.7% YTD.

Key insight: The HACK ETF volume spike (3.0σ on April 11) signals growing institutional interest. The AI threat escalation is permanent — every AI agent deployed expands the attack surface. Cybersecurity is an AI beneficiary, not victim. The cheapest way to play this is CHKP at 14x P/E or PANW at ~55x (vs CRWD 81x), not CRWD.

4d. Infrastructure Software (Long Thesis — IMPAIRED) — 40% Confidence, DOWN from 45%

The bifurcation within the bifurcation is now undeniable. The five names in this category occupy three entirely different competitive positions:

DDOG ($109.80): The Guggenheim flip — same analyst downgraded to Sell ($105 PT) in July 2025, then upgraded to Buy ($175 PT) in April 2026 — crystallizes the uncertainty. OpenAI migration is real ($150M+ revenue hole) but slower than feared. AI-native customers ex-OpenAI growing 220% YoY. At $110, trading at 48% off highs with consumption model aligned to AI workload growth. $57.4M in insider selling over 90 days (CEO, CTO, CFO, CPO, CRO all selling) is the heaviest selling per dollar of any tracked name. OpenTelemetry is commoditizing the instrumentation layer. Verdict: Hold at half size. The open question is whether OpenAI builds a general-purpose observability competitor, not just an internal tool.

NET ($181.52): The April 9 crash was tariff-driven, not company-specific. NET's AI Gateway + MCP server toolkit positions it as the infrastructure substrate for agentic AI — routing, caching, authentication, and billing for AI model calls from Anthropic, OpenAI, Google, and xAI. Major companies (Asana, Atlassian, Block, PayPal, Sentry, Stripe) are building Claude integrations via Cloudflare Workers. CEO Prince sold $64M in Q1 2026 (pre-scheduled but optically terrible). The strategic positioning is strong but monetization is unproven. Verdict: Monitoring only. If AI inference traffic grows as projected, NET has structural upside. Wait for earnings.

SNOW ($132.10): Databricks has won the growth battle on every metric: $5.4B ARR (+65% YoY) vs Snowflake $4.47B (+29%); AI revenue $1.4B vs ~$100M; NRR >140% vs declining. Apache Iceberg has neutralized Snowflake's data lock-in moat. Insiders sold $108.8M in 90 days; Slootman lifetime sales total $1.43B. Databricks IPO (likely H2 2026, $134B private valuation) will create a direct public comparable that highlights Snowflake's structural disadvantage. Verdict: EXIT. Thesis is broken. Do not catch falling knife.

PLTR ($131.73): At $312B market cap on $3.2B trailing revenue (97x P/S), the valuation requires perfection that the competitive landscape no longer supports. Burry's Anthropic thesis has structural merit — Anthropic now has its own "Applied AI Engineers" replicating the FDE model, and Palantir's own AI FDE platform has first-class support for Anthropic, OpenAI, Google, and xAI models, effectively conceding model-layer commoditization. Government contracts ($10B Army deal, Maven Program of Record) provide a floor, but the commercial thesis is severely impaired. Trump's Truth Social endorsement (first time a sitting president endorsed a stock by ticker) failed to reverse the weekly -15% decline. Verdict: EXIT. The government floor is real but the commercial premium driving 97x P/S is evaporating.

MDB ($229.49): Atlas revenue +29% growth with native vector search for RAG applications makes this the only genuine AI-demand beneficiary in the group. Consumption-model economics are positively correlated with AI workload growth. At -48% from 52-week high, the stock is approaching value territory. Verdict: Hold with moderate conviction. Best risk/reward in the infrastructure group.

4e. Energy Security & Geopolitics — 85% Confidence, UP from 78%

The ceasefire collapse changes everything. The April 12 failure is not a return to the pre-April 8 status quo — it is worse. The US has escalated from "negotiating while constraining" to "blockading while negotiating has failed." Key sticking points were non-negotiable on both sides: Iran refuses to halt enrichment; the US demands it. The Strait of Hormuz has 400+ tankers queued; mine-clearing operations are nascent (first sweeps began April 11-12); even optimistic scenarios require 4-8 weeks for safe passage plus 2-4 weeks for insurance reinstatement.

Oil: WTI ~$95.50 and Brent ~$96 as of Friday close will gap up Monday. The IEA estimated the conflict has removed ~11 million bpd from global supply. OPEC+ increased production by 206,000 bpd in April — symbolic when multiple members physically cannot ship via Hormuz. Saudi's East-West Pipeline runs at 7 million bpd capacity but Yanbu port can only load ~4.5 million bpd. The supply deficit is structural, not speculative.

Helium: Semiconductor-grade helium from Qatar (30% of global supply) remains offline since March 2. Spot prices surged 40-100%. Asian fabs are now operating on depleted buffers. No viable substitute exists for EUV lithography cooling. New helium projects (Canada, Tanzania) are years away. This is the sleeper risk that could force hardware thesis reassessment.

Defense: NATO committed to 5% of GDP by 2035; all 32 allies now exceed the prior 2% target. Germany pledging to double defense spending to EUR 162B. US FY2026 proposed at $1.01T (+13%). LMT +28.7% YTD with $194B backlog; RTX $268B backlog (sector largest). The Iran escalation feeds directly into defense spending acceleration.

Nuclear: CEG closed $16.4B Calpine acquisition, creating the nation's largest electricity producer (~55 GW). TMI restart faces transmission delays (FERC waiver filed). VST Meta nuclear PPA for 2,609 MW. Oklo targeting first criticality July 4, 2026. The Iran crisis has made energy independence a national security priority, accelerating nuclear timelines.

What ends the energy premium: Full mine clearance (4-8 weeks minimum), insurance reinstatement (2-4 more weeks), sustained safe transit (50+ ships/day for 2+ weeks). Even optimistic scenarios don't unwind the premium before June.

4f. Credit & Capital Flows — 30% Confidence (HOSTILE), DOWN from 35%

The ceasefire collapse makes stagflation the base case. Headline CPI already at 3.3% (gasoline +21.2% MoM); the US blockade will add another leg to energy prices. Core inflation (2.6%) remains contained, but consumer inflation expectations have de-anchored: UMich 1-year expectations surged from 3.8% to 4.8% in a single month — the largest one-month jump since April 2025. When consumers expect inflation, they demand wages, and the supply-shock bifurcation collapses into broad inflation.

The yield curve tells the story: The longest inversion in US history (27 months) has ended — but via a bear steepener, with 10Y yields pushing past 4.30%. This is the market pricing fiscal concerns and inflation persistence, not growth optimism. Atlanta Fed GDPNow tracking Q1 at 1.3%, well below the Fed's 2.4% annual projection.

Credit says "not yet": HY OAS at 294 bps is near cycle-tights. Either credit is right (no recession) and equities are oversold, or credit is lagging (spreads widen) and equities are early. Bank earnings this week (JPM April 14, GS, WFC) will resolve this: if loan loss provisions spike, credit follows.

Fed paralysis: Rate at 4.25-4.50%. April 29 FOMC is 94.8% hold. Powell term expires May 15. Warsh confirmation delayed again — earliest hearing April 21. Tillis blocking all Fed nominees. Institutional vacuum at the worst possible time.

SpaceX IPO: S-1 filed April 1, targeting $75B raise at $1.75T valuation. June 8 roadshow. 30% retail allocation ($22.5B from retail investors alone). This is the largest IPO in history by a wide margin and will vacuum capital from growth/venture allocations in June-July. Application-layer SaaS faces both AI disruption AND a liquidity headwind.

5. ADVERSARIAL CHALLENGE

Cross-thesis contradiction identified (updated): 1. Short thesis NEEDS cheap compute → Hardware long NEEDS scarce compute. If helium disrupts chip production, AI deployment slows, which paradoxically helps short targets (AI agents delayed) while hurting hardware longs. The portfolio has a helium correlation risk. 2. Energy long benefits from Hormuz staying closed → Short thesis benefits from Hormuz reopening (helium restores → chips flow → AI accelerates → seats get cancelled). The portfolio can't win on both simultaneously. 3. The ceasefire collapse helps energy longs but hurts everything else through higher inflation, tighter monetary conditions, and consumer spending compression.

6. INSIDER ACTIVITY

Pattern: 100% sells across every tracked ticker except CRM (one director purchase). Aggregate net insider selling: ~$185M in the trailing 30 days. CRM's single $500K purchase is the ONLY insider buy in the entire tracked universe — notable as a contrarian signal but immaterial in dollar terms vs the selling wave.

7. SUPPLY CHAIN & RESOURCE BOTTLENECKS

8. MACRO LANDSCAPE & SECTOR FLOWS

CNN Fear & Greed Components

Trajectory: Crashed from 64 (greed, mid-Feb) → 13.7 (extreme fear, April 7) → 39.9 (recovering on ceasefire hope). The ceasefire collapse will push this back toward extreme fear Monday. Safe haven demand at 93.4 (extreme greed) confirms gold/treasury flight.

Sector Rotation Rankings (Pipeline data, 20-day returns)

Key insight: Infrastructure software is the worst-performing sector over 5d and 20d, underperforming even core short targets. The market is telling us the "infra is safe harbor from SaaS disruption" narrative has failed for DDOG/NET/SNOW/PLTR. Only semiconductors, power, materials, and cloud infrastructure remain in genuine uptrends.

Rates & Fed

9. PREDICTION MODEL STATUS

Training Report (2026-04-12)

Directional Accuracy by Window (Cross-Validated)

Calibration Report (78 resolved predictions)

Direction Bias (Critical for Trade Use)

Source Comparison

Claude synthesis significantly outperforms the ML model. The ML model's primary value is in its anomaly detector and feature importance rankings, not directional predictions.

Top Features by Horizon

Key insight: Yield curve spread and short interest become top-5 features at 10-day horizons — macro and positioning factors increasingly dominate as the time horizon extends. The bear steepener is now the second-most-important signal for 10-day predictions.

10. RECOMMENDED TRADES

HIGH Confidence (>70%) — Core Positions

MEDIUM Confidence (50-70%) — Satellite Positions

LOW Confidence (<50%) — Tail Bets

Why These Trades

1. NVDA/WDAY Pair (R:R 2.0:1): Structurally hedged — AI acceleration benefits NVDA and hurts WDAY simultaneously. WDAY CTO defected to Anthropic; Duffield dumping $14M/week; Morningstar moat downgrade. NVDA: Blackwell 3.6M unit backlog, SOX ATH.

2. CEG Long (R:R 2.7:1): Best entry since thesis inception — down 34% from highs after guidance miss, but Calpine closed ($16.4B), 55 GW fleet, TMI restart on track for Microsoft. Nuclear is energy independence + AI power demand. The ceasefire collapse makes nuclear more valuable, not less.

3. RTX/LMT Long (R:R 2.1-2.4:1): NATO 5% GDP target; Germany doubling defense budget; US $1.01T proposal; Iran escalation feeds directly into missile/defense procurement. RTX $268B backlog is the largest in defense. These positions are policy-insulated from recession and AI disruption.

4. XOM Long (R:R 2.5:1): Ceasefire collapse → oil gap up Monday. 400 tankers queued, US blockade announced, 11M bpd removed from supply. Q1 earnings April 24 — first full quarter reflecting war pricing. At $153, still below $176 ATH.

5. GLD Long (R:R 1.8:1): The anomaly detector flagged gold with a 6.6σ volume surge while price was flat — classic institutional accumulation before a move. Gold at $4,760/oz with JPMorgan targeting $5,055 Q4 average. Central banks buying record volumes. Peace talks collapsed. De-dollarization accelerating (gold now exceeds US Treasuries as global reserve asset).

6. PANW Long (R:R 2.0:1): CyberArk $25B acquisition closed. NGS ARR $6.3B (+33%). Glasswing founding member. EU AI Act August 2 enforcement. At ~55x forward P/E, cheaper than CRWD (81x) with better growth trajectory.

11. PRICE TARGETS (3-Month Outlook)

Short Targets

Long Targets

12. RISK MATRIX

13. WHERE THIS ANALYSIS COULD BE WRONG

1. Enterprise AI adoption is 88% failure rate at production (40% probability). If AI agent deployment continues to fail at scale, the seat compression thesis extends by 2-3 years. Per-seat SaaS gets a reprieve. WDAY/PATH rally 30%+ from oversold levels. The CTO defection and SAP statement become premature signals. The mechanism: AI costs are real but ROI is illusory — enterprises discover agents hallucinate, require supervision, and don't actually reduce headcount.

2. Helium crisis forces a global chip production slowdown that crashes the hardware thesis (30% probability). If Asian fabs can't source helium for 3+ months, defect rates spike, yield drops, and TSMC/MU miss Q2 guidance by 10%+. The SOX index corrects 20% from ATH. The mechanism: semiconductor-grade helium has zero substitutes, new supply is years away, and the US blockade ensures Hormuz stays closed longer than the market expects.

3. Credit spreads are right and equities are wrong (25% probability). HY OAS at 294 says default risk is minimal. If bank earnings this week show clean loan books and stable provisions, equity fear is overdone and a 10-15% rally follows across all sectors. The mechanism: the sell-off was positioning-driven (extreme put/call), not fundamental — once positioning normalizes, prices revert to credit-implied levels.

4. Ceasefire revives faster than expected (15% probability). If backchannel diplomacy produces a workable framework within 1-2 weeks, oil drops $20/bbl, energy stocks retrace, and the risk-on trade that Friday started resumes with full force. The mechanism: both sides face domestic pressure (US elections, Iranian economic collapse) that could force a deal the principals publicly reject.

5. AI model providers vertically integrate infrastructure (20% probability). If OpenAI, Anthropic, and Google build their own observability, databases, and security tooling, the "infra benefits from AI" thesis breaks down entirely. The customer becomes the competitor. Evidence: OpenAI already building DDOG replacement. The mechanism: frontier lab margins are thin (OpenAI 33% gross margin) and they seek to reduce vendor dependency.

14. ML INTELLIGENCE DIGEST

Anomalies Investigated

61 total anomalies detected on April 13 (reflecting April 11 trading session).

Surprises Explained

1 surprise detected: DOCN predicted +1.47%, realized +6.01% (gap 4.54%). Direction correct but magnitude missed. DOCN has 66.7% hit rate (reliable ticker per the model). Likely driven by the broad regime-herding rally, not company-specific catalyst.

Contrarian Signals

The calibration data confirms ML contrarian signals should be taken seriously: UP predictions hit at 16% (worse than coin flip), meaning ML-predicted rallies are actually reliable DOWN signals with ~84% reliability. On April 11, the model predicted UP for most tickers — and they did rally. But the calibration says this rally is 84% likely to reverse. Given the ceasefire collapse, the contrarian interpretation is strongly supported.

Model Reliability This Cycle