AI Agent · Investing · 2026
Situational-X
Conviction, audited weekly.
A thesis investor for the AI-infrastructure buildout - power, datacenters, storage - with what conviction funds skip: eight falsifiable KPIs scored from primary sources, and a valuation gate no narrative can override.
8
falsifiable thesis KPIs
3
tranches per entry - never all into strength
90
paper days before live
The problem
Thesis investing has a famous failure mode: conviction without disconfirmation. The AI-infrastructure trade - long the power plants, datacenters, and storage behind the buildout - is the most crowded conviction of the decade, and the funds running it publish no "I am wrong if..." conditions. Retail copying makes it worse: 13F filings arrive forty-five days late, so imitators buy last quarter's trade after the move, without the hedges. The question: can an agent hold a long-horizon thesis and audit it like an engineer instead of defending it like a believer?
The approach
Situational-X emulates the logic of the bottleneck trade - own the scarce physical links, not the crowded chip names - and adds the missing machinery. A Thesis Health Monitor scores eight observable KPIs every week from primary sources with citations: hyperscaler capex guidance, gigawatts of new power deals, transformer and turbine lead times, neocloud pricing, capability milestones, even a euphoria proxy that counts crowding against the thesis. The composite maps deterministically to a posture - ACCUMULATE, HOLD, or DE-RISK - and no eloquent narrative can outvote it. Entries come in three tranches, never all into strength; a valuation gate blocks any buy trading above the 90th percentile of its own two-year history; a permanent 25% cash reserve and a 20% drawdown breaker stand where hedges would, because at this scale honest ballast beats imitation options.
Architecture & build
Same chassis as its sibling Sentinel-X, opposite engine. Both agents share one ring-fenced brokerage account, partitioned into ledger-enforced sleeves that cannot spend each other's cash. Situational-X thinks weekly - a deep Claude run every Friday reads earnings calls, capex guidance, and power-deal announcements, scores the KPIs, and makes tranche decisions through the same validator discipline - while a daily health check rides along on Sentinel-X's closing run. Here the LLM is the primary analyst rather than a veto gate, but the rule holds: Claude proposes, code disposes. Results publish to a Zero Trust dashboard showing the thesis scoreboard next to the equity curve, benchmarked against SPY and against simply buying and holding the initial basket.
The outcome
In paper mode since July 2026, on a 90-trading-day clock. Going live requires beating both benchmarks net of modeled slippage, keeping drawdown under twenty percent, zero integrity failures, and one thing no conviction fund publishes: proof that the Thesis Health Monitor actually drove at least one documented posture decision. If the thesis cracks, the dashboard will say so before the account does.
- Universe
- AI power, datacenters, grid equipment, storage (~30 names)
- Cadence
- weekly deep run Fri 15:00 ET + daily sleeve check
- Risk
- 20% position cap, 25% cash floor, -20% sleeve breaker, no leverage ever
- Gate
- 90 paper days, beat SPY and the buy-and-hold basket
Personal research systems, not investment products. Nothing here is financial advice.