Forecasts · 6 bets · 9 refused
The AI's bets, stated in advance.
When the radar surfaces a cluster, the AI shapes it into a bet — naming the core names and stating a win-probability — or refuses it. Every bet is fixed before the outcome and resolved in public: a binary result and a categorical verdict against the mechanical baseline and the market. This is an experiment at full transparency, not a proven edge.
Open bets · 6
v5 theme-bets the model has entered, each with a stated win-probability up front. Distinct from the names held on the live book and the research theses scored on the track record.
- IRDM core ATEXMRCY
Sovereign and critical-infrastructure connectivity is re-rating as utilities, governments, and defense buyers harden and modernize their networks: satellite communications (IRDM), private-wireless spectrum (ATEX), and secure defense electronics (MRCY) are bid together on rising grid-security and defense capex. Broad participation across the group plus heavy volume confirm a genuine theme rather than a single-name squeeze.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
- MU coreSTX core SIMOWDCUMC
The AI data-center buildout is driving a genuine memory and storage supercycle: DRAM/HBM and NAND pricing are recovering while nearline HDD demand from hyperscalers stays tight. Micron and Seagate are the highest-quality, cash-generative center of this theme, with SIMO's NAND controllers and WDC riding the same storage upcycle and UMC providing broader foundry participation.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
- DELL coreNTAP core HPEOKTASNX
Enterprise AI-infrastructure buildout: hyperscaler and enterprise capex is pouring into AI-optimized servers, storage and IT distribution, and Dell is the volume leader turning that demand into record backlog. NetApp, HPE, the channel (SNX) and identity/security (OKTA) all ride the same datacenter refresh cycle real earnings, real breadth (4 peers +30% in 3m, 1.82x volume), not a single squeeze.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
- SEZL coreDAVE core HNGEXMTRTXG
Broad breadth (7 peers +30% in 3m) and 1.8x volume participation confirm real theme demand, not a single squeeze.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
- CNC coreHUM core OSCRHNGE
Managed-care and ACA-marketplace insurers are re-rating together on a margin-recovery narrative, with sector-wide breadth and heavy volume rather than a single-name squeeze.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
- RXO coreXPO core ARCBLSTRJBHT
A genuine freight-cycle upturn after a multi-year trucking recession: brokerage, LTL and truckload volumes and spot rates are re-accelerating, lifting asset-light and asset-based carriers together. RXO leads as the highest-beta brokerage play; XPO's LTL execution is the durable compounding anchor.
Invalidates if Invalidated on a sustained break of the cluster's long-term trend, or a daily close below the AI's pre-stated invalidation level.
What the AI refused · 9
Clusters the radar flagged but the AI declined to bet — co-movement that isn't a real theme. A refusal is a call too: curation only earns its keep if it beats betting every cluster.
- ATEX RefusedATEXIRDMATENMBX
Single-name re-rating masquerading as breadth. Members span four unrelated sectors 900MHz spectrum leasing (ATEX), LEO satellite (IRDM), application networking (ATEN), endocrine biotech (MBX) with only weak pairwise corr (~0.43–0.49) and the leader (+364%) far outrunning peers. No shared narrative driver; MBX is a binary-event biotech. Correlation-artifact grab-bag.
- DFTX RefusedDFTXAPGECGONKYMRBHVNACA
binary-event biotech (FDA/readout lottery) basket APGE/CGON/KYMR/BHVN are unrelated clinical-stage names (atopic derm, bladder cancer onco-virus, protein degradation, neuro) with no shared modality/indication; co-movement is XBI risk-on beta, not a narrative theme. Weak corr (0.48/0.48/0.44/0.41), only 3 peers >30% in 3m, and a lone +235% leader = single readout run masquerading as breadth.
- CORT RefusedCORTACHCBTSGDNLIRPRX
Correlation artifact: healthcare grab-bag, no shared narrative. Leader CORT (cortisol pharma, idiosyncratic ROSELLA ovarian-cancer catalyst) carries it; peers are unrelated models ACHC behavioral facilities, BTSG home-health services, DNLI binary neuro biotech, RPRX royalty aggregator. Uniform low corr ~0.40 = sector/factor beta, not a theme; breadth weak (2/4 >30%).
- SEZL RefusedSEZLTWSTDAVEDDOGSNEXVACLFSTXMTR
factor/beta momentum artifact, not a narrative theme. Members span 6 unrelated sectors (SEZL/DAVE fintech, SNEX brokerage, TWST synth-DNA biotech, DDOG SaaS, VAC timeshare, LFST mental-health, XMTR manufacturing) with only 0.40-0.58 corr to leader. Co-movement is risk-on high-beta ripping together, no shared story. Shaping would require dropping 5 of 8 names and manufacturing conviction.
- DVA RefusedDVAFLEXTGTXGEOLTHDOCJAZZPFGC
Momentum-factor beta basket, no shared narrative. 8 unrelated sectors: dialysis (DVA), contract electronics (FLEX), private prisons (GEO), MS biotech (TGTX), gyms (LTH), healthcare REIT (DOC), pharma (JAZZ), food distro (PFGC). Moderate ~0.5 pairwise corr = common momentum/beta, not a driver; each top name has its own story. No theme center to curate.
- DFTX RefusedDFTXAPGECGONKYMRBHVN
binary-event biotechs (FDA/readout lotteries) grab-bag of clinical-stage biotechs across unrelated indications (APGE immunology, CGON bladder-cancer, KYMR protein degradation, BHVN neuro), no shared compounding narrative. Co-movement is XBI sector-ETF flow beta plus idiosyncratic readout spikes; sub-0.5 corr on 3 of 4 peers = sector twins, not a theme.
- SYRE RefusedSYREORKARVMDMKSIAPGECGNXARWESI
binary-event biotech (FDA/readout lottery): leader SYRE and the two top-return narrative-center peers ORKA/APGE are pre-revenue clinical-stage I&I antibody names; remaining "breadth" (MKSI, CGNX, ARW, ESI) are unrelated profitable industrials sharing only risk-on beta (corr 0.41-0.54). Two incompatible sub-themes glued by beta, binary-biotech core.
- PTRN RefusedPTRNDDOGFTNTLFSTXMTRWRBY
index/ETF-flow twins heterogeneous high-beta growth basket (DDOG observability, FTNT cybersecurity, LFST behavioral health, XMTR manufacturing marketplace, WRBY eyewear) sharing no end-market or narrative; weak ~0.41-0.47 leader correlations = co-movement is momentum/growth-ETF factor flow, not a real theme.
- DVA RefusedDVAFLEXTGTXGEOVECODOCLTHVCYT
spurious cross-sector correlation / factor-beta grab-bag no shared narrative theme. Members span 6+ unrelated industries: DVA dialysis, FLEX electronics EMS, TGTX MS biotech, GEO private prisons, VECO semi equipment, DOC healthcare REIT, LTH gyms, VCYT diagnostics. Moderate corr (0.40–0.66) reflects common market beta, not a compounding theme. No coherent bet to shape.
The honest frame
An experiment, scored in the open.
Every mechanical harvest of this pattern lost to simply owning the market. The one untested part is whether the AI choosing the names, the membership, and the timing adds anything — and that can only be answered forward. So it ships as a forecast log with a stated probability on every bet, feeding the public Brier score, under pre-registered kill gates that shut it down if it doesn't work.
Common questions
- What is a theme forecast?
- When the weekly radar surfaces a cluster of stocks moving together, the AI either shapes it into a bet — picking the core names, and stating a probability that the theme plays out — or rejects it. Each shaped bet is an immutable, dated forecast: the probability is fixed before the outcome, so it can be scored.
- How is a bet scored?
- Two ways, both categorical and non-monetary. Against the raw mechanical cluster (equal-weighting every name the scan flagged) — did the AI's curation add anything? And against matched-window SPY — did the bet beat the market? Each resolved bet shows beat or trailed by a whole number of percentage points; the underlying returns are never published.
- Why publish the rejections?
- Because a refusal is a call. The radar is mechanical and flags plenty of co-movement that isn't a real theme — index twins, merger arb, a grab-bag of biotech readout lotteries. Showing what the AI declined, and why, is part of the honest record: the curation only earns its place if it beats simply betting every cluster.
- Is this investment advice?
- No. These are research forecasts, published to be scored in public — not recommendations, not a trade to copy. No position sizes, prices, or account details are ever shown.