The proof · live registry
Our portfolios beat the world ETF, and everything else checks out.
The best of our model portfolios is ahead of the world ETF since its composition, and the detail sits below, served live by the engine. Behind the performance, a method that checks itself: every asset carries a dated view, and every forecast is logged then scored once the outcome is in. What we do not beat, we write down in the same place.
Our AI portfolios against the world ETF
Each model portfolio is measured since its composition date against the MSCI World ETF, a basket that tracks the global stock market. The figure is served live by the engine, so it stays current; “simulated” means the net asset value replays the composition at constant weights since its date, before fees and taxes. Portfolios that trail stay shown at the same contrast as the rest.
Reading the model portfolios…
Our grades hold, the consensus's invert
Grade discipline
B › C › Dorder held
A good grade should earn more than a bad one. With us it does, with the consensus it does not.
Our B grades earn more than our C grades, which earn more than our D grades. With the consensus, it is the D grades, its worst, that earn the most.
Every forecast is scored, and replayable
Calibration · Brier score
0.205on 32 resolved forecasts
Every forecast is logged with its probability, then scored once the outcome is in.
The Brier score compares each stated probability with the real outcome: the lower it is, the truer the probabilities were. The honest reference is 0.25, the score you get by answering with a coin flip.
On the 5 assets where our view diverges sharply from the consensus, we were right 2 times. Here are four, outcome stopped on 2 February 2026: two won, two missed. The outcome is the asset's return over the horizon.
| Asset | Consensus view | Our view | Outcome | Our call |
|---|---|---|---|---|
| StellantisSTLAP | Consensus viewneutral | Our viewavoid | Outcome-41.3% | Our callwon |
| TeslaTSLA | Consensus viewneutral | Our viewtrim | Outcome-23.6% | Our callwon |
| BoeingBA | Consensus viewbuy | Our viewavoid | Outcome+0.2% | Our callmissed |
| IntelINTCthe miss we publish on purpose | Consensus viewneutral | Our viewavoid | Outcome+86.4% | Our callmissed |
And when we say buy
Three summer buys, taken at their word.
Cards published on 12 July 2026, verdict and same-day price on the record. Here is what they have done since.
MicrosoftMSFT
+28.5%
Starter position · rated B on 12 July 2026 at $385.10
$495.04 as of 26 August 2026
PublicisPUB
+18.2%
Starter position · rated B on 12 July 2026 at €87.16
€103.00 as of 26 August 2026
ThalesHO
+12.2%
Starter position · rated B on 12 July 2026 at €223.30
€250.50 as of 26 August 2026
Three examples chosen from the 24 buy calls published on 12 July 2026: 16 have risen since, 8 have fallen, the worst sits at -11.4 %. Prices frozen as of 26 August 2026; every card stays dated and replayable.
An investment recommendation within the meaning of Regulation (EU) 596/2014. Produced by Finagus, methodology published on the method page, production date carried by each card. No retrocessions, no in-house products. Investing carries a risk of capital loss.
Anti-cheat probe
PASS · 0 / 6We asked the engine 6 questions whose answer was a future price it could not know. Zero correct answers, exactly what we want: it does not guess the future, it forecasts it blind.
Frozen and replayable
Grades are placed as of 2 February 2026 and counted by a rule written in advance. Anyone can replay the computation and get the same figures, nothing is adjusted after the fact.
The Challenger
Before publication, every thesis goes through an independent challenge round: an adversarial agent rereads the orders and the raw evidence, without our narrative, and looks for three concrete reasons not to execute. The card ships only after that round.
Finagus live registry · final reading · grades placed as of 2 February 2026, outcome stopped on 3 August 2026 · universe 28 assets, horizon 6 months. Past returns do not predict future returns.
The full table of 28 assets, dated and verifiable.
Same script, same frozen cards, zero touch-ups. Join the private beta to follow it asset by asset.
Comparison perimeter: the 11 test assets where the analyst consensus had a position reconstructed as of 2 February 2026, set against the more than 600 assets rated in the product. Consensus rebuilt from the dated recommendations of research firms. Past or simulated performance does not predict future performance. Finagus publishes general information, never personalised investment advice. Figures frozen on 23 August 2026.