First principles, then a million simulations.
Most research prices the world as it is. We price the world that's being built. Every reading answers two questions — is the thesis real? and is that already in the price?— with seven evidence-cited factors and a one-million-path Monte Carlo, read against the whole market's temperature. No free-text opinions. No black box. Every number cited, versioned, and reproducible.
Thesis Quality — seven factors, each scored by an agent against cited evidence, combined by a geometric mean so a single broken dimension can't be averaged away.
The price algorithm — a million-path Monte Carlo of where the company could be worth in five years, then the honest read of what today's price is already demanding.
Seven questions that separate creators from rearrangers.
Rearranging the present optimizes what exists. Creating the future builds what didn't. Each company is scored on seven dimensions drawn from Deutsch, Thiel and first-principles engineering — never a free-text opinion, always a cited number that moves only when the evidence moves.
Does leadership see something the consensus doesn't — and are they executing a multi-decade contrarian thesis, not chasing the quarter?
A track record of hard milestones actually hit, at scale. Ambition is cheap; delivery is the filter.
How hard is the problem? Physics-constrained and genuinely difficult means defensible. A commodity gets competed away.
Why won't this be commoditized in ten years? Every moat is stress-tested and returns a verdict — survives, mixed, or fails.
Deutsch's test of explanatory reach: does the core capability generalize into new domains, or is it a single trick?
Does the company solve the bottleneck binding its thesis today?
And the bottleneck likely to bind by 2028–2030? Weighted heavier — the future constraint matters more than the present one.
Combined by geometric mean — a weak moat or a missed decade drags the whole score down. There is no averaging away a broken thesis.
We only score companies that could matter.
Before a single factor is scored, a company has to clear five thesis-fit tests. It's the discipline that keeps the list to businesses becoming essential to the next economy — and it's the part with the longest track record.
We don't pick a price target. We simulate the distribution.
A point estimate pretends the future is a single number. It isn't. For every company we run a real one-million-path Monte Carlo of its value five years out, then read the whole distribution — the bear case, the base case, the bull case, and the tail where the great outcomes live.
Each of the million paths is drawn from correlated, empirically anchored assumptions — not round numbers. The building blocks:
Creators diverge upward; rearrangers de-rate. A creator-intensity score blends each company between the frontier it could ride and its own organic path.
EBITDA, enterprise value and net debt pulled straight from EDGAR filings — not a multiple of a proxy of a guess.
Power-law survival rates from the study of nearly every US stock since 1926. Most of the market's return comes from a tiny tail — so we model that tail explicitly.
Wright's Law cost declines by domain — AI accelerators, batteries, solar, robotics — so cost and volume compound the way they actually have.
Excess returns fade on a half-life set by the moat verdict. Monopolies hold for two decades; commodities fade in five years.
The law of large numbers, made explicit — a size-damped ceiling so no company is allowed to grow 30% forever.
What is today's price already assuming?
Turn the distribution around. Given where the stock trades now, what five-year growth rate is each outcome demanding? That single inversion places every company on one honest ladder — from priced below its own downside to priced beyond the tail.
Because the classification reuses the exact thresholds behind the four-tier signal, the read and the verdict can never disagree.
Four verdicts. And a hard line between buying and selling.
Quality, price and the market backdrop collapse into one shape-aware entry signal — read straight off where today's price sits in the modeled five-year range.
The whole modeled five-year range clears today's price — you're buying below its own downside.
The base-case upside isn't in the price yet. Room to build a position.
Fairly valued for the quality on offer — own it, don't chase it.
The future has caught up to the price. Patience is the position.
The signal only ever sizes an entry — Strong Buy, Accumulate, Hold, or Wait. It never says sell. Selling is governed by pre-committed thesis-break triggers: falsifiable conditions you set before you buy. A position can be fully valued and still held for years — until the thesis itself breaks, on rules written in advance, never on a nervous model.
Every score is read against the market's temperature.
A brilliant company bought at a manic top is still a bad entry. Two public gauges frame every reading — one structural, one tactical.
The discipline is in what we refuse to claim.
Error correction is the whole thesis. So the model is built to be wrong out loud, not confident in silence.
The distribution's validated job is catching failures — real busts fall below the bear band — and staying internally coherent. We do not sell you a next-quarter target or a promised hit-rate.
Every factor is evidence-cited, every constant is versioned, every snapshot is reproducible. When a score moves, its citation moves with it — and nothing touches the scoring logic without a backtest and a version bump.
The signal sizes entries only. Exits belong to thesis-break triggers you commit to in advance. A model should never panic you out of a position it can't re-underwrite.
The whole method, on one company the crowd wrote off.
Everything above — the seven factors, the million-path distribution, the implied read, the exit rules — applied to a company you already know. Click the price bands. Open any factor.
+3% / yr · Search holds, Cloud keeps compounding — and it still clears today's price.
DeepMind's discoveries generalize — AlphaFold rewrote structural biology; the same lab now does weather, materials and mathematics. Explanatory reach, not one product.
The market prices Alphabet as a search company being disrupted by AI. First principles say the opposite — it owns the deepest AI stack on earth: custom TPUs since 2015, DeepMind, Gemini, plus distribution to billions and Waymo's lead in autonomy. Even a bear case that gives Waymo zero credit and assumes search merely holds clears today's $3.0T.
At $3.0T, the price demands only ~3%/yr — below the base case. Upside to the tail is 4.5×.
Search query share falls for two consecutive quarters — the disruption thesis actually materializing.
Gemini falls behind the frontier on independent evals — losing the model race.
A forced structural breakup fragments the ad-and-distribution engine.
Illustrative reading, for demonstration. Live readings re-run on material events and are reserved for subscribers.
The method is the moat. The reports are the product.
The few companies becoming foundational to the next economy — each one scored, simulated a million ways, and put on the permanent record.