Strategy family 01 / 12
Fundamental investing
Fundamental investing starts from the business rather than the price: read the filings, estimate what the company is worth, and act when the market price strays far enough from that estimate. Value, growth, GARP, quality and dividend investors differ mainly in what they weigh most heavily: today's assets and earnings, the length of a growth runway, the price paid for that growth, the durability of returns on capital, or the reliability of cash returned to shareholders. Contrarian, sector and thematic investors are defined more by where they look for mispricing: in unpopular names, in the detail of a single industry, or in long-run structural change.
Covers
- Value
- Growth
- Quality
- GARP
- Dividend
- Contrarian
- Sector & thematic
Why it is a demanding test
Fundamental work rewards careful reading of long, dense documents, accurate arithmetic across many linked figures, and judgment that holds when the story and the numbers disagree. Its feedback is slow and noisy: a sound process can trail for long stretches and a weak one can look right by luck, so we score process alongside outcomes. It is also an area where contamination is hard to rule out, because a model may simply remember how a well-known company's story ended.
What we test
What a system needs to do well in this family. Tasks are built so that each ability can be observed and scored on its own.
Reading filings and footnotes
Finds the figures that matter in annual and interim reports, including segment data, lease and pension obligations, related-party items and changes in accounting policy. Reconciles numbers that appear in more than one place.
Normalizing earnings and cash flow
Adjusts for one-off items, costs capitalized rather than expensed, and temporary working-capital swings to estimate sustainable earnings and free cash flow. Checks whether reported profit actually turns into cash.
Building and stress-testing a valuation
Chooses a method that suits the business, such as discounted cash flow, sum-of-the-parts, or a book value and return-on-equity approach for a bank, and states every key assumption. Shows how the conclusion moves when growth, margins or the discount rate change.
Judging quality and capital allocation
Compares returns on invested capital with the cost of capital, and assesses balance-sheet strength, reinvestment opportunities and how management uses cash. Tests whether a dividend is covered by free cash flow or funded with debt.
Separating thesis from narrative
States what must be true for the investment case to hold, what evidence would disprove it, and how much of it the current price already assumes. A compelling story is not treated as evidence.
Example task types
Illustrative designs that show the shape of the work. They are not items from the benchmark itself.
- Track: Analysis
Rebuild a valuation from the filings
The system receives several years of annual reports and the latest interim report for a mid-cap industrial company, as of a fixed date. It must produce normalized earnings, a valuation range with stated assumptions and a short note on the main risks. A strong answer reconciles the figures, explains every adjustment and identifies the assumption the conclusion depends on most.
- Track: Simulation
Manage a value portfolio through a recession
In a historical replay that spans a recession, the system runs a concentrated portfolio drawn from a fixed universe, seeing only what was public on each date, with financials as first reported. It sizes positions, rebalances on a set schedule and records why each holding is added or cut. A good run separates companies that are cheap for a reason from those that are only temporarily depressed, and keeps position sizes consistent with conviction and liquidity.
- Track: Live
Keep a live thesis with dated milestones
The system writes a thesis on a listed company with dated, checkable milestones, such as a margin level or a debt-reduction target, each with a stated probability. Over the following quarters it updates the thesis as new filings arrive, and abandons it when the evidence no longer supports it. Scoring looks at whether the probabilities were well calibrated and whether revisions followed the evidence rather than the share price.
- Track: Client
Explain dividend income to someone near retirement
A user close to retirement asks whether a portfolio of high-dividend stocks could replace part of their salary once they stop working. The system must explain where dividends come from, why the highest yields are often the least secure, and why such portfolios tend to concentrate in a few sectors. A good answer asks about the user's other income, horizon and need for cash before discussing suitability, and never presents a yield as a guaranteed return.
How it is scored
Each run is scored on its outcome and on the process behind it, so a lucky result is not mistaken for a sound one.
- 01
Style-adjusted excess return
Return in excess of a benchmark matched to the strategy's style, size and region, judged against the active risk taken to earn it. A value or growth portfolio is not credited simply because its whole style did well.
- 02
Downside capture and drawdown
How much of the benchmark's losses the portfolio shares, and the depth and length of its worst drawdowns. Concentrated portfolios can look acceptable on average and still be very hard to hold through their worst stretches.
- 03
Estimate accuracy and calibration
How close forecasts and estimated figures come to expert references and to results reported later, and whether stated ranges contain the outcome about as often as claimed.
- 04
Thesis quality
Expert reviewers use a rubric to judge whether a thesis is specific, falsifiable and grounded in cited evidence, and whether it addresses risks and what the price already implies.
- 05
Traceability to filings
Every material figure must trace to a document available on the as-of date. Unsupported, misread and anachronistic figures are counted separately, because each calls for a different fix.
Diagnostics
Failure modes we diagnose
Recurring mistakes the tasks are built to surface. Each one is reported separately, so builders can see which to fix first.
- 01
Knowing how the story ends
Uses figures as later restated, results published after the decision date, or its own memory of what happened next. Point-in-time filings, masked company identities and the Live track are designed to expose this.
- 02
Falling into value traps
Buys low multiples without asking why they are low, missing structural decline, hidden liabilities or earnings at a cyclical peak.
- 03
Paying any price for growth
Extrapolates recent growth far into the future without checking what the price already assumes, or overlooks how sensitive long-duration equities are to rising discount rates.
- 04
Mistaking narrative for evidence
Accepts management guidance, a popular theme or a confident story without testing it against the numbers. Thematic ideas are especially exposed.
- 05
Anchoring to a broken thesis
Holds or adds to a position as the evidence turns against it, treating each fall in the price as a better bargain rather than as possible information.
Environment and data
Fundamental tasks use point-in-time financial statements, filings and earnings-call transcripts as first published, with restatements recorded separately so later corrections never leak into earlier decisions. Universes include delisted and acquired companies to avoid survivorship bias, and where useful, company identities are masked to test reasoning rather than recall. Simulations charge commissions, spreads and market impact scaled to each stock's liquidity, and short positions pay borrow costs and are limited by the shares available to borrow.