Method Statement
Published in full, as required for any recommendation disclosed under EU Market Abuse Regulation article 20 — minus the per-ticker numbers.
The formula
weighted_sum = sum(Graham) x1 + sum(Buffett) x2 + sum(Lynch) x3 + penalties (negative); score = max(0, weighted_sum) / denominator x 100
Score per criterion — met / partial / not met: 1 / 0.5 / 0.
Frameworks and weights
| Framework | Focus | Weight | Criteria |
|---|---|---|---|
| Graham | Safety and value | ×1 | 10 |
| Buffett | Quality and durability of the business | ×2 | 10 |
| Lynch | Growth at a reasonable price, understandability | ×3 | 10 |
The 30 criteria
| # | Framework | Criterion | What it looks for | Weight |
|---|---|---|---|---|
| 1 | Graham | Current earnings multiple, against its own sector | priced in the cheapest tier of its own sector | ×1 |
| 2 | Graham | Price to book — or cash generation, for intangible-heavy sectors | book multiple low by the standards of its sector; for technology, semiconductors, pharma and e-commerce a high free-cash-flow yield replaces it, because book value means little there | ×1 |
| 3 | Graham | Earnings multiple times book multiple | the combined earnings-and-book multiple below a Graham-style ceiling | ×1 |
| 4 | Graham | Dividend actually paid | a dividend paid without interruption over the last five years | ×1 |
| 5 | Graham | Dividend yield | a dividend yield well above what the average listed company pays | ×1 |
| 6 | Graham | Leverage | debt below equity, or a net cash position; for banks and insurers this is judged on regulatory capital instead | ×1 |
| 7 | Graham | Return on invested capital against the cost of that capital | returns on invested capital clearly above what the capital costs; not computable for financials, which is why their denominator differs | ×1 |
| 8 | Graham | Earnings growth over five years | earnings per share higher than they were five years ago | ×1 |
| 9 | Graham | Consistency of profit | profitable in every one of the last five years | ×1 |
| 10 | Graham | Price against the Graham Number | price below the Graham Number computed from earnings and book value | ×1 |
| 11 | Buffett | A durable moat | a competitive advantage that is clear and defensible — brand, network, switching cost, licence or cost position | ×2 |
| 12 | Buffett | How long the moat lasts | an advantage that should still be standing a decade from now | ×2 |
| 13 | Buffett | Return on equity, sustained | a high return on equity held for years without leaning on debt, measured after one-off gains are normalised away | ×2 |
| 14 | Buffett | Margins against the sector | net or operating margins above the sector average, by a clear gap | ×2 |
| 15 | Buffett | Quality of management | a management team with a long track record of allocating capital well | ×2 |
| 16 | Buffett | Skin in the game | management owns a meaningful part of the company and is not selling | ×2 |
| 17 | Buffett | Predictability | a business simple enough that next year can be reasoned about | ×2 |
| 18 | Buffett | Retained earnings put to work | every unit of profit kept in the business creates more than a unit of value | ×2 |
| 19 | Buffett | Pricing power | able to pass inflation on to customers without losing them | ×2 |
| 20 | Buffett | Margin of safety against the valuation models | price meaningfully below where the triangulated models put intrinsic value; for banks the dividend discount model is used, never analyst consensus | ×2 |
| 21 | Lynch | Growth against the multiple paid for it | the multiple paid is small relative to the growth rate bought | ×3 |
| 22 | Lynch | Rate of profit growth | earnings per share compounding fast over the last three years, after one-off and cyclical gains are normalised away | ×3 |
| 23 | Lynch | Fits a Lynch category | clearly a fast grower, a stalwart or an asset play — not an ambiguous cyclical or a risky turnaround | ×3 |
| 24 | Lynch | An unfashionable corner of the market | in a business or a sector nobody is excited about | ×3 |
| 25 | Lynch | A story that can be told simply | the investment case can be explained in a couple of minutes, without jargon | ×3 |
| 26 | Lynch | Insider ownership | founders or insiders hold a large stake and have been buying, not selling | ×3 |
| 27 | Lynch | A solid balance sheet | net cash, or debt far below equity | ×3 |
| 28 | Lynch | Market share | taking share from competitors rather than merely defending it | ×3 |
| 29 | Lynch | Analyst coverage | followed by few institutional analysts, if any | ×3 |
| 30 | Lynch | Dilution from share-based pay | share-based compensation consumes only a marginal slice of revenue each year | ×3 |
Maximum points by instrument type
| Instrument | Max points | Note |
|---|---|---|
| Standard equity | 60 | 10 Graham x1 + 10 Buffett x2 + 10 Lynch x3. |
| Financials (banks, insurers) | 59 | ROIC has no meaning for a bank, so criterion G7 is removed from the numerator and from the maximum. |
| Closed-end fund | 42 | Nine criteria have no operating counterpart in a fund. They are forced to 0 and removed from the maximum together, so dropping them cannot inflate the percentage. |
| Passive index ETF | 13 | Only about nine criteria apply at all. |
Penalties
- Red flags. Accounting red flags found in the filings subtract from the weighted sum before the division. The number of points is not published.
- Valuation. A price far above every model in the triangulation subtracts as well: a great business at an indefensible price is not a candidate.
- Floor. The weighted sum is floored at zero, so penalties cannot produce a negative score.
- Evidence. A second, shadow axis scores how well each claim is backed by a primary document. It is tracked but, in the current method version, it does not enter the published score.
Verdict bands
| Verdict | Range |
|---|---|
| Strong buy candidate | 82–100 |
| Buy candidate | 72–82 |
| Interesting | 62–72 |
| Monitor | 48–62 |
| Speculative | 32–48 |
| Avoid | 0–32 |
Notes
- The criteria and their weights are published because a published methodology is required (EU Market Abuse Regulation, art. 20). Each criterion is described in words; the exact numeric cut-off behind it, and the per-ticker score against it, are not published.
- The published score is a 10-point band, never the exact percentage.
- Sector exceptions: for intangible-heavy sectors (tech, semiconductors, pharma, e-commerce) criterion G2 uses FCF yield instead of P/B; the denominator stays 60.
- A company with negative earnings is flagged 'outside the framework': the Graham criteria cannot be computed at all, so the low score is arithmetic, not judgement.
Who produces this
Behind this site there is one person — the author — and a software system he wrote and maintains. There is no research department, no sales desk, no brokerage, no client mandates and nobody paying for coverage.
The division of labour is stated openly, because it is what tells you where this site is strong and where it is not:
The method is human. The thirty criteria, their weights, the cut-offs, what gets penalised, which company is worth a week of reading — and, at the end of all of it, whether anything is bought at all — are the author’s decisions, made once and then applied identically to everyone.
The mechanical part belongs to the software. It pulls the filings, computes the criteria, runs the valuation models and the simulations, and stores every evaluation with its date.
The deep reports are drafted by a large language model — Claude, from Anthropic — reading the company’s primary filings under one rule it does not get to bend: every figure in the text points at the document and the page it came from. What cannot be shown in a filing does not go into the report. The author decides what is published.
This is written here, in the open, because it is the page’s best argument rather than a weakness to hide. A model has no client, no relationship with management, no sales target, and does not get tired at the three-hundredth filing: it reads that one with the same attention as the first and applies the same thirty criteria the same way. That is exactly what a human analyst, however honest, cannot do — and without it, comparing two companies on two different pages would mean nothing. Consistency is the kind of objectivity this site can offer: not neutrality, reproducibility. It is also why a company is re-evaluated when new filings appear, rather than once a year when a budget allows it.
The limit is stated just as plainly. A language model can be confidently wrong, and can produce a figure that sounds right without being right; it answers for nothing and has nothing to lose. That is why every numeric claim in a report carries its source next to it, why the verdict is recomputed from stored subtotals instead of being read out of a paragraph’s conclusion, and why the last step — the decision — is human and is not optional.
Everything here is general investment research: addressed to no particular person, taking account of nobody’s circumstances, containing no personal recommendation. The vocabulary is deliberate. A company that scores well is a candidate for further work, and the label on it is a verdict on an assessment, not an instruction.
The five steps
The diagram at the top of this page is the whole process; this is the same thing in words, with the part that a picture cannot carry.
1 — Filters. A few hundred listed companies across a dozen-plus exchanges are put through the same screen before anything else happens. The families used, without their cut-offs: price against earnings and against book value, both taken relative to the company’s own sector rather than absolutely; total debt against equity and against operating cash flow; return on invested capital against the cost of that capital; operating and net margins against the sector; dilution of the share count over three years; insider and founder ownership, and whether it is being added to or sold; and a liquidity floor, because a name nobody can trade is not a candidate whatever it scores. Companies with negative earnings are not filtered out — they are flagged outside the framework, since several Graham tests cannot be computed for them at all. Most companies fail several filters, which is the point: the useful output of a screen is the rejection, not the ranking.
2 — Thirty criteria. What survives is scored against thirty criteria drawn from Benjamin Graham, Warren Buffett and Peter Lynch — ten each, weighted one, two and three. Each criterion is scored met, partly met or not met; the scores are weighted, penalties subtracted, and the result divided by the maximum available for that kind of instrument. The same pass also produces a first-draft intrinsic value, which is not the answer — it is the number the deep work is later checked against.
3 — Valuation range. Several discounted-cash-flow models are built for the same company and triangulated against one another, with an earnings-power estimate and an asset-based check where those apply. A Monte Carlo simulation then re-runs the valuation many thousands of times with the key assumptions drawn from distributions instead of fixed at a point, producing a range and a shape rather than a number. A single figure presented as the value of a company is a false precision this site will not sell.
4 — Deep evaluation. What survives the range gets read properly: annual and interim filings, segment notes, the cash-flow statement line by line, insider ownership, compensation, prior restatements. It ends with a thesis in one sentence and a list of things that, if they happened, would show the thesis was wrong. Those falsifiers are written before the verdict, not after. This is the slow step, and the three before it exist to earn it.
5 — Read the quarterly yourself. Optional, and strongly recommended — and it is the reader’s step, not the author’s. Open the latest filing and read it before acting on anything here. A score is a summary and a summary is not understanding. The aim of the whole exercise is to end up owning a business you can follow for years, and the only way to find out whether this is one of them is to read a quarter of it in your own words. If the industry still makes no sense to you afterwards, that is an answer: it is not your company, whatever the score says.
The three legs a decision stands on
The steps above produce two independent measurements of the same company, and they are deliberately not the same measurement twice.
Leg one — the score. Thirty criteria, weighted and penalised, expressed as a band. This says what the company is: how cheap, how profitable, how durable, how honestly financed, measured the same way as every other company in the universe. It is comparative by construction — its whole value is that a 70–80 here means the same thing it means on the next page.
Leg two — the intrinsic value. Five models and twenty thousand simulations, expressed as a range against today’s price. This says what the company might be worth, and it is computed from cash flows and assumptions, not from the scorecard. The two legs can and do disagree: an excellent business at a demanding price scores well and prices badly, a mediocre one at a distressed price does the reverse. That disagreement is information, and collapsing the two into a single number would destroy it.
Leg three — your judgement of the risks. The deep report names the specific things that could break the thesis. What it cannot do is tell you how much each of them weighs for you: whether a founder-controlled share structure is a dealbreaker or a detail, whether a business whose revenue depends on one government programme is a risk you understand well enough to carry for ten years. Nobody can supply this leg from the outside, and it is not optional. A high score and a wide margin of safety on a business you do not understand is not a case; it is two numbers.
The three frameworks, and how they are weighted
The thirty criteria split evenly across three frameworks, ten each, and each framework carries a different weight in the sum. The weights are published because a disclosed methodology is required of anyone disseminating investment research in the EU.
Graham, weight one — safety and price. Cheapness against earnings and against book value, sector-relative; the two together inside the classic combined limit; a dividend paid consistently and worth having; modest debt or net cash; returns on invested capital above the cost of that capital; earnings higher than five years ago and positive throughout; a price below the Graham number.
Buffett, weight two — quality and durability. A competitive advantage that plausibly survives a decade; high returns on equity without leverage doing the work; margins above the sector; a real record of capital allocation; insiders holding a serious stake and not selling; a predictable business; retained earnings that produce more than they cost; pricing power against inflation; a margin against the discounted-cash-flow estimate.
Lynch, weight three — growth at a sensible price, and understandability. Growth cheap relative to its own rate; earnings compounding quickly over three years; a recognizable Lynch category; a sector nobody is looking at; a story explainable in two minutes; founders holding a large block and buying; a strong balance sheet; market share being won; thin institutional coverage; stock-based compensation that is not quietly diluting shareholders.
The numeric thresholds behind each criterion are the calibration, and the calibration is the product. The criteria are named above; the cut-offs, the penalty sizes and the parameter distributions used in the simulation are not. That is the line the regulation itself draws — disclose the method, not the proprietary tuning.
Two structural rules. Where a company’s value sits in intangibles rather than book assets, the price-to-book test is replaced by a cash-flow yield test. And a company with negative earnings is flagged outside the framework: several Graham criteria cannot be computed at all, so a low score there is arithmetic, not an opinion.
Maximum points by instrument type
The denominator changes with what is being scored, so percentages stay comparable. A standard operating company is scored out of sixty. A bank or insurer out of fifty-nine: return on invested capital has no meaning for a balance-sheet business, so that criterion leaves both the numerator and the maximum. A closed-end fund out of forty-two — nine criteria have no operating counterpart in a fund, and they leave the maximum with the numerator, so dropping them cannot inflate the result. A passive index tracker out of thirteen.
What a verdict means
A verdict is a band, not a score, and it describes the status of a candidate in the author’s own research queue.
Strong buy candidate and Buy candidate: the assessment cleared the top bands, and the company would go on the author’s shortlist for further work. Interesting: it cleared the framework, but not by enough to move ahead of what is already in the queue. Monitor: the business is understood and tracked, without today’s assessment supporting more. Speculative: the case depends on something that has not happened yet. Avoid: the assessment found enough against the company that no further work is planned. Not scored means exactly that.
None of these labels tells any reader to do anything. They describe where a name sits in one analyst’s process, at a stated date, on stated criteria.
Sources
Primary sources are the companies’ own filings — annual and interim reports, and the statutory disclosures of whatever regime the issuer reports under: the SEC’s electronic filing system for US issuers, the equivalent national or exchange systems elsewhere. Ownership and compensation data comes from the same filings. Market data comes from commercial market-data providers under their own terms. This site publishes no live prices, which keeps it clear of the licensing question and of showing figures already stale by the time they are read. Where a report leans on an industry source rather than a filing, it says so at the point of use.
How often this is refreshed
The evaluation database is regenerated daily and the site rebuilt from it, so verdicts, bands and evaluation dates are never more than a day behind. A deep report is not on a schedule: it is rewritten when the filings change the argument — for most companies, at annual results and after any event that touches the thesis. Every company page shows both dates; where they are far apart, the gap is itself information.
What the free version withholds, and why
Published for everyone: the universe with verdicts and score bands, this methodology, the twelve-month record, the one-sentence thesis, the key risks, what would change the verdict, the shape of the valuation range and of the simulated outcomes, and the first chapter of the deep report.
Withheld: the exact score behind the band, the framework subtotals, the per-criterion scores and their evidence, the intrinsic-value estimate, the margin of safety, the simulated probabilities and percentiles, the valuation assumptions, and the rest of the report.
The reason, stated plainly: the arithmetic took years to calibrate and it is what a paying reader will eventually pay for. That is a commercial decision, disclosed here so nobody mistakes an ellipsis in an excerpt for the author’s own hesitation. Five companies a week have their full report and figures opened to everyone, and one a week has the whole method shown end to end.
The twelve-month record
Every past verdict is published with its date, because a methodology without an outcome record is a marketing document. Read the track record before reading any verdict here: it says the record is a paper one, that it shows no demonstrated edge over the universe it was drawn from, and that the horizon measured so far is far too short for a framework built for years.
Each entry is one verdict, for one company, on one day, recomputed from the stored framework subtotals rather than read from the stored verdict column, which is an append-only audit trail never corrected retroactively. Evaluations missing a subtotal are excluded, and the number excluded is disclosed. A company re-evaluated twice in a day counts once.
Conflicts of interest
The author may hold positions in the securities covered here, and in some cases the interest in a company began with a position rather than with the screen. That is a real conflict, not a theoretical one, and it is disclosed rather than avoided.
The policy: where the author holds a position in a covered security, that fact is shown on the company’s own page alongside the verdict, and updated when the position changes. No position is opened or closed in the twenty-four hours around publishing new material on it. The author takes no payment from any covered company, any brokerage, or anyone with an interest in the price of a covered security, and accepts no sponsored coverage.
Disclaimer
This site publishes general information and general investment research. It is not investment advice, not a personal recommendation, not an offer or solicitation, and it takes no account of any reader’s objectives, situation or risk tolerance. No verdict here tells any reader to buy, sell or hold anything. Past performance, real or hypothetical, does not predict future results. Data may be delayed, incomplete or wrong, and the models rest on assumptions that may be wrong. Every decision a reader makes is that reader’s own. See the full disclaimer.