One developer, not a company
StockScorer rates stocks by disclosed, rule-based criteria instead of gut feeling. This page covers who runs the project, where the data comes from, how a score is built, and where the backtests honestly reach their limits.
Sunday night, 11:12 pm. Fourteen open tabs, two spreadsheets, three opinions from three forums, and in the end it still comes down to a feeling. Sounds familiar. That's exactly why this software exists.
Metrics, prices, and index compositions come from leeway, a commercial market data provider. An automated pipeline updates them once a day after market close. Scores and metrics are based on closing prices, there are no real-time scores. Only in the portfolio does StockScorer fetch a current price from leeway on demand (briefly cached), so your portfolio value is right. For a rule set that looks at annual figures, a faster cadence would only be noise anyway.
Every stock page shows how fresh the data is: with the price date, not with "live".
Every data run measures its own success rate. If it drops below 80%, say during a provider outage, downstream steps like sector averages and the score snapshot are skipped automatically. A score from yesterday beats a wrong one from today. This is how a nightly run went, shown here as a historical example (step names and values from the pipeline's run record):
A scoring profile is a list of rules: each rule checks one metric against a value you set with a slider, and awards points. You can build your own profile from over 100 metrics, take the StockScorer profile as-is, or clone a profile from the library and tune it to your taste. Evaluation is purely rule-based, no black-box AI, no hidden weighting. The AI summary of the annual report on some stock pages does not feed into the evaluation.
Return on capital, margins, leverage, growth, valuation. You decide what counts and what you ignore.
Return on capital (ROC) at least 20, gross margin greater than 40, debt-to-equity less than 0.5. Every rule gets its own point value, negative points included. Rules can be grouped, negated, or set as a hard exclusion.
The score is the sum of the triggered rules on your profile's scale, and it can be negative. Four points mean four points. Public stock pages and comparisons across rule sets additionally show a derived 0-to-10 display. For it, the points are divided by the maximum attainable points, multiplied by 10 and rounded to one decimal place. A negative sum is shown as 0. Example: 5 of 11 points gives 4.5 of 10.
You decide how many points make a stock a low, medium, or high match with your profile, split by large, mid, and small cap if you want (the example profile here uses one threshold for all). This classification is deliberately neutral and is not a buy or sell recommendation: it only describes how well a stock fits a rule set.
The "Moat Quality" profile looks for companies with durably high returns on capital, a defended gross margin, and strong cash flow. Nine rules, exactly as stored in the profile. All rules except the return-on-equity group and the last one apply to non-financial companies only: banks, insurers and real estate are not meaningful on these metrics and get the last rule instead. ROC stands for return on capital employed.
Every line is yours to touch: drag the threshold slider, change the points, negate a rule, compare against another metric instead of a constant (this year's gross margin against last year's), or mark a rule as a hard exclusion. Missing data is its own case and is never silently scored as "passed".
That's your call, not mine. Two thresholds per profile, the upper and the lower score threshold. Optionally they can be split by cap class, because the same point total can mean something different for a small cap than for a large-cap giant. The "Moat Quality" profile does not split: upper threshold 5, lower 2, for all cap classes.
On a tie, a metric of your choice decides, ROC in this example, higher is better. And if you have no idea where to start: the Strategy Optimizer (Premium) backtests other thresholds and rules and proposes concrete changes. Purely informational, you adopt them or you don't.
Check it yourself
This is the full math behind the card at the top of the page, with SAP SE's real metrics. Nine rules, four of them triggered, five not. SAP sits one point below the upper threshold, and that is exactly what the breakdown shows: nothing gets smoothed over until it fits.
The raw sum is only comparable within the same profile. A 7 from one profile and a 7 from another are two different things, and that's the point: it's your scale. Only the derived 0-to-10 display is comparable across profiles.
Check it yourself
A profile that gives every second stock top marks helps no one. This is how the scores of the "Moat Quality" profile spread across 16,222 stocks, one bar per point value reached. On the left are the names with negative points, on the right the few with the highest value (7). Mathematically up to 10 points are possible, but rules that exclude each other cap the real maximum at 7.
Measured 2026-10-01 (example) · profile "Moat Quality" · real distribution across all active stocks
Check it yourself
Historical simulations almost always suffer from the same two problems. StockScorer addresses both explicitly, not because it's convenient, but because a backtest that looks too good is worse than none at all.
Looking only at companies that still exist today writes the failures out of history. StockScorer reconstructs index membership at every point in time: delisted companies stay in the history and exit at their last known price instead of simply disappearing. If no exit price exists, the backtest assumes proceeds of 70% of the last valuation price (an assumption, shown in the result). This applies to index universes with stored membership history. One gap remains: companies that left the market early and for which no quarterly data exists are missing from the backtest, which favours the result.
A period's score and entry rely solely on data that was actually available at that time. Returns are only measured in the following period, never in the same window the selection came from.
Every backtest also shows its data quality and its gaps. Once enough data points exist, a strategy page additionally shows a real live measurement (out-of-sample) since the start of the current profile version. That one runs forward, not backward. A rule change starts a new version.
Check it yourself
Metrics, prices, and scores once a day after market close, via an automated nightly pipeline. There are no real-time scores. Only in the portfolio and when pre-filling a price does StockScorer fetch a current price from leeway on demand (briefly cached). Every stock page shows the price date of the data.
No. A score only expresses how strongly a stock matches a profile's rules. The presentation is purely informational and does not replace individual investment advice.
Both are the same quantity on two scales. The score is the raw sum of the points your rules award, and it can go negative too. In the rule table you see for every stock exactly which rule contributed which point. The main number on the stock page is derived from it: the points are divided by the maximum attainable points, multiplied by 10 and rounded to one decimal place, so 5 of 11 points becomes 4.5 of 10. A negative sum is shown as 0. The raw sum is only comparable within one profile, while the 0-to-10 display is comparable across profiles.
No. You can use the StockScorer profile as-is, clone a profile from the library, or build your own from over 100 metrics. Most people start with a ready-made profile and tune it after a few weeks.
On purpose: the StockScorer Score is deliberately geared towards very large companies. Within the same rating, companies worth 300bn euros or more come first, then companies from 25bn euros, and only after that does the point total decide. In the ten-year backtest, more than half of the positions were therefore companies worth 300bn euros or more, and companies below 25bn euros practically never entered the portfolio. Companies of every size still receive a rating. If you want smaller companies up front, copy the profile and change the size classes and the priority in the rule editor.
Only in part. For index universes with stored membership history, composition is reconstructed at every point in time, delisted companies stay in the history and exit at their last known price. If none exists, 70% of the last valuation price is assumed (an assumption, shown in the result). However, companies that left the market early and for which no quarterly data exists are missing, which favours the result, especially in early years. The backtest states the date from which point-in-time composition applies; for universes without history a warning appears instead.
Create a profile, run it against 16,222 stocks, check it against the history. Start for free, no sales call, there's no one who could make one.