The Benya whitepaper

What Benya does

Benya turns a repeatable stock-ranking process into portfolios you can inspect and follow. The AI reads public information and produces ratings. Portfolio rules decide which stocks to hold, how to weight them, and when to rebalance. Your brokerage holds your money and executes your trades.

This is a forward-tracked experiment, not a promise that AI can beat the market. A rising portfolio can reflect a rising market, company size, sector exposure, or chance. The record must be judged against appropriate benchmarks, trading costs, and the length of the observation period.

This whitepaper explains the live method and how to read it in the iOS app. It does not report results for strategies still under evaluation.

From public information to a ranking

Daneel, the live model, evaluates the Nasdaq-100 universe each week. Its web-backed analysis considers recent news, earnings, analyst guidance, and market reactions, with an approximately 30-day investment horizon. A weekly rating update is separate from the rebalance schedule of any portfolio built from it.

Each company receives a structured assessment, including a score and a finer ranking value. Portfolios sort by the finer ranking value first, then the integer score, then an alphabetical tie-break. A rating is not a probability of profit, a price target, or a guarantee that a higher-ranked company will outperform a lower-ranked one.

The inputs and generated analysis can be incomplete, stale, or wrong. Explanations make the model’s reasoning inspectable, but a convincing explanation does not establish predictive skill. Changing a prompt, model, or portfolio rule changes what is being tested and needs its own clearly identified record.

From a ranking to a portfolio

A model and a portfolio are different things. The model supplies the ranking; a portfolio applies a stated selection, weighting, and rebalance rule to that ranking. Daneel’s public range includes portfolios holding the top 20, 10, 5, or 1 stocks, with weekly, monthly, quarterly, or yearly rebalances. Read the rule shown for the particular portfolio you follow.

At a scheduled rebalance, the portfolio applies its rule to the available ratings and updates its target holdings. Equal weighting assigns the same target share to each selected stock at rebalance; price movements make those weights drift between rebalances. More frequent rebalancing creates more opportunities for costs and execution friction.

Holding more names reduces dependence on one company, but does not remove market or sector risk. A Nasdaq-100 portfolio can still be heavily exposed to technology and to a few related industries. A single-stock portfolio can lose much more abruptly than a broad index.

The live method, in numbers

Daneel’s rating schedule is weekly, with Monday as the configured UTC rebalance day. The analysis looks at roughly 30 days of public context. Its integer score runs from −5 to +5; scores of +2 or higher map to buy, −1 through +1 to hold, and −2 or lower to sell. These labels describe the model’s assessment, not an instruction to trade independently of your portfolio’s rule.

The finer ranking value runs from 0 to 1 and is ordinal: higher means more attractive within that week’s universe. It is not calibrated as a return forecast across weeks. Confidence also runs from 0 to 1, but is the model’s self-assessment, not a measured probability of correctness. The portfolio ordering is ranking value descending, score descending, then symbol alphabetically.

The calculated portfolio track starts with $10,000. Its trading-cost assumption is 15 basis points (0.15%) per unit of turnover. Turnover is half the sum of absolute changes between the old and new portfolio weights. A complete replacement has turnover 1; a turnover of 0.20 incurs a modeled cost of 0.03% of portfolio value.

Costs compound multiplicatively: net growth factor = gross growth factor × (1 − turnover × 0.0015). Initial entry is treated as a full buy-in with turnover 1 and zero gross return, so the initial cost is $15 on the $10,000 reference capital. Between rebalances, the model charges no rebalance cost and continues marking the holdings to market.

The daily track uses recorded market prices, with corporate-action adjustments for events such as splits and confirmed special distributions. Ordinary cash dividends are not reinvested by this calculation. The Nasdaq-100 and S&P 500 benchmark lines use index closes, not dividend-reinvested total-return indexes. These are price-based comparisons, not a complete simulation of every cash flow in a brokerage account.

Total return = ending equity ÷ starting capital − 1. Maximum drawdown is the largest percentage drop from a previous equity peak. Headline Sharpe uses week-over-week simple returns sampled at ISO-week closes, regardless of the portfolio’s rebalance cadence: mean weekly return ÷ weekly return standard deviation × √52. This implementation does not subtract a risk-free rate. Decision-cadence Sharpe, where shown, instead annualizes by √52, √12, √4, or √1 for weekly, monthly, quarterly, or yearly observations.

Model records and your returns

The model portfolio record is a calculated track from inception, using recorded holdings, market prices, and the method’s cost assumptions. It is not a brokerage account, an audited client return, or evidence that those trades were filled at those prices. “Since inception” refers to that model portfolio’s history, not the day you joined.

Your personal track begins when you enter and depends on what you actually do. Different purchase times, unfilled or partial orders, cash left over, deposits, withdrawals, fees, and taxes can all separate your outcome from the model record. Practice balances and manually completed checklists are records in Benya, not confirmation of a brokerage fill.

A model’s average summarizes its portfolio set. It is not a portfolio you can buy, and changing the set changes the average. The best portfolio is selected with hindsight when ranked by past return; its record does not tell you which portfolio will be best next.

Benchmarks, risk, and costs

Compare a Nasdaq-100 strategy with the Nasdaq-100 over the same dates first. The S&P 500 adds broader-market context, but beating it can reflect a technology or growth tilt rather than stock-selection skill. A simple portfolio of large companies without AI is another useful research control.

Total return measures change from the track’s starting capital. Drawdown measures a fall from an earlier peak. The displayed Sharpe relates returns to their variability without subtracting a risk-free rate; it is sensitive to the sampling window and assumptions. Annualized figures extrapolate a rate from the period observed and become especially easy to misread over short histories.

The calculated portfolio record deducts an assumed transaction cost based on turnover at rebalances. That is a model of trading friction, not a quote from your brokerage. Actual spreads, slippage, commissions, fractional-share restrictions, minimum order sizes, and delays can differ. Returns are pre-tax, and a modeled trading-cost deduction should not be read as including every cost of using the service or your account.

Compare like with like: the same dates, comparable price or total-return treatment, and clearly stated costs. Missing or delayed market data can also affect the record. A daily chart is not a live executable quote.

Following and trading at your brokerage

Following a portfolio does not itself buy its holdings. The available workflow depends on the brokerage, the permissions on your particular connection, and which capabilities are enabled in Benya. A brokerage connection may support account data without supporting in-app orders.

Where in-app ordering is available, review the destination account, amounts, and orders before submitting. Elsewhere, Benya provides a checklist for trades you make at your brokerage. Marking an item complete records your action; it does not verify that the brokerage executed it.

An order request or acceptance is not the same as a fill. Orders can wait, be partially filled, or be rejected or canceled. Check the reported order status and your brokerage’s confirmations. Market hours, buying power, order minimums, and connection permissions can prevent a portfolio from being reproduced exactly.

Benya does not take custody of your cash or securities. Removing only the follow record leaves your brokerage holdings in place. For eligible connected portfolios, Delete can instead offer a separately confirmed sell-and-delete flow with an order review. Read the confirmation before proceeding: submitting those sell orders still does not guarantee a fill. Choose a rebalance cadence you can realistically follow, especially if you place trades manually.

What would count as evidence

There are two separate research questions: does the ranking contain information about future returns, and does a portfolio built from it remain useful after costs? A profitable portfolio alone cannot answer the first question.

Ranking diagnostics compare subsequent returns across score groups and examine the relationship between scores and later returns. These require sample-size and uncertainty context. A score-to-return slope is not market beta, and a favorable result across many overlapping portfolios is not many independent experiments.

Research variants are evaluated against rules written before their scheduled read. Replaying a new rule on old ratings is exploratory research, not a new live track record. Choosing a rule after seeing a period and then presenting that same period as proof would overstate the evidence.

Benya has not established that its AI ranking delivers persistent stock-selection skill. Short records, shared holdings, tied ratings, sector concentration, and market regimes limit what can be concluded. Strategies under evaluation are not treated as validated because an early chart looks promising.

Before you follow

Read the portfolio’s holdings, concentration, cadence, inception date, and benchmark. Distinguish its calculated record from your own entry and execution. Check the capabilities of your actual brokerage connection, and understand which actions still require you to place trades.

You can lose money, including a substantial part of what you invest. Past performance does not predict future results. Benya does not know your full financial circumstances; a displayed portfolio or ranking is not a personalized assessment of suitability.