Skip to main content

Methodology

How the models trade, and how we measure them.

The rules behind Alpha Balanced and Alpha Max, how every live-tracked account and backtest figure on this site is computed, and where the backtest falls short.

Large and mid caps, long-only
US
Rebalance
Weekly
Trailing-stop check
Daily
Per side, backtest costs
6 bps

The process

Six rules, run the same way in the live-tracked accounts and in the backtest.

Long-only, trend-filtered momentum across US large- and mid-cap stocks, with a weekly rebalance and a daily trailing stop. Alpha Balanced adds volatility management and a cap on exposure in weak markets; Alpha Max is more concentrated.

  1. Universe

    A fixed list of US large- and mid-cap stocks. A stock is eligible only while it is in a long-term uptrend and trades above a price floor. SPY is never traded; it is only the market gauge.

    US large and mid caps · uptrend filter

  2. Signal

    Eligible stocks are ranked on a blend of past returns over several horizons, measured on prior closes. There is no volatility penalty, no sector tilt and no machine learning.

    Momentum across several horizons

  3. Sizing

    The top-ranked names are weighted so that calmer stocks carry more weight, with a cap on any one name. Orders are in whole shares, and very small orders are skipped.

    Volatility-aware weights · single-name cap

  4. Rebalance

    Weekly. Names that leave the top list are sold and the rest are resized. There is no fixed weekday.

    Weekly, no fixed weekday

  5. Exits

    Every trading day, each position is checked against a trailing stop that follows its highest close since entry. A stopped-out name is not bought back the same day.

    Daily trailing stop, on the prior close

  6. Risk overlay

    Alpha Max has none beyond the stop. Alpha Balanced adds volatility management: it sets exposure from the market’s trend, scales it down when markets turn volatile, and caps it at a low level in weak markets, checked daily.

    Alpha Balanced only

Two profiles

Same process, two levels of risk.

Universe, signal, weights, rebalance and stops are identical. Only concentration and exposure differ.

Alpha Max

Concentrated and mostly invested. Higher return potential, deeper drawdowns.

Holdings
Fewer names: the concentrated model
Gross exposure
About 90% invested on average
Volatility management
None
Weak-market cap
None
Cash after a stop
Reinvested at the next weekly rebalance

Alpha Balanced

Broader and exposure-managed. Lower risk, and lower return potential.

Holdings
More names than Alpha Max
Gross exposure
Set by the market’s trend
Volatility management
Exposure scaled down when markets turn volatile
Weak-market cap
A low cap on exposure in weak markets, checked daily

Stated plainly

What the models do not do.

Each of these is a common assumption about quantitative models. None of them is part of Alpha Balanced or Alpha Max.

  • No mean reversion

    Both models buy strength. Neither buys a stock because it has fallen.

  • No daily rebalancing

    Positions are rebalanced weekly. The only daily action is the stop check (and, for Balanced, the weak-market cap).

  • No floor or defensive manager

    A profit-locked floor was tested in backtests and left out: it lowered risk-adjusted returns.

  • No sector caps or tilts

    Names are ranked on momentum alone; the only concentration limit is a cap on each name.

  • No machine learning

    The ranking is a fixed blend of past returns.

  • No shorts or leverage

    Long-only, and never more than fully invested.

Execution and costs

Decided on the close, traded at the open.

The live-tracked accounts and the backtest take the same decisions from the same data cut-off. The backtest charges a flat cost on every trade.

Decision
Signals, stops and target weights use data through the previous close.
Orders
Market orders, placed shortly after the open, in a separate live-tracked account for each model.
Completion
A rebalance or stop counts as done only when every order has filled; rejected or partial orders are retried on the next check.
Backtest
The same decisions once a day, filled at that day’s open, net of a flat 6 bps per side. Taxes, market impact and borrow costs are not modelled.

Backtest trading cost

6 bps

per side, on every simulated fill

Each model page shows how the backtest changes at 20 bps per side. Live-tracked account fills are the broker-reported prices.

Controls

What the trading systems check.

The operational controls around both live-tracked accounts, described as the code runs them today.

Trading mode
Every run must be told explicitly which kind of account it trades. If that setting is missing or invalid, the run stops before it contacts the broker.
Read-only switch
One setting puts a model into read-only mode: it still works out and logs the orders it would place, and sends none. Removing its schedule stops it entirely.
State from the broker
Each run rebuilds what has been done (the last rebalance, today’s stop check, each position’s high) from the broker’s own order and fill history. If that history cannot be read, the run does not trade.
Position limits
Long-only, never more than fully invested, and no stock above a single-name cap when positions are sized. The market gauge (SPY) is never bought. A run refuses to start if its list of eligible stocks is missing or far shorter than expected.
Holdings check
Once a session, each account’s actual holdings are compared with what the model should hold. A difference beyond a small tolerance, or any short position, is flagged on a private monitoring dashboard and sent as an alert.
Monitoring
The runs report to that dashboard after the open and at the close. A model that has not reported shortly after either is flagged.
Releases
Only tagged releases are deployed, each from a clean copy of the code, and the previous release stays available for rollback. Rule changes are dated in the change log below.
Keys and access
Each model has its own account and its own broker keys, held as secrets on the hosting services and not in the code. This site reads the accounts on its server; the keys never reach a browser.

Measuring performance

How every figure is computed.

Live-tracked account figures come from the broker accounts and are labelled Live-tracked. Backtest figures are simulations on historical data and are labelled Backtest.

Live-tracked return

Live-tracked

Equity now − Equity at startdivided byEquity at start

Broker-reported account equity. Gross of fees.

Sharpe ratio

Annualised return − 3.6%divided byAnnualised volatility

The same 3.6% risk-free rate for live-tracked accounts and backtest. 252 trading days a year.

Live-tracked return
Simple cumulative return on broker-reported account equity from the close before each model’s live-tracked record started, against SPY, QQQ and GLD over the same period. Until the record has about three months of sessions (63) it is marked as too short to judge, and no figure is annualised. It is not time-weighted or money-weighted. Each account is simulated, with no deposits or withdrawals, so there are no external cash flows to adjust for. Alpha Balanced’s record starts in January 2026 and includes trading under earlier rules before 16 Oct 2026; the model change log says what changed. The live-tracked record page also shows a challenger under test, a proposed change to Alpha Max in its own live-tracked account, measured the same way; it is not a model.
Risk metrics
Live-tracked account volatility, Sharpe, Sortino, maximum drawdown, beta to SPY and up days are computed from daily equity changes over the live-tracked period only. Volatility, Sharpe, Sortino and beta are annualised with 252 trading days and shown only after about three months of sessions (63).
Sharpe
Live-tracked account and backtest Sharpe both use a 3.6% risk-free rate: annualised return minus 3.6%, divided by annualised volatility.
Sortino
The backtest Sortino also uses the 3.6% risk-free rate. The live-tracked account Sortino uses none: annualised return divided by the downside deviation of losing days.
Benchmarks
SPY, QQQ and GLD are dividend-adjusted daily closes from a public market-data source, rebased to 100 at each model’s live-tracked account start; the current session’s latest close is the final point. Adjusted closes include dividends, so benchmark figures are total returns.
Fees
Live-tracked returns are simple returns on broker-reported equity, before any fees. Qentari publishes no fee terms and does not manage money for anyone.
Live-tracked and backtest
They are kept apart. The live-tracked record starts on each model’s record start date and is never joined onto the backtest; every backtest figure is labelled as one.
Freshness
Live-tracked account figures refresh about every minute while a page is open and the US market is open, and each reading shows the time it was taken. Outside market hours (weekends, exchange holidays, overnight) they read "Market closed · as of" the last close. Benchmark closes update through the session, about every 15 minutes.

Data sources

Where the numbers come from.

Three sources, each used for one job.

  1. 01

    Live-tracked accounts

    Account equity comes straight from the broker for the two live-tracked accounts, one per model; holdings and fills are recorded the same way and shared with partners under NDA. The accounts trade at live market prices but hold no real capital; actual results would differ because of fills, slippage and capacity. Nothing is entered by hand.

  2. 02

    Benchmarks

    SPY, QQQ and GLD daily closes, dividend-adjusted, from a public market-data source. The same series feeds every live-tracked account comparison on the site.

  3. 03

    Backtest prices

    Split- and dividend-adjusted daily prices from a public market-data source. Index membership on each date is rebuilt from public records of S&P index changes and archived constituent lists, so it can differ slightly from the official history. The live-tracked accounts’ signals use the broker’s market data, which can differ slightly from these consolidated prices.

What we do not claim

No audit has been performed on anything on this site, and nothing is independently verified. We publish no “verified results” badges or hashes.

How every figure is computed

Backtest limits

Read the backtest with these six limits.

The point-in-time backtest has six known limits. Read every backtest number with them in mind.

  1. Universe

    On each date the universe is that day’s S&P 500 and S&P MidCap 400 members, including companies later removed or acquired. About 8% of member-days have no price data, mostly companies since acquired, so some exits are missing from the simulation. The live-tracked accounts trade a fixed list of large- and mid-cap stocks chosen in 2026, which carries a selection bias this backtest avoids.

  2. No out-of-sample period

    The parameters (including the number of names and the stop) were chosen in 2026 using 2023–26 data, so no part of the period is out-of-sample.

  3. Path dependence

    Shifting the weekly rebalance day moves both models’ results by several points a year, and the Alpha Max headline sits at the top of its range.

  4. Concentration

    In both models about 10 stocks produced all of the profit; the other names together lost money. Results depend on the 2025–26 rally in AI and memory stocks, and both models’ results through the end of 2025 are markedly lower.

  5. Costs

    Costs are a flat 6 bps per side; at higher costs both models’ results are lower, as each model page shows. Taxes, market impact and borrow costs are not modelled.

  6. Method

    Decisions use the prior close and fills use the next open, on split- and dividend-adjusted daily data. The results are simulated on historical data.

Model change log

Every rule change, dated.

What changed, in plain words, and what it means for the live-tracked record. This log is updated whenever a rule changes.

How model rules change.

The rules of Alpha Max and Alpha Balanced are frozen.

A proposed change runs first as a challenger in its own live-tracked account for at least four weekly rebalances, alongside the current model, and is judged against criteria written down before the test starts.

A change that is adopted takes effect on a dated rebalance, is recorded in this log, and is marked on every chart.

Live-tracked accounts are not closed or reset. If an account has to move for operational reasons, its full history carries over and the move is marked on the record.

  1. Alpha Balanced

    Alpha Balanced’s published record now shows its live-tracked account’s full history since January 2026, when it began trading. The period before 16 Oct 2026 ran under earlier rules and is marked as such. The rules did not change.

    Record: Starts on 2 Jan 2026, the first session of 2026. Trading before 16 Oct 2026 is marked as earlier rules.

  2. Alpha Max

    Alpha Max moved to a new live-tracked account. The previous live-tracked account was closed. The rules did not change.

    Record: Restarted on 12 Nov 2026 for Alpha Max. Its figures from the previous account are no longer published.

  3. Alpha Balanced

    After its stop exits at the open on 28 Oct 2026, Alpha Balanced’s live-tracked account did not trade until 12 Nov 2026, because of a software fault in the program that runs it. For ten sessions no daily stop check and no weekly rebalance ran; the account held three names and was mostly in cash. The fault was fixed and trading resumed on 12 Nov 2026. The rules did not change.

    Record: The published record includes this period.

  4. Alpha Max

    Alpha Max moved to a new live-tracked account that ran the corrected rules from its first day.

    Record: Restarted on 16 Oct 2026 for Alpha Max. That account was replaced on 12 Nov 2026 (above), and its figures are no longer published.

  5. Alpha Balanced

    The corrected rules took effect on Alpha Balanced’s live-tracked account, which kept trading.

    Record: The published record continues across this date, which the chart marks “Rules corrected”. Trading before it ran under earlier rules.

  6. Both models

    Rule corrections. The trailing stop is checked once a day on the prior close, and a stock it sells is not bought back the same day. A rebalance counts as done only when every order has filled. The market gauge (SPY) is never held. Alpha Balanced’s weak-market cap is applied every day, not only on rebalance days. The backtest was rerun without look-ahead and with each model’s own settings.

    Record: Alpha Balanced’s record includes the period before the corrections, marked as earlier rules. Alpha Max’s figures from before them are no longer published.

  7. Both models

    Fault fixed. The trading runs had been rebalancing at every 15-minute check instead of weekly, and the trailing stop could not trigger.

    Record: Alpha Balanced’s record includes the faulty period, marked as earlier rules. Alpha Max’s figures from it are no longer published.

  8. Alpha Balanced

    From this date the account runs the Alpha Balanced rules. A later review found that before it, for about nine months, the account had been run by a different program.

    Record: The published record includes the period before this date, marked as earlier rules.

  9. Both models

    Current settings for both models chosen, using data from 2023 to 2026. As a result, no part of the backtest period is out-of-sample.

    Record: Alpha Balanced’s record includes the period before 16 Oct 2026, marked as earlier rules. Alpha Max’s figures from that period are no longer published.

Qentari Insights

One idea, one chart and one change-log note a week, on how founders can run their wealth as carefully as their company. Educational only; not advice. Unsubscribe any time.

We use your email only to send Qentari Insights. See the Privacy Policy.

Next step

Start with a conversation.

Tell us who you are and what you would like to talk about. The founder replies.

Live-tracked: broker accounts trading at live market prices with no real capital. Actual results would differ because of fills, slippage and capacity, and this is not a record of managed client money. Past performance does not guarantee future results.

Backtest: simulated, net of 6 bps per side costs, not results of actual trading.

Nothing on this site is an offer or a recommendation. Risk disclosures