Most "investment newsletter" track records are unfalsifiable. The author tells you the strategy returned X% last year, but the recommended trades are not timestamped, the rebalance dates are foggy, and the entry prices are picked after the fact. There is no way to audit whether the strategy actually predicted what it claims to have predicted.
We built the live model portfolio to fix exactly that, and it shipped this week. Here is the design, why each design decision exists, and how to verify the numbers.
What the model portfolio is
The model portfolio is a 30-stock basket, picked from the 7-of-7 framework consensus tier, capped to 30 by market cap. As of the latest rebalance, the 30 stocks span healthcare, industrial gases, software, specialty chemicals, payment networks, and a handful of high-quality consumer-staples names. Every position is visible at /track-record/ under the Live Conviction Portfolio section.
For each holding, the page shows:
- Entry date (the calendar date the stock entered the portfolio)
- Entry price (the close on the entry date)
- Current price (refreshed daily at market close)
- Position weight (equal-weight initially; drifts with price)
- Days held
- Total return since entry
The portfolio NAV is plotted against SPY (the S&P 500 ETF) on the same time axis, normalised to 100 at the portfolio inception date. Both lines update together every trading day.
Why this is harder to fake than most track records
Three architectural choices make this difficult to backdate or massage:
1. Entry timestamps are server-side and immutable
When a stock enters the portfolio, the database insert stamps the entry timestamp from the server clock (Frankfurt / eu-central-1). The stamp is the source of truth. We could not, even if we wanted to, retroactively claim "actually we bought this last week" because the row would have to be deleted and re-inserted, and the audit log would show the delete.
In practice, the entries happen during the quarterly rebalance cron. The rebalance is a single SQL transaction; either all 30 stocks land at the same timestamp, or none do.
2. Entry prices are end-of-day close, sourced from Financial Modeling Prep
The entry price for each stock is the official end-of-day close on the entry date, from Financial Modeling Prep's licensed price data feed. Not an intraday tick, not "our average fill", not "approximately X". Anyone with an FMP account (or Yahoo Finance, or any broker's historical data) can pull the same close price and verify.
3. The NAV calculation is a deterministic transform
NAV(t) = sum over 30 positions of (current_price[i] / entry_price[i]) × position_weight[i] × 100.
There are no smoothing, no recovery basis, no "look-through" adjustments. The formula is published at /methodology/. A user can pull our entry prices + current prices and reproduce the NAV to four decimal places.
What this is not
It is not a real fund. We do not custody money. Returns are paper returns on a published model, not realized after-fee, after-tax returns on actual capital.
It is not a back-test. Back-tests have the temptation to fit the parameters to past returns. The model portfolio is forward-tracking: the entry happens now, the return accumulates from now, and the future return is unknown to everyone, us included.
It is not a recommendation. The 30 stocks are the mechanical output of the 7-of-7 framework consensus screen, capped by market cap. They are the stocks the screen would hold; whether you should hold them is your own call.
It is not survivor-curated. If a stock enters the portfolio and then crashes, it stays in the portfolio (with its negative return) until the next quarterly rebalance ejects it on the screen's own logic. We do not retroactively remove embarrassing positions.
Why we shipped the rule-based backtest first
The model portfolio went live this week. The 5-year rule-based backtest (showing the 7-of-7 cohort returned +73.6% above the S&P 500 over a rolling 5-year window) went live last week.
The two answer different questions:
- Back-test answers: "Would this screen have outperformed in the past 5 years?" - directional, helpful, but at the mercy of selection bias and curve-fit risk.
- Live model portfolio answers: "Does this screen outperform going forward?" - slow to build evidence, but the only number that is genuinely unfalsifiable.
Both numbers will be published in parallel forever. If the live portfolio underperforms over the next year, we will publish that. If it outperforms, we will publish that too. The whole point of publishing both is that the comparison between them - back-test versus live - is itself an honesty signal. If the live portfolio dramatically underperforms the back-test, that's evidence the back-test was curve-fit. If it tracks closely, that's evidence the screen has real signal.
How the portfolio gets rebuilt
Every quarter (end-March / end-June / end-September / end-December), the screen re-runs against the full universe with the latest quarterly fundamentals. The new 7-of-7 set is computed. The portfolio holds:
- Any stock still in the 7-of-7 set: keep
- Any stock no longer in the 7-of-7 set: sell at next-close
- Any new entrant: buy at next-close
We did not optimise the rebalance frequency for return; quarterly is the cleanest cadence that matches the cadence of new quarterly fundamentals data. Annual would be too sluggish; monthly would burn signal on noise.
The first rebalance after the quarterly date is logged with a timestamp, the position changes are dumped to a CSV that anyone can request via zaid@invest-like.com, and the NAV chart shows a small marker on the rebalance day so the kink in performance is visible.
What to watch as a skeptic
If you want to test whether this portfolio is actually working, four things to track:
-
NAV vs SPY at the 3-month mark. After three months of forward tracking, is the portfolio at least matching the benchmark? If it has dramatically underperformed by 3-5%+, the back-test was probably overfit and we should say so publicly.
-
Drawdown vs SPY during the next correction. High-quality compounders are supposed to draw down less than the index in panicks. If the portfolio draws down more than SPY during a 10%+ correction, the screen is picking up something other than quality.
-
Position turnover at quarterly rebalances. Should be low (maybe 2-5 positions changing per quarter). If turnover is 10+ positions per quarter, the screen is too sensitive to noise.
-
The honesty signal. If a position in the portfolio publicly blows up (fraud, accounting restatement, major regulatory action), do we keep it in the portfolio with its losses until the rebalance? Yes - published as a transparency commitment. The track record only counts if the losses count too.
Where to find this on the site
- The live portfolio + NAV chart: /track-record/
- The 5-year rule-based back-test (sibling page, same screen, historical version): /track-record/
- The methodology (full ruleset + rebalance logic + scoring code link): /methodology/
- The individual stock pages for each position: /buffett/aapl/ and similar
The macro stats are public for everyone. The full stock list with entry dates is visible to signed-in users (free or Pro) - the gate is just to prevent automated scraping; signup is free, no card.
Disclosure
Educational analysis tool. The live model portfolio is a paper portfolio, not a fund. Returns are pre-fee, pre-tax, and not personalised. Past performance is not a forecast. The screen is documented; the screen could change in the future (any change will be timestamped and published). Built by Zaid Ghazal in Kiel, Germany.