The fastest way to lie with backtests is to leave the methodology out. This post walks through exactly how invest-like's 7-framework consensus scoring did against the S&P 500 over a 5-year window, with the universe filter, the rebalance assumption, the survivorship guard, and the full per-tier cohort table.
If you only read one number: stocks that passed 7 of 7 value-investing frameworks (Buffett, Graham, Fisher, Lynch, Greenblatt, Munger, Smith) returned +73.6 percentage points above the S&P 500 over 5 years, by median. The full table is below, and the live numbers update daily at /track-record.
If you only read two numbers: that same 7-of-7 cohort had only 42 stocks in it at any given quarterly rebalance — small cohorts amplify both signal and noise. Read this as a directional result, not a proven edge.
What the screen actually does
The screen is not an opinion. It runs the documented rules of seven published investor frameworks against every stock in the coverage universe (currently ~12,000 listed tickers; ~3,700 pass the Tier-1 quality floor that excludes ETFs, REITs, preferred shares, bond funds, and most cyclical commodity producers).
For each stock, each framework returns a verdict — strong fit, partial fit, or weak fit — based on its published criteria. The seven verdicts get tallied. A stock with 7-of-7 strong-fit reads is in the highest-conviction tier; a stock with 6-of-7 is in the next tier; and so on.
The seven framework rule sets are reproduced verbatim from the methodology page:
- Warren Buffett: ROE ≥ 15%, P/E ≤ 12 for stable businesses, durable competitive advantage, conservative balance sheet
- Benjamin Graham: P/E ≤ 15, P/B ≤ 1.5, current ratio ≥ 2, 5-year profitability, dividend track record
- Philip Fisher: Gross margin ≥ 50%, revenue CAGR ≥ 10%, sustained R&D, management quality signals
- Peter Lynch: PEG ≤ 1.0, EPS growth ≥ 15%, debt/equity ≤ 0.5, recognisable business
- Joel Greenblatt: Magic Formula — earnings yield ≥ 10% AND ROIC ≥ 25% (rank-based)
- Charlie Munger: ROIC ≥ 18% sustained 5+ years, P/E up to 30 for true quality, low debt
- Terry Smith: ROCE ≥ 20%, FCF / Net Income ≥ 95%, gross margin ≥ 45%, no banks or cyclicals
Every rule is published, every rule is auditable from the open scoring code, and every per-stock pass/fail is shown on the individual stock page. There is no proprietary black box.
The 5-year cohort results
Here is the median return delta versus the S&P 500 over a rolling 5-year window, by consensus tier:
| Consensus tier | Cohort size (avg) | Median 5y return vs S&P 500 | % of cohort that beat the index |
|---|
| 7 of 7 frameworks agree | 42 | +73.6% | 81% |
| 6 of 7 frameworks agree | 168 | +41.2% | 67% |
| 5 of 7 frameworks agree | 412 | +18.4% | 56% |
| 4 of 7 frameworks agree | 681 | +6.1% | 52% |
| 3 of 7 frameworks agree | 893 | −2.4% | 47% |
| 2 of 7 frameworks agree | 712 | −9.8% | 41% |
| 1 of 7 frameworks agree | 487 | −16.3% | 36% |
| 0 of 7 frameworks agree | 305 | −28.5% | 28% |
The cohort sizes are averaged across the rolling quarterly rebalances over the 5-year window. The "% of cohort that beat the index" is the share of stocks in that tier whose 5-year total return exceeded the S&P 500's over the same period.
The pattern that matters: consensus is monotonic. Higher tier → higher median return, higher hit rate. There is no inversion, no zig-zag. That is what you would expect if framework agreement carries real information, and it is not what you would expect from random noise (random tiers would scatter).
What this does not prove
A 5-year window with 42 stocks per quarter in the top tier is not a proven edge. It is a directional result. Here is what could explain it away:
- Selection bias from the universe filter. We pre-filter to ~3,700 quality stocks before scoring. That filter excludes most of the destruction (cyclical commodity producers, REITs that blew up, micro-caps). Some of the +73.6% is the universe filter, not the consensus scoring.
- Crowded high-quality factor. From 2020 to 2025, the "high-quality compounder" factor outperformed by a wide margin in general. Some of the result is the factor, not the screen.
- No transaction costs, no tax drag. The backtest assumes quarterly rebalancing at close. Real-world friction would shave returns.
- Survivorship. Delisted stocks are excluded from the universe at the time of scoring (you cannot score a stock that does not have current fundamentals). This biases the result mildly upward.
The honest read: the 7-of-7 tier likely outperforms the index, the magnitude is probably real but not the literal +73.6%, and the monotonicity of the cohort table is the part of the result that is hardest to explain away by bias.
What we did not do
We did not curve-fit the rules. The seven rule sets are taken verbatim from their public sources (Buffett's letters, Graham's Intelligent Investor, Fisher's Common Stocks and Uncommon Profits, Lynch's One Up on Wall Street, Greenblatt's Little Book that Beats the Market, Munger's published talks, Smith's Investing for Growth). No threshold was tuned against historical returns. The "fit" definitions are documented at /methodology/.
We did not back-test only US stocks. The 3,700-stock universe is global; the S&P 500 comparison is just the most-recognised benchmark.
We did not include the Buffett Brain LLM verdicts. Buffett Brain is a separate product (an AI verdict that synthesises the framework scores plus the actual Berkshire letters into a single 0-100 grade with reasoning). The backtest above uses only the deterministic 7-framework rule scoring. Buffett Brain's live track record — much smaller cohort, still growing — is published separately at the same /track-record page under section 2.
The 30 stocks in the live conviction portfolio
The published conviction portfolio is the 7-of-7 cohort at the current rebalance, capped to 30 stocks by market cap. Entry timestamps are locked at portfolio inclusion and cannot be backdated (the database stamp is the source of truth). The NAV is recomputed daily against SPY. As of the latest rebalance, it includes:
- Several high-ROIC technology compounders (the Munger/Fisher overlap that the Buffett-only framework would have rejected as too expensive)
- Multiple consumer-staples names (the Graham/Buffett overlap)
- A small number of financials that pass even Terry Smith's strict no-banks-no-cyclicals filter (e.g. Visa, Mastercard which Smith treats as toll-bridges not banks)
- Two energy names that pass on cash-generation despite Smith's cyclical exclusion
The full list, with entry date, entry price, current NAV, and 5-year backtest position size, is visible at /track-record/ for any visitor. No paywall on the macro numbers.
How to verify this yourself
Three checks anyone can run:
- Pick any single stock in the 7-of-7 tier and open its Buffett Brain page. Every framework's pass/fail is listed with the underlying metric (e.g. "Buffett rule on ROE: 31.4% — pass"). The screen is reproducible per stock.
- Open /methodology/ and read the exact rule for each framework. Compare to the original published source. The point is that there is no proprietary judgement.
- Compare against Yahoo Finance or Morningstar. Pull the same stock, look at the same metrics, apply the same rules manually. The verdict should match.
If any verification fails, please email me (the founder, zaid@invest-like.com). The point of publishing the methodology and the track record together is that any user who finds an error should be able to flag it.
Disclosure
This is an educational analysis tool. Backtested performance is not a forecast. Past performance does not guarantee future returns. The framework scores are a deterministic mapping from published rules to fundamental data — they are not investment advice, not a recommendation, and not personalised to any visitor's situation. The German legal disclaimer applies: this is a Bildungstool, keine Anlageberatung.
The author is Zaid Ghazal, indie founder of invest-like, based in Kiel, Germany. The app runs on Next.js + Supabase out of Frankfurt (eu-central-1) for GDPR locality. The full company background and EEAT signals are documented at /about/.