Simulations

Honest limitations

• Stock simulations are an approximation. The price metrics (1-yr and 5-yr gains, trend) and the TradingView rating (rebuilt bar-by-bar from TradingView's own open-source formula, validated within ±0.10 of the live screener) are reconstructed exactly. Analyst grades, price targets, and forward P/E come from harvested point-in-time records — archived Yahoo and MarketWatch pages and dated analyst-action history — where coverage exists; a metric the era's data can't feed for at least half the universe drops out, and verdict thresholds scale to the points in play (each result reports its average coverage). Grades and targets reach back through the whole period, but forward P/E doesn't — the archives simply carry it too rarely — so historical runs score 27 of the rubric's 32 points rather than the full board the live site uses. Those sources don't grade alike — the panel behind the 2016–17 years runs about a quarter-point more bullish on the 1–5 consensus scale — so each is shifted onto the scale the rubric's grade bands were drawn for. That shift is measured directly where the panels overlap, in 2018–22, and applied to 2016–17 on the assumption the panel's scale didn't move: treat those two years as the least certain. From the site's launch onward, simulations use our own stored daily snapshots and stop approximating. ETF simulations replay the full rubric exactly throughout.
• Point-in-time universe. The stock list is rebuilt as it stood at the start of each month — the 100 largest US-listed companies by market cap, ADRs included, from archived market-cap tables — so companies later acquired or delisted are in it and are sold at their last traded price. 2016–19 is an approximation, assembled from archived S&P 500 weights (float-adjusted, standing in for full caps) and a top-50-by-cap table, so foreign ADRs ranked 51–100 in those years are missed, and that table printed foreign caps in local currency through 2018–19, so those two years carry its ADRs from the last clean 2017 table; from mid-2020 the tables are exact (January to June 2020 use the same approximate method), though ADRs outside the global top 100 can still be missed. Three members never get a price — PetroChina, China Life, and Dow Chemical's 2016–17 months under a symbol Dow Inc. reused — so they sit on the list unscoreable, are never bought, and drop out of the equal-weight line; EMC and Monsanto are left out entirely because their symbols now belong to other securities. The ETF list is point-in-time too, from September 2026: the 50 largest US ETFs by assets as of each month, bond and money-market funds excluded, read from archived etfdb.com tables (Common Crawl and the Wayback Machine). 84 of its 120 months come from a capture within the preceding two months; the rest carry the previous list forward, the longest such gap running July 2024 to January 2025; from 2026 it is today's list. Seventy-nine funds were members at some point, thirty-one of them no longer in today's fifty. That change cut the ten-year ETF figure from about $12,350 per $1,000 to about $4,650, against $3,280 for equal weight and $4,040 for SPY: most of what the old list showed was the list. The equal-weight line applies zero skill to the same monthly list, so any part of the method's return it matches is the list, not the verdicts.
• The rubric's bands were calibrated recently, with this same history in view — the backtest is partly grading its own homework. A walk-forward check softens that: rerun with the bands as first transcribed from the 2013–19 sheet, fixed before 2020's data existed, the stock method returned $3,099 per $1,000 over January 2020 to July 2026 against $2,600 for equal weight and $2,546 for SPY, and beat equal weight in 81 of 92 rolling three-year windows. Trust that number more than the ten-year one. The stock verdict rule was re-tuned on 3 September 2026 against this record — the line moved from 20/13 to 16 of 32 points, and BUY became the ten highest totals at or above it rather than everything above a fixed line — chosen on start months before 2021 and checked on those after; the bands were tried the same way and none beat the existing ones out of sample, so they stand. Every simulated figure on this page uses today's cutoffs throughout, and the forward record from launch is the test they have not yet passed.
• Dividends are automatically reinvested (all prices are total-return adjusted closes) and shares are fractional. The headline numbers carry no taxes, fees or slippage; the “with costs” figures deduct 0.05% slippage on every trade and a 20% tax on each realised gain (losses not offset). Equal weight and SPY never trade, so they carry none of it.
• Monthly granularity: verdicts are computed and trades execute at the daily close of the last trading day on or before the 1st of each month; the weekly candle stands in only where daily data is missing.
• Instruments too new to score (the board's LIMITED state) are skipped until enough history exists; anything already held is simply kept while unscoreable.
• Price history comes from an unofficial Yahoo Finance API; the eleven delisted companies' series were fetched once from Tiingo and stored.
• Past performance is not indicative of future results. This is a published record of one method — not investment advice.