Do our valuations work?

The public methodology, the backtest results and the model changelog. Every score links here; every change is announced before it lands.

Ten thousand, ten years, by valuation fifthten years · monthly rebalance · point-in-time data · ten thousand invested

Most undervalued fifth$59,939

Second$31,210

Middle$19,840

Fourth$14,120

Most overvalued fifth$9,870

The index$17,040

The upright rule is the $10,000 stake · governed backtest v418, point-in-time data, monthly rebalance, no survivorship · illustrative

Ten thousand in the most undervalued fifth became $59,939 against $17,040 in the index; the most overvalued fifth lost money. Directional accuracy at one year is 63.6%, rising with the horizon. Point-in-time data, monthly rebalance, no survivorship.

Hit ratedirection right at one year

One year63.6%

Two years66.9%

Three years70.2%

By region

United States64.9%

Europe61.2%

Asia-Pacific58.7%

By sectorweakest last

Financial services68.0%

Industrials65.4%

Software62.1%

Semiconductors57.9%

Biotech51.3%

Model changelogwhat changed, why, how many names moved

v419 lands 9 Sep 2026. Terminal growth capped at the ten-year nominal GDP forecast rather than a fixed 3%. It is expected to lower 38 values by up to 6%; every affected name is listed in the letter before it lands.

Version	What changed	Why	Affected
v41825 Aug 2026	Years one to two revenue growth reads the guided range from the transcript when prepared remarks give one.	PYPL: “FY27 revenue growth six to seven per cent, up from four to five.” Blended value $109 → $118 on that name; 14 names moved more than 3%.	14 names
v41711 Aug 2026	Blume adjustment on the headline beta; bottom-up beta kept as a cross-check.	Raw regression betas overstated the cost of equity on thin trading histories; 61 names moved, median 1.8%.	61 names
v41621 Jul 2026	Industry and country multiples switched from means to medians.	A single 4,000× outlier described the outlier rather than the set; the relative valuation on 203 names changed.	203 names
v41530 Jun 2026	Retail sentiment made fade-only in every strategy.	Mention velocity never supports a long thesis; it can only weaken one. No valuation changed.	strategies only

A value that moves without explanation is the complaint behind every one-star review of a valuation site. Every change here names the version, the sentence that caused it, and the names it touched; the letter carries the ones on your list.

Methodologypublic

Eleven signals, blended with weights that are shown on every score card. Four discounted models, a reverse DCF, disclosure flows, filings read, calls parsed, revisions and cross-checked news. Every run records its id, model version, tunables and input vintage; a figure without those four is not one of ours.

Trial ledgerQuant

Trials recorded448

Deflated Sharpe, flagship1.31

Out-of-sample embargotwelve months

Survivorshipnone · delisted names kept

run_id 8f2c1e04-…-a91bmodel dcf@1.9.2 · v418tunables v4vintage 2026-09-02inputs backtest engine 2.3, point-in-time universe, model changelogstatus provisional · previous run shown

Design preview: illustrative figures throughout, invented and tied to no live run and no real company’s accounts. Analysis is not advice and nothing here is a recommendation to buy or sell any security.

Insider, fund and congressional disclosures are shown with the lag between transaction and filing, because that lag is part of what a row means. Valuations are model output under stated conventions, reproducible only from a run’s own recorded inputs, version and vintage.