Stock

Method

Where the numbers come from, what is computed here, and how the investment score is built.

Every figure on this site comes from a filing, a company release or a named data source, and the analysis says which one. This page sets out where the numbers come from, what is calculated here rather than copied, and where the method has limits.

Sources

  • SEC EDGAR — registration statements (S-1, F-1 and their amendments), final prospectuses (424B4), quarterly and annual reports. Lockup terms are read from the prospectus, not estimated.
  • Company investor-relations pages and press releases, with the date of the release.
  • Fund filings (Form N-PORT) — US registered funds disclose the value they carry each holding at, including private companies.
  • Market data for prices, market capitalisation and trading volume.

Where a fact is not documented, it is left out. An article will say that a figure is unavailable rather than estimate it. Anthropic’s prospectus, for example, discloses that its two largest customers accounted for about a fifth of first-half 2026 revenue but does not name them — so no names appear here either.

What is calculated here

Lockup supply against trading volume. A lockup expiry matters in proportion to how much stock is released relative to how much normally trades. The shares unlocking are divided by average daily volume to give the number of trading days that supply represents. Prospectus terms vary — 90, 160 or 180 days, sometimes conditional on earnings or price — so the dates are read from the document rather than counted from the listing date.

Per-share private pricing from fund filings. Value divided by share count gives the price a fund carried a private holding at, on the date of that filing. Three cautions apply, and each is applied before publication: funds in the same family are not independent opinions; a gap larger than about twofold usually means a different share class rather than a disagreement; and funds have different fiscal year ends, so the comparison is made on one reference date. The filing date is always stated, because these marks are quarterly and lag.

Documented exposure routes. For a company that has not listed, the analysis lists the listed companies with a documented connection — investors, suppliers, customers, substitutes — with the filing or release that establishes each one. An undocumented connection is not published.

Earnings direction. For a listed company the last five quarterly reports of revenue and net income are reduced to a direction rather than a level: whether revenue rose in each of the last three quarters, whether growth is accelerating or slowing, whether the net result of the last two quarters improved on the two before them, and whether a loss turned into a profit. Percentage changes are computed only from a meaningful base, because a quarter near zero produces readings that look dramatic and mean nothing; where fewer than four quarters are available, nothing is inferred. This measure was added on 14 November 2026 and feeds the entry scoring of the model portfolio, set out under rule 7 on the Rules page. It does not enter the ten-factor score described below, which is unchanged.

The investment score

Each company analysis ends with ten factors scored out of ten, each with the evidence behind the score. The factors are chosen for the company but fall into the same categories: market opportunity, the quality of the core asset, differentiation, management and ownership, financial position and runway, revenue and profitability, competitive position, regulatory and execution risk, valuation at the current price, and institutional or analyst coverage.

The overall score is the arithmetic mean of the ten factor scores, to one decimal place. Nothing is weighted and nothing is adjusted afterwards, so you can check the number yourself against the table above it.

A score is a summary of evidence, not a recommendation. It does not say whether to buy.

Scenarios, not price targets

Twelve-month sections set out scenarios with what would have to happen in each, and name the catalyst that decides between them. They are not forecasts and iposight does not issue price targets of its own. Where an analyst consensus target is quoted, the number of covering firms and the date are given, because a target from four firms is not the same as one from twenty.

How often the data is refreshed

The IPO calendar, the key-dates calendar and the filing pipeline are refreshed weekly, on Friday evening — the point in the week when the data is freshest, because US listings price and filings arrive late in the week. Fund-filing pricing is refreshed weekly but the underlying filings are quarterly. Every published figure carries the date it was read.

How the articles are produced

Data collection and drafting use software tools, including AI. Every article is read, checked against its sources and approved by a named editor before it is published.

Where the method has limits

  • Fund-filing marks are quarterly and lag by weeks. A private valuation reported in the press may be newer than the mark shown here.
  • A company that lists drops out of the fund-filing dataset, so coverage of it thins after the IPO.
  • Holdings held through a pooled vehicle give a share count that is not comparable, and those rows are marked rather than divided.
  • Newly listed companies have no quarterly report yet, so some ratios cannot be calculated at all for the first weeks.

Corrections

Errors are corrected on the article with the date of the change. If you find one, write in through the Contact page.