How the grade works.
And when it's wrong.

Every stock starts at 50 points — a C−. It then earns or loses points on six questions any careful investor would ask, using audited numbers from the company's own filings:

• Is revenue growing? Sales up more than 20% year-over-year earns the most points; shrinking sales lose points.
• Are profits growing? Same idea, applied to the bottom line.
• Is the debt manageable? Low debt compared to what shareholders own earns points; heavy leverage loses them.
• Does management earn a good return on your money? Return on equity — profit per dollar shareholders have in the company.
• Can it pay its bills? A healthy cash cushion against short-term obligations.
• What do professional analysts think? Price-target upside gets a small voice — deliberately the smallest of the six.

Add it up, cap it between 0 and 100, and map to a letter: 90+ is an A+, below 35 is an F. The same numbers always produce the same grade — no black box, no "the model felt bearish." Growth and profitability dominate, balance-sheet safety matters nearly as much, and market opinion gets a voice but never a veto. A company can't score an A on hype.

Want the exact point bands, straight from the production code? The complete algorithm is published on our engineering site: Anatomy of a Grade. Every grade in your dashboard also shows its own arithmetic — hover it and see which factors moved and why.

The grade never travels alone.

Alongside every letter grade you'll see independent signals computed separately, so one lens can't hide another: an ML health score (a machine-learned second opinion — when it disagrees with the grade, that disagreement is itself worth your attention), analyst sentiment from real ratings, and profitability badges — including a green "cash-positive despite losses" flag for companies running the deliberate-reinvestment playbook that accounting income makes look worse than it is.

In August 2026, a GMI report briefly showed Taiwan Semiconductor with a market value of $50 trillion — roughly half the economic output of the entire planet. The cause: our data vendor reports some figures in each company's home currency, and one derived number silently mixed Taiwan dollars with US dollars.

It was caught within hours — because the founder runs his own family portfolio through the platform every week, and $50 trillion is personally unbelievable. But the fix mattered more than the catch: GMI now computes its valuation and quality metrics in-house from raw audited statements instead of trusting the vendor's pre-computed figures. When a number can't be computed honestly, your report shows a "—", never an invented value.

Full detail for the technically curious: Data Provenance & Accuracy — every source, every refresh cadence, every validation rule.

When we break something, we publish it.

Every production system has incidents. The only variables are whether the operator learns from them — and whether you get to see the learning. GMI keeps a public incident log with real dates and real root causes: Production Postmortems. The bug count isn't zero. The same-bug-twice count is.